From 9a44a2ace192982cbe448863539be240b419d108 Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Fri, 26 Sep 2025 10:04:53 -0700
Subject: [PATCH 01/14] ollama support

---
 backend/onyx/llm/factory.py                   |  34 ++++-
 backend/onyx/llm/llm_provider_options.py      |  26 ++++
 backend/onyx/server/manage/llm/api.py         |  80 +++++++++++
 backend/onyx/server/manage/llm/models.py      |   6 +
 .../llm/LLMProviderUpdateForm.tsx             | 131 +++++++++++++++++-
 web/src/lib/hooks.ts                          |  28 ++++
 6 files changed, 302 insertions(+), 3 deletions(-)

diff --git a/backend/onyx/llm/factory.py b/backend/onyx/llm/factory.py
index 985b8e129d9..6f6a07da4b6 100644
--- a/backend/onyx/llm/factory.py
+++ b/backend/onyx/llm/factory.py
@@ -13,6 +13,8 @@
 from onyx.llm.chat_llm import DefaultMultiLLM
 from onyx.llm.exceptions import GenAIDisabledException
 from onyx.llm.interfaces import LLM
+from onyx.llm.llm_provider_options import OLLAMA_API_KEY_CONFIG_KEY
+from onyx.llm.llm_provider_options import OLLAMA_PROVIDER_NAME
 from onyx.llm.override_models import LLMOverride
 from onyx.llm.utils import get_max_input_tokens_from_llm_provider
 from onyx.llm.utils import model_supports_image_input
@@ -30,7 +32,29 @@ def _build_extra_model_kwargs(provider: str) -> dict[str, Any]:
     For now, just using the GEN_AI_MODEL_FALLBACK_MAX_TOKENS value.
     TODO: allow model-specific values to be configured via the UI.
     """
-    return {"num_ctx": GEN_AI_MODEL_FALLBACK_MAX_TOKENS} if provider == "ollama" else {}
+    return (
+        {"num_ctx": GEN_AI_MODEL_FALLBACK_MAX_TOKENS}
+        if provider == OLLAMA_PROVIDER_NAME
+        else {}
+    )
+
+
+def _build_provider_extra_headers(
+    provider: str, custom_config: dict[str, str] | None
+) -> dict[str, str]:
+    if provider != OLLAMA_PROVIDER_NAME or not custom_config:
+        return {}
+
+    raw_api_key = custom_config.get(OLLAMA_API_KEY_CONFIG_KEY)
+
+    api_key = raw_api_key.strip() if raw_api_key else None
+    if not api_key:
+        return {}
+
+    if not api_key.lower().startswith("bearer "):
+        api_key = f"Bearer {api_key}"
+
+    return {"Authorization": api_key}
 
 
 def get_main_llm_from_tuple(
@@ -272,6 +296,12 @@ def get_llm(
 ) -> LLM:
     if temperature is None:
         temperature = GEN_AI_TEMPERATURE
+
+    extra_headers = build_llm_extra_headers(additional_headers)
+    provider_extra_headers = _build_provider_extra_headers(provider, custom_config)
+    if provider_extra_headers:
+        extra_headers.update(provider_extra_headers)
+
     return DefaultMultiLLM(
         model_provider=provider,
         model_name=model,
@@ -282,7 +312,7 @@ def get_llm(
         timeout=timeout,
         temperature=temperature,
         custom_config=custom_config,
-        extra_headers=build_llm_extra_headers(additional_headers),
+        extra_headers=extra_headers,
         model_kwargs=_build_extra_model_kwargs(provider),
         long_term_logger=long_term_logger,
         max_input_tokens=max_input_tokens,
diff --git a/backend/onyx/llm/llm_provider_options.py b/backend/onyx/llm/llm_provider_options.py
index 462e7003e16..844e6740023 100644
--- a/backend/onyx/llm/llm_provider_options.py
+++ b/backend/onyx/llm/llm_provider_options.py
@@ -97,6 +97,9 @@ class WellKnownLLMProviderDescriptor(BaseModel):
 ][::-1]
 BEDROCK_DEFAULT_MODEL = "anthropic.claude-3-5-sonnet-20241022-v2:0"
 
+OLLAMA_PROVIDER_NAME = "ollama"
+OLLAMA_API_KEY_CONFIG_KEY = "OLLAMA_API_KEY"
+
 IGNORABLE_ANTHROPIC_MODELS = [
     "claude-2",
     "claude-instant-1",
@@ -160,6 +163,7 @@ class WellKnownLLMProviderDescriptor(BaseModel):
     BEDROCK_PROVIDER_NAME: BEDROCK_MODEL_NAMES,
     ANTHROPIC_PROVIDER_NAME: ANTHROPIC_MODEL_NAMES,
     VERTEXAI_PROVIDER_NAME: VERTEXAI_MODEL_NAMES,
+    OLLAMA_PROVIDER_NAME: [],
 }
 
 _PROVIDER_TO_VISIBLE_MODELS_MAP = {
@@ -167,6 +171,7 @@ class WellKnownLLMProviderDescriptor(BaseModel):
     BEDROCK_PROVIDER_NAME: [BEDROCK_DEFAULT_MODEL],
     ANTHROPIC_PROVIDER_NAME: ANTHROPIC_VISIBLE_MODEL_NAMES,
     VERTEXAI_PROVIDER_NAME: VERTEXAI_VISIBLE_MODEL_NAMES,
+    OLLAMA_PROVIDER_NAME: [],
 }
 
 
@@ -185,6 +190,27 @@ def fetch_available_well_known_llms() -> list[WellKnownLLMProviderDescriptor]:
             default_model="gpt-4o",
             default_fast_model="gpt-4o-mini",
         ),
+        WellKnownLLMProviderDescriptor(
+            name=OLLAMA_PROVIDER_NAME,
+            display_name="Ollama",
+            api_key_required=False,
+            api_base_required=True,
+            api_version_required=False,
+            custom_config_keys=[
+                CustomConfigKey(
+                    name=OLLAMA_API_KEY_CONFIG_KEY,
+                    display_name="Ollama API Key",
+                    description="Optional API key used when connecting to Ollama Cloud (API base must be https://ollama.com).",
+                    is_required=False,
+                    is_secret=True,
+                )
+            ],
+            model_configurations=fetch_model_configurations_for_provider(
+                OLLAMA_PROVIDER_NAME
+            ),
+            default_model=None,
+            default_fast_model=None,
+        ),
         WellKnownLLMProviderDescriptor(
             name=ANTHROPIC_PROVIDER_NAME,
             display_name="Anthropic",
diff --git a/backend/onyx/server/manage/llm/api.py b/backend/onyx/server/manage/llm/api.py
index 59bdd4b91da..ff3aefdb1ba 100644
--- a/backend/onyx/server/manage/llm/api.py
+++ b/backend/onyx/server/manage/llm/api.py
@@ -4,6 +4,7 @@
 from datetime import timezone
 
 import boto3
+import httpx
 from botocore.exceptions import BotoCoreError
 from botocore.exceptions import ClientError
 from botocore.exceptions import NoCredentialsError
@@ -40,6 +41,7 @@
 from onyx.server.manage.llm.models import LLMProviderUpsertRequest
 from onyx.server.manage.llm.models import LLMProviderView
 from onyx.server.manage.llm.models import ModelConfigurationUpsertRequest
+from onyx.server.manage.llm.models import OllamaModelsRequest
 from onyx.server.manage.llm.models import TestLLMRequest
 from onyx.server.manage.llm.models import VisionProviderResponse
 from onyx.utils.logger import setup_logger
@@ -474,3 +476,81 @@ def get_bedrock_available_models(
         raise HTTPException(
             status_code=500, detail=f"Unexpected error fetching Bedrock models: {e}"
         )
+
+
+def _extract_model_names(response_json: dict) -> set[str]:
+    models = response_json.get("models") or response_json.get("model") or []
+    if not isinstance(models, list):
+        return set()
+
+    model_names: set[str] = set()
+    for model in models:
+        if not isinstance(model, dict):
+            continue
+        candidate = model.get("model") or model.get("name")
+        if isinstance(candidate, str) and candidate:
+            model_names.add(candidate)
+    return model_names
+
+
+@admin_router.post("/ollama/available-models")
+def get_ollama_available_models(
+    request: OllamaModelsRequest,
+    _: User | None = Depends(current_admin_user),
+) -> list[str]:
+    """Fetch the list of available models from an Ollama server."""
+
+    cleaned_api_base = request.api_base.rstrip("/")
+    if not cleaned_api_base:
+        raise HTTPException(
+            status_code=400, detail="API base URL is required to fetch Ollama models."
+        )
+
+    headers: dict[str, str] = {}
+
+    endpoints = ["api/tags", "api/models"]
+    last_error: Exception | None = None
+
+    for endpoint in endpoints:
+        url = f"{cleaned_api_base}/{endpoint}"
+        try:
+            response = httpx.get(url, headers=headers, timeout=10.0)
+            response.raise_for_status()
+            try:
+                response_json = response.json()
+            except ValueError as e:
+                last_error = e
+                continue
+
+            model_names = _extract_model_names(response_json)
+            if model_names:
+                return sorted(model_names)
+
+            last_error = ValueError("No models returned from Ollama response")
+        except httpx.HTTPStatusError as e:
+            last_error = e
+            # If Ollama doesn't support the endpoint, try the next one
+            if e.response.status_code in {404, 405}:
+                continue
+            raise HTTPException(
+                status_code=e.response.status_code,
+                detail=(
+                    f"Failed to fetch Ollama models: "
+                    f"{e.response.text or e.response.reason_phrase}"
+                ),
+            )
+        except httpx.RequestError as e:
+            raise HTTPException(
+                status_code=400,
+                detail=f"Failed to reach Ollama server at {cleaned_api_base}: {e}",
+            )
+
+    error_detail = (
+        str(last_error)
+        if last_error
+        else "Unexpected response from Ollama when listing models"
+    )
+    raise HTTPException(
+        status_code=400,
+        detail=f"Failed to fetch Ollama models: {error_detail}",
+    )
diff --git a/backend/onyx/server/manage/llm/models.py b/backend/onyx/server/manage/llm/models.py
index 027769a0929..83de2809b57 100644
--- a/backend/onyx/server/manage/llm/models.py
+++ b/backend/onyx/server/manage/llm/models.py
@@ -196,3 +196,9 @@ class BedrockModelsRequest(BaseModel):
     aws_secret_access_key: str | None = None
     aws_bearer_token_bedrock: str | None = None
     provider_name: str | None = None  # Optional: to save models to existing provider
+
+
+class OllamaModelsRequest(BaseModel):
+    api_base: str
+    api_key: str | None = None
+    provider_name: str | None = None
diff --git a/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx b/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx
index e3ca2ff9779..e38372d1048 100644
--- a/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx
+++ b/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx
@@ -58,7 +58,9 @@ export function LLMProviderUpdateForm({
     name:
       existingLlmProvider?.name || (firstTimeConfiguration ? "Default" : ""),
     api_key: existingLlmProvider?.api_key ?? "",
-    api_base: existingLlmProvider?.api_base ?? "",
+    api_base:
+      existingLlmProvider?.api_base ??
+      (llmProviderDescriptor.name === "ollama" ? "http://127.0.0.1:11434" : ""),
     api_version: existingLlmProvider?.api_version ?? "",
     // For Azure OpenAI, combine api_base and api_version into target_uri
     target_uri:
@@ -277,6 +279,97 @@ export function LLMProviderUpdateForm({
     }
   };
 
+  const fetchOllamaModels = async (values: any, setFieldValue: any) => {
+    if (!values.api_base) {
+      setFetchModelsError("API Base is required to fetch Ollama models");
+      return;
+    }
+
+    setIsFetchingModels(true);
+    setFetchModelsError("");
+
+    try {
+      const response = await fetch("/api/admin/llm/ollama/available-models", {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+        },
+        body: JSON.stringify({
+          api_base: values.api_base,
+          provider_name: existingLlmProvider?.name,
+        }),
+      });
+
+      if (!response.ok) {
+        let errorMessage = "Failed to fetch models";
+        try {
+          const errorData = await response.json();
+          errorMessage = errorData.detail || errorMessage;
+        } catch {
+          // ignore JSON parsing errors and use the fallback message
+        }
+
+        throw new Error(errorMessage);
+      }
+
+      const availableModels: string[] = await response.json();
+
+      const updatedModelConfigs = availableModels.map((modelName) => {
+        const existingConfig = llmProviderDescriptor.model_configurations.find(
+          (config) => config.name === modelName
+        );
+
+        return {
+          name: modelName,
+          is_visible: existingConfig?.is_visible ?? false,
+          max_input_tokens: null,
+          supports_image_input: false,
+        };
+      });
+
+      llmProviderDescriptor.model_configurations = updatedModelConfigs;
+
+      const previouslySelectedModels = values.selected_model_names || [];
+      const stillAvailableSelectedModels = previouslySelectedModels.filter(
+        (modelName: string) => availableModels.includes(modelName)
+      );
+
+      setFieldValue("selected_model_names", stillAvailableSelectedModels);
+
+      if (
+        (!values.default_model_name ||
+          !availableModels.includes(values.default_model_name)) &&
+        availableModels.length > 0
+      ) {
+        setFieldValue("default_model_name", availableModels[0]);
+      }
+
+      if (
+        values.fast_default_model_name &&
+        !availableModels.includes(values.fast_default_model_name)
+      ) {
+        setFieldValue("fast_default_model_name", null);
+      }
+
+      setFieldValue("_modelListUpdated", Date.now());
+
+      setPopup?.({
+        message: `Successfully fetched ${availableModels.length} models from Ollama.`,
+        type: "success",
+      });
+    } catch (error) {
+      const errorMessage =
+        error instanceof Error ? error.message : "Unknown error";
+      setFetchModelsError(errorMessage);
+      setPopup?.({
+        message: `Failed to fetch models: ${errorMessage}`,
+        type: "error",
+      });
+    } finally {
+      setIsFetchingModels(false);
+    }
+  };
+
   return (
     <Formik
       initialValues={initialValues}
@@ -548,6 +641,42 @@ export function LLMProviderUpdateForm({
             </div>
           )}
 
+          {llmProviderDescriptor.name === "ollama" && (
+            <div className="flex flex-col gap-2">
+              <Button
+                type="button"
+                onClick={() =>
+                  fetchOllamaModels(
+                    formikProps.values,
+                    formikProps.setFieldValue
+                  )
+                }
+                disabled={isFetchingModels || !formikProps.values.api_base}
+                className="w-fit"
+              >
+                {isFetchingModels ? (
+                  <>
+                    <LoadingAnimation size="text-sm" />
+                    <span className="ml-2">Fetching Models...</span>
+                  </>
+                ) : (
+                  "Fetch Available Ollama Models"
+                )}
+              </Button>
+
+              {fetchModelsError && (
+                <Text className="text-red-600 text-sm">{fetchModelsError}</Text>
+              )}
+
+              <Text className="text-sm text-gray-600">
+                Ensure your Ollama server is accessible from Onyx and that the
+                requested models are pulled (e.g. via <code>ollama pull</code>).
+                Provide the server&apos;s base URL and optional API key (when
+                using Ollama Cloud) before fetching the available models.
+              </Text>
+            </div>
+          )}
+
           {!firstTimeConfiguration && (
             <>
               <Separator />
diff --git a/web/src/lib/hooks.ts b/web/src/lib/hooks.ts
index 6a042c5617a..406210496e9 100644
--- a/web/src/lib/hooks.ts
+++ b/web/src/lib/hooks.ts
@@ -955,6 +955,33 @@ const MODEL_DISPLAY_NAMES: { [key: string]: string } = {
   "ai21.jamba-instruct-v1:0": "Jamba Instruct",
   "ai21.j2-ultra-v1": "J2 Ultra",
   "ai21.j2-mid-v1": "J2 Mid",
+
+  // Ollama default models
+  // These can be directly served from the Ollama CLI
+  // User may also add arbitrary open source models
+  "llama3.2": "Llama 3.2",
+  "llama3.2:1b": "Llama 3.2 1B",
+  "llama3.2:3b": "Llama 3.2 3B",
+  "llama3.2:11b": "Llama 3.2 11B",
+  "llama3.2:90b": "Llama 3.2 90B",
+  "llama3.1": "Llama 3.1",
+  "llama3.1:8b": "Llama 3.1 8B",
+  "llama3.1:70b": "Llama 3.1 70B",
+  llama3: "Llama 3",
+  "llama3:8b": "Llama 3 8B",
+  "llama3:70b": "Llama 3 70B",
+  phi3: "Phi-3",
+  "phi3:mini": "Phi-3 Mini",
+  "phi3:medium": "Phi-3 Medium",
+  "phi3:medium-4k": "Phi-3 Medium 4K",
+  "phi3:medium-128k": "Phi-3 Medium 128K",
+  "mistral-small": "Mistral Small",
+  "mistral-large": "Mistral Large",
+  "mistral-nemo": "Mistral Nemo",
+  "smollm2:135m": "SmolLM2 135M",
+  "smollm2:360m": "SmolLM2 360M",
+  "qwen2.5:3b": "Qwen 2.5 3B",
+  "qwen2.5:7b": "Qwen 2.5 7B",
 };
 
 export function getDisplayNameForModel(modelName: string): string {
@@ -993,6 +1020,7 @@ export const defaultModelsByProvider: { [name: string]: string[] } = {
     "anthropic.claude-3-7-sonnet-20250219-v1:0",
   ],
   anthropic: ["claude-3-opus-20240229", "claude-3-5-sonnet-20241022"],
+  ollama: ["llama3.2", "phi3:mini", "mistral-small"],
 };
 
 // Get source metadata for configured sources - deduplicated by source type

From b45a5c9d63de69a53259cf2cfe5c186f00b7c679 Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Sun, 28 Sep 2025 12:37:54 -0700
Subject: [PATCH 02/14] update ollama model names

---
 web/src/lib/hooks.ts | 54 ++++++++++++++++++++++++--------------------
 1 file changed, 30 insertions(+), 24 deletions(-)

diff --git a/web/src/lib/hooks.ts b/web/src/lib/hooks.ts
index 406210496e9..4ab0c8f9a91 100644
--- a/web/src/lib/hooks.ts
+++ b/web/src/lib/hooks.ts
@@ -956,32 +956,38 @@ const MODEL_DISPLAY_NAMES: { [key: string]: string } = {
   "ai21.j2-ultra-v1": "J2 Ultra",
   "ai21.j2-mid-v1": "J2 Mid",
 
-  // Ollama default models
-  // These can be directly served from the Ollama CLI
-  // User may also add arbitrary open source models
-  "llama3.2": "Llama 3.2",
-  "llama3.2:1b": "Llama 3.2 1B",
-  "llama3.2:3b": "Llama 3.2 3B",
-  "llama3.2:11b": "Llama 3.2 11B",
-  "llama3.2:90b": "Llama 3.2 90B",
-  "llama3.1": "Llama 3.1",
-  "llama3.1:8b": "Llama 3.1 8B",
-  "llama3.1:70b": "Llama 3.1 70B",
+  // Ollama cloud models
+  "gpt-oss:20b-cloud": "gpt-oss 20B Cloud",
+  "gpt-oss:120b-cloud": "gpt-oss 120B Cloud",
+  "deepseek-v3.1:671b-cloud": "DeepSeek-v3.1 671B Cloud",
+  "qwen3-coder:480b-cloud": "Qwen3-Coder 480B Cloud",
+
+  // Ollama models in litellm map (disjoint from ollama's supported model list)
+  // https://models.litellm.ai --> provider ollama
+  codegeex4: "CodeGeeX 4",
+  codegemma: "CodeGemma",
+  codellama: "CodeLLama",
+  "deepseek-coder-v2-base": "DeepSeek-Coder-v2 Base",
+  "deepseek-coder-v2-instruct": "DeepSeek-Coder-v2 Instruct",
+  "deepseek-coder-v2-lite-base": "DeepSeek-Coder-v2 Lite Base",
+  "deepseek-coder-v2-lite-instruct": "DeepSeek-Coder-v2 Lite Instruct",
+  "internlm2_5-20b-chat": "InternLM 2.5 20B Chat",
+  llama2: "Llama 2",
+  "llama2-uncensored": "Llama 2 Uncensored",
+  "llama2:13b": "Llama 2 13B",
+  "llama2:70b": "Llama 2 70B",
+  "llama2:7b": "Llama 2 7B",
   llama3: "Llama 3",
-  "llama3:8b": "Llama 3 8B",
   "llama3:70b": "Llama 3 70B",
-  phi3: "Phi-3",
-  "phi3:mini": "Phi-3 Mini",
-  "phi3:medium": "Phi-3 Medium",
-  "phi3:medium-4k": "Phi-3 Medium 4K",
-  "phi3:medium-128k": "Phi-3 Medium 128K",
-  "mistral-small": "Mistral Small",
-  "mistral-large": "Mistral Large",
-  "mistral-nemo": "Mistral Nemo",
-  "smollm2:135m": "SmolLM2 135M",
-  "smollm2:360m": "SmolLM2 360M",
-  "qwen2.5:3b": "Qwen 2.5 3B",
-  "qwen2.5:7b": "Qwen 2.5 7B",
+  "llama3:8b": "Llama 3 8B",
+  mistral: "Mistral", // Mistral 7b
+  "mistral-7B-Instruct-v0.1": "Mistral 7B Instruct v0.1",
+  "mistral-7B-Instruct-v0.2": "Mistral 7B Instruct v0.2",
+  "mistral-large-instruct-2407": "Mistral Large Instruct 24.07",
+  "mixtral-8x22B-Instruct-v0.1": "Mixtral 8x22B Instruct v0.1",
+  "mixtral8x7B-Instruct-v0.1": "Mixtral 8x7B Instruct v0.1",
+  "orca-mini": "Orca Mini",
+  vicuna: "Vicuna",
 };
 
 export function getDisplayNameForModel(modelName: string): string {

From 44fa7bc1663753a308a99baf599553372433f1a6 Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Sun, 28 Sep 2025 17:39:43 -0700
Subject: [PATCH 03/14] handle max tokens and vision capabilities for ollama

---
 ...add_image_input_support_to_model_config.py |  30 +
 backend/onyx/db/llm.py                        |   1 +
 backend/onyx/db/models.py                     |   3 +
 backend/onyx/llm/factory.py                   |  22 +-
 backend/onyx/llm/utils.py                     |  25 +
 backend/onyx/server/manage/llm/api.py         | 128 +--
 backend/onyx/server/manage/llm/models.py      |  26 +-
 .../configuration/llm/FetchModelsButton.tsx   | 112 +++
 .../llm/LLMProviderUpdateForm.tsx             | 750 ++++++------------
 .../app/admin/configuration/llm/interfaces.ts |   5 +-
 web/src/app/admin/configuration/llm/utils.ts  | 170 ++++
 web/src/lib/hooks.ts                          |   1 +
 12 files changed, 710 insertions(+), 563 deletions(-)
 create mode 100644 backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py
 create mode 100644 web/src/app/admin/configuration/llm/FetchModelsButton.tsx

diff --git a/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py b/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py
new file mode 100644
index 00000000000..02a8c9a2a43
--- /dev/null
+++ b/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py
@@ -0,0 +1,30 @@
+"""Add image input support to model config
+
+Revision ID: 64bd5677aeb6
+Revises: 2b75d0a8ffcb
+Create Date: 2025-09-28 15:48:12.003612
+
+"""
+
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision = "64bd5677aeb6"
+down_revision = "2b75d0a8ffcb"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "model_configuration",
+        sa.Column("supports_image_input", sa.Boolean(), nullable=True),
+    )
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_column("model_configuration", "supports_image_input")
+    # ### end Alembic commands ###
diff --git a/backend/onyx/db/llm.py b/backend/onyx/db/llm.py
index e1c49349139..42e6db57077 100644
--- a/backend/onyx/db/llm.py
+++ b/backend/onyx/db/llm.py
@@ -112,6 +112,7 @@ def upsert_llm_provider(
                 name=model_configuration.name,
                 is_visible=model_configuration.is_visible,
                 max_input_tokens=model_configuration.max_input_tokens,
+                supports_image_input=model_configuration.supports_image_input,
             )
             .on_conflict_do_nothing()
         )
diff --git a/backend/onyx/db/models.py b/backend/onyx/db/models.py
index f897bcdca42..89d9b2c513e 100644
--- a/backend/onyx/db/models.py
+++ b/backend/onyx/db/models.py
@@ -2353,6 +2353,9 @@ class ModelConfiguration(Base):
     # - The end-user is configuring a model and chooses not to set a max-input-tokens limit.
     max_input_tokens: Mapped[int | None] = mapped_column(Integer, nullable=True)
 
+    # Whether this model supports image input
+    supports_image_input: Mapped[bool | None] = mapped_column(Boolean, nullable=True)
+
     llm_provider: Mapped["LLMProvider"] = relationship(
         "LLMProvider",
         back_populates="model_configurations",
diff --git a/backend/onyx/llm/factory.py b/backend/onyx/llm/factory.py
index 6f6a07da4b6..5059f44c814 100644
--- a/backend/onyx/llm/factory.py
+++ b/backend/onyx/llm/factory.py
@@ -1,8 +1,5 @@
-from typing import Any
-
 from onyx.chat.models import PersonaOverrideConfig
 from onyx.configs.app_configs import DISABLE_GENERATIVE_AI
-from onyx.configs.model_configs import GEN_AI_MODEL_FALLBACK_MAX_TOKENS
 from onyx.configs.model_configs import GEN_AI_TEMPERATURE
 from onyx.db.engine.sql_engine import get_session_with_current_tenant
 from onyx.db.llm import fetch_default_provider
@@ -26,19 +23,6 @@
 logger = setup_logger()
 
 
-def _build_extra_model_kwargs(provider: str) -> dict[str, Any]:
-    """Ollama requires us to specify the max context window.
-
-    For now, just using the GEN_AI_MODEL_FALLBACK_MAX_TOKENS value.
-    TODO: allow model-specific values to be configured via the UI.
-    """
-    return (
-        {"num_ctx": GEN_AI_MODEL_FALLBACK_MAX_TOKENS}
-        if provider == OLLAMA_PROVIDER_NAME
-        else {}
-    )
-
-
 def _build_provider_extra_headers(
     provider: str, custom_config: dict[str, str] | None
 ) -> dict[str, str]:
@@ -298,6 +282,10 @@ def get_llm(
         temperature = GEN_AI_TEMPERATURE
 
     extra_headers = build_llm_extra_headers(additional_headers)
+
+    # NOTE: this is needed since Ollama API key is optional
+    # User may access Ollama cloud via locally hosted instance (logged in)
+    # or just via the cloud API (not logged in)
     provider_extra_headers = _build_provider_extra_headers(provider, custom_config)
     if provider_extra_headers:
         extra_headers.update(provider_extra_headers)
@@ -313,7 +301,7 @@ def get_llm(
         temperature=temperature,
         custom_config=custom_config,
         extra_headers=extra_headers,
-        model_kwargs=_build_extra_model_kwargs(provider),
+        model_kwargs={},
         long_term_logger=long_term_logger,
         max_input_tokens=max_input_tokens,
     )
diff --git a/backend/onyx/llm/utils.py b/backend/onyx/llm/utils.py
index ffcfb46bae4..b0d1d3419b2 100644
--- a/backend/onyx/llm/utils.py
+++ b/backend/onyx/llm/utils.py
@@ -16,6 +16,7 @@
 from langchain.schema.messages import BaseMessage
 from langchain.schema.messages import HumanMessage
 from langchain.schema.messages import SystemMessage
+from sqlalchemy import select
 
 from onyx.configs.app_configs import LITELLM_CUSTOM_ERROR_MESSAGE_MAPPINGS
 from onyx.configs.app_configs import MAX_TOKENS_FOR_FULL_INCLUSION
@@ -26,6 +27,8 @@
 from onyx.configs.model_configs import GEN_AI_MAX_TOKENS
 from onyx.configs.model_configs import GEN_AI_MODEL_FALLBACK_MAX_TOKENS
 from onyx.configs.model_configs import GEN_AI_NUM_RESERVED_OUTPUT_TOKENS
+from onyx.db.engine.sql_engine import get_session_with_current_tenant
+from onyx.db.models import ModelConfiguration
 from onyx.file_store.models import ChatFileType
 from onyx.file_store.models import InMemoryChatFile
 from onyx.llm.interfaces import LLM
@@ -37,6 +40,8 @@
 from onyx.utils.logger import setup_logger
 from shared_configs.configs import LOG_LEVEL
 
+# Database imports for OLLAMA provider check
+
 
 if TYPE_CHECKING:
     from onyx.server.manage.llm.models import LLMProviderView
@@ -640,6 +645,26 @@ def get_max_input_tokens_from_llm_provider(
 
 
 def model_supports_image_input(model_name: str, model_provider: str) -> bool:
+    # TODO: Add support to check model config for any provider
+    # TODO: Circular import means OLLAMA_PROVIDER_NAME is not available here
+    # NOTE: Theoretically, if user has model configured through both Ollama and another provider,
+    # this may erroenously return False because this checks just one database entry
+    # and it may be the other provider's entry, which would be None
+    if model_provider == "ollama":
+        try:
+            with get_session_with_current_tenant() as db_session:
+                model_config = db_session.scalar(
+                    select(ModelConfiguration).where(
+                        ModelConfiguration.name == model_name
+                    )
+                )
+                if model_config and model_config.supports_image_input is not None:
+                    return model_config.supports_image_input
+        except Exception as e:
+            logger.warning(
+                f"Failed to query database for Ollama model {model_name} image support: {e}"
+            )
+
     model_map = get_model_map()
     try:
         model_obj = find_model_obj(
diff --git a/backend/onyx/server/manage/llm/api.py b/backend/onyx/server/manage/llm/api.py
index ff3aefdb1ba..9809a7aa659 100644
--- a/backend/onyx/server/manage/llm/api.py
+++ b/backend/onyx/server/manage/llm/api.py
@@ -12,10 +12,12 @@
 from fastapi import Depends
 from fastapi import HTTPException
 from fastapi import Query
+from pydantic import ValidationError
 from sqlalchemy.orm import Session
 
 from onyx.auth.users import current_admin_user
 from onyx.auth.users import current_chat_accessible_user
+from onyx.configs.model_configs import GEN_AI_MODEL_FALLBACK_MAX_TOKENS
 from onyx.db.engine.sql_engine import get_session
 from onyx.db.llm import fetch_existing_llm_provider
 from onyx.db.llm import fetch_existing_llm_providers
@@ -41,6 +43,8 @@
 from onyx.server.manage.llm.models import LLMProviderUpsertRequest
 from onyx.server.manage.llm.models import LLMProviderView
 from onyx.server.manage.llm.models import ModelConfigurationUpsertRequest
+from onyx.server.manage.llm.models import OllamaModelDetails
+from onyx.server.manage.llm.models import OllamaModelResponse
 from onyx.server.manage.llm.models import OllamaModelsRequest
 from onyx.server.manage.llm.models import TestLLMRequest
 from onyx.server.manage.llm.models import VisionProviderResponse
@@ -478,79 +482,97 @@ def get_bedrock_available_models(
         )
 
 
-def _extract_model_names(response_json: dict) -> set[str]:
-    models = response_json.get("models") or response_json.get("model") or []
-    if not isinstance(models, list):
-        return set()
+def _get_ollama_available_model_names(api_base: str) -> set[str]:
+    """Fetch available model names from Ollama server."""
+    tags_url = f"{api_base}/api/tags"
+    try:
+        response = httpx.get(tags_url, timeout=5.0)
+        response.raise_for_status()
+        response_json = response.json()
+    except Exception as e:
+        raise HTTPException(
+            status_code=400,
+            detail=f"Failed to fetch Ollama models: {e}",
+        )
 
-    model_names: set[str] = set()
-    for model in models:
-        if not isinstance(model, dict):
-            continue
-        candidate = model.get("model") or model.get("name")
-        if isinstance(candidate, str) and candidate:
-            model_names.add(candidate)
-    return model_names
+    models = response_json.get("models", [])
+    return {model.get("name") for model in models if model.get("name")}
 
 
 @admin_router.post("/ollama/available-models")
 def get_ollama_available_models(
     request: OllamaModelsRequest,
     _: User | None = Depends(current_admin_user),
-) -> list[str]:
+) -> list[OllamaModelResponse]:
     """Fetch the list of available models from an Ollama server."""
 
-    cleaned_api_base = request.api_base.rstrip("/")
+    cleaned_api_base = request.api_base.strip().rstrip("/")
     if not cleaned_api_base:
         raise HTTPException(
             status_code=400, detail="API base URL is required to fetch Ollama models."
         )
 
-    headers: dict[str, str] = {}
+    model_names = _get_ollama_available_model_names(cleaned_api_base)
+    if not model_names:
+        raise HTTPException(
+            status_code=400,
+            detail="No models found from your Ollama server",
+        )
 
-    endpoints = ["api/tags", "api/models"]
-    last_error: Exception | None = None
+    models_with_context_size: list[OllamaModelResponse] = []
+    show_url = f"{cleaned_api_base}/api/show"
 
-    for endpoint in endpoints:
-        url = f"{cleaned_api_base}/{endpoint}"
+    for model_name in model_names:
+        context_limit: int | None = None
         try:
-            response = httpx.get(url, headers=headers, timeout=10.0)
-            response.raise_for_status()
-            try:
-                response_json = response.json()
-            except ValueError as e:
-                last_error = e
-                continue
+            show_response = httpx.post(
+                show_url,
+                json={"model": model_name},
+                timeout=5.0,
+            )
+            show_response.raise_for_status()
+            show_response_json = show_response.json()
 
-            model_names = _extract_model_names(response_json)
-            if model_names:
-                return sorted(model_names)
+            # Parse the response into the expected format
+            ollama_model_details = OllamaModelDetails.model_validate(show_response_json)
 
-            last_error = ValueError("No models returned from Ollama response")
-        except httpx.HTTPStatusError as e:
-            last_error = e
-            # If Ollama doesn't support the endpoint, try the next one
-            if e.response.status_code in {404, 405}:
+            # Check if this model supports completion/chat
+            if not ollama_model_details.supports_completion():
                 continue
-            raise HTTPException(
-                status_code=e.response.status_code,
-                detail=(
-                    f"Failed to fetch Ollama models: "
-                    f"{e.response.text or e.response.reason_phrase}"
-                ),
+
+            # Optimistically access. Context limit is stored as "model_architecture.context" = int
+            architecture = ollama_model_details.model_info.get(
+                "general.architecture", ""
             )
-        except httpx.RequestError as e:
-            raise HTTPException(
-                status_code=400,
-                detail=f"Failed to reach Ollama server at {cleaned_api_base}: {e}",
+            context_limit = ollama_model_details.model_info.get(
+                architecture + ".context_length", None
+            )
+            supports_image_input = ollama_model_details.supports_image_input()
+        except ValidationError as e:
+            logger.warning(
+                "Invalid model details from Ollama server",
+                extra={"model": model_name, "validation_error": str(e)},
+            )
+        except Exception as e:
+            logger.warning(
+                "Failed to fetch Ollama model details",
+                extra={"model": model_name, "error": str(e)},
             )
 
-    error_detail = (
-        str(last_error)
-        if last_error
-        else "Unexpected response from Ollama when listing models"
-    )
-    raise HTTPException(
-        status_code=400,
-        detail=f"Failed to fetch Ollama models: {error_detail}",
-    )
+        # If we fail at any point attempting to extract context limit,
+        # still allow this model to be used with a fallback max context size
+        if not context_limit:
+            context_limit = GEN_AI_MODEL_FALLBACK_MAX_TOKENS
+
+        if not supports_image_input:
+            supports_image_input = False
+
+        models_with_context_size.append(
+            OllamaModelResponse(
+                name=model_name,
+                max_input_tokens=context_limit,
+                supports_image_input=supports_image_input,
+            )
+        )
+
+    return models_with_context_size
diff --git a/backend/onyx/server/manage/llm/models.py b/backend/onyx/server/manage/llm/models.py
index 83de2809b57..c63141c6107 100644
--- a/backend/onyx/server/manage/llm/models.py
+++ b/backend/onyx/server/manage/llm/models.py
@@ -1,3 +1,4 @@
+from typing import Any
 from typing import TYPE_CHECKING
 
 from pydantic import BaseModel
@@ -140,6 +141,7 @@ class ModelConfigurationUpsertRequest(BaseModel):
     name: str
     is_visible: bool | None = False
     max_input_tokens: int | None = None
+    supports_image_input: bool | None = None
 
     @classmethod
     def from_model(
@@ -149,6 +151,7 @@ def from_model(
             name=model_configuration_model.name,
             is_visible=model_configuration_model.is_visible,
             max_input_tokens=model_configuration_model.max_input_tokens,
+            supports_image_input=model_configuration_model.supports_image_input,
         )
 
 
@@ -200,5 +203,24 @@ class BedrockModelsRequest(BaseModel):
 
 class OllamaModelsRequest(BaseModel):
     api_base: str
-    api_key: str | None = None
-    provider_name: str | None = None
+
+
+class OllamaModelResponse(BaseModel):
+    name: str
+    max_input_tokens: int
+    supports_image_input: bool
+
+
+class OllamaModelDetails(BaseModel):
+    """Response model for Ollama /api/show endpoint"""
+
+    model_info: dict[str, Any]
+    capabilities: list[str] = []
+
+    def supports_completion(self) -> bool:
+        """Check if this model supports completion/chat"""
+        return "completion" in self.capabilities
+
+    def supports_image_input(self) -> bool:
+        """Check if this model supports image input"""
+        return "vision" in self.capabilities
diff --git a/web/src/app/admin/configuration/llm/FetchModelsButton.tsx b/web/src/app/admin/configuration/llm/FetchModelsButton.tsx
new file mode 100644
index 00000000000..1c1ea6a3b85
--- /dev/null
+++ b/web/src/app/admin/configuration/llm/FetchModelsButton.tsx
@@ -0,0 +1,112 @@
+import { Button } from "@/components/ui/button";
+import { LoadingAnimation } from "@/components/Loading";
+import Text from "@/components/ui/text";
+import { fetchModels } from "./utils";
+import { WellKnownLLMProviderDescriptor, LLMProviderView } from "./interfaces";
+import { PopupSpec } from "@/components/admin/connectors/Popup";
+
+interface FetchModelsButtonProps {
+  llmProviderDescriptor: WellKnownLLMProviderDescriptor;
+  existingLlmProvider?: LLMProviderView;
+  values: any;
+  setFieldValue: any;
+  isFetchingModels: boolean;
+  setIsFetchingModels: (loading: boolean) => void;
+  fetchModelsError: string;
+  setFetchModelsError: (error: string) => void;
+  setPopup?: (popup: PopupSpec) => void;
+}
+
+interface ProviderConfig {
+  buttonText: string;
+  loadingText: string;
+  helperText: string | React.ReactNode;
+  isDisabled: (values: any) => boolean;
+}
+
+const providerConfigs: Record<string, ProviderConfig> = {
+  bedrock: {
+    buttonText: "Fetch Available Models for Region",
+    loadingText: "Fetching Models...",
+    helperText: (
+      <>
+        Enter your AWS region, then click this button to fetch available Bedrock
+        models.
+        <br />
+        If you&apos;re updating your existing provider, you&apos;ll need to
+        click this button to fetch the latest models.
+      </>
+    ),
+    isDisabled: (values) => !values.custom_config?.AWS_REGION_NAME,
+  },
+  ollama: {
+    buttonText: "Fetch Available Ollama Models",
+    loadingText: "Fetching Models...",
+    helperText: (
+      <>
+        Ensure your Ollama server is accessible from Onyx and that the requested
+        models are pulled (e.g. via <code>ollama pull</code>). Provide the
+        server&apos;s base URL and optional API key (when using Ollama Cloud)
+        before fetching the available models.
+      </>
+    ),
+    isDisabled: (values) => !values.api_base,
+  },
+};
+
+export function FetchModelsButton({
+  llmProviderDescriptor,
+  existingLlmProvider,
+  values,
+  setFieldValue,
+  isFetchingModels,
+  setIsFetchingModels,
+  fetchModelsError,
+  setFetchModelsError,
+  setPopup,
+}: FetchModelsButtonProps) {
+  const config = providerConfigs[llmProviderDescriptor.name];
+
+  // Only render if the provider supports model fetching
+  if (!config) {
+    return null;
+  }
+
+  const handleFetchModels = () => {
+    fetchModels(
+      llmProviderDescriptor,
+      existingLlmProvider,
+      values,
+      setFieldValue,
+      setIsFetchingModels,
+      setFetchModelsError,
+      setPopup
+    );
+  };
+
+  return (
+    <div className="flex flex-col gap-2">
+      <Button
+        type="button"
+        onClick={handleFetchModels}
+        disabled={isFetchingModels || config.isDisabled(values)}
+        className="w-fit"
+      >
+        {isFetchingModels ? (
+          <>
+            <LoadingAnimation size="text-sm" />
+            <span className="ml-2">{config.loadingText}</span>
+          </>
+        ) : (
+          config.buttonText
+        )}
+      </Button>
+
+      {fetchModelsError && (
+        <Text className="text-red-600 text-sm">{fetchModelsError}</Text>
+      )}
+
+      <Text className="text-sm text-gray-600">{config.helperText}</Text>
+    </div>
+  );
+}
diff --git a/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx b/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx
index e38372d1048..f9c107f7db0 100644
--- a/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx
+++ b/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx
@@ -17,9 +17,10 @@ import { useState } from "react";
 import { useSWRConfig } from "swr";
 import {
   LLMProviderView,
-  ModelConfigurationUpsertRequest,
+  ModelConfiguration,
   WellKnownLLMProviderDescriptor,
 } from "./interfaces";
+import { FetchModelsButton } from "./FetchModelsButton";
 import { PopupSpec } from "@/components/admin/connectors/Popup";
 import * as Yup from "yup";
 import isEqual from "lodash/isEqual";
@@ -53,6 +54,13 @@ export function LLMProviderUpdateForm({
 
   const [showAdvancedOptions, setShowAdvancedOptions] = useState(false);
 
+  // Helper function to get current model configurations
+  const getCurrentModelConfigurations = (values: any): ModelConfiguration[] => {
+    return values.fetched_model_configurations?.length > 0
+      ? values.fetched_model_configurations
+      : llmProviderDescriptor.model_configurations;
+  };
+
   // Define the initial values based on the provider's requirements
   const initialValues = {
     name:
@@ -102,6 +110,9 @@ export function LLMProviderUpdateForm({
           .filter((modelConfiguration) => modelConfiguration.is_visible)
           .map((modelConfiguration) => modelConfiguration.name) as string[]),
 
+    // Store fetched model configurations in form state instead of mutating props
+    fetched_model_configurations: [] as ModelConfiguration[],
+
     // Helper field to force re-renders when model list updates
     _modelListUpdated: 0,
   };
@@ -176,6 +187,7 @@ export function LLMProviderUpdateForm({
     is_public: Yup.boolean().required(),
     groups: Yup.array().of(Yup.number()),
     selected_model_names: Yup.array().of(Yup.string()),
+    fetched_model_configurations: Yup.array(),
   });
 
   const customLinkRenderer = ({ href, children }: any) => {
@@ -186,190 +198,6 @@ export function LLMProviderUpdateForm({
     );
   };
 
-  const fetchBedrockModels = async (values: any, setFieldValue: any) => {
-    if (llmProviderDescriptor.name !== "bedrock") {
-      return;
-    }
-
-    setIsFetchingModels(true);
-    setFetchModelsError("");
-
-    try {
-      const response = await fetch("/api/admin/llm/bedrock/available-models", {
-        method: "POST",
-        headers: {
-          "Content-Type": "application/json",
-        },
-        body: JSON.stringify({
-          aws_region_name: values.custom_config?.AWS_REGION_NAME,
-          aws_access_key_id: values.custom_config?.AWS_ACCESS_KEY_ID,
-          aws_secret_access_key: values.custom_config?.AWS_SECRET_ACCESS_KEY,
-          aws_bearer_token_bedrock:
-            values.custom_config?.AWS_BEARER_TOKEN_BEDROCK,
-          provider_name: existingLlmProvider?.name, // Save models to existing provider if editing
-        }),
-      });
-
-      if (!response.ok) {
-        const errorData = await response.json();
-        throw new Error(errorData.detail || "Failed to fetch models");
-      }
-
-      const availableModels: string[] = await response.json();
-
-      // Update the model configurations with the fetched models
-      const updatedModelConfigs = availableModels.map((modelName) => {
-        // Find existing configuration to preserve is_visible setting
-        const existingConfig = llmProviderDescriptor.model_configurations.find(
-          (config) => config.name === modelName
-        );
-
-        return {
-          name: modelName,
-          is_visible: existingConfig?.is_visible ?? false, // Preserve existing visibility or default to false
-          max_input_tokens: null,
-          supports_image_input: false, // Will be determined by the backend
-        };
-      });
-
-      // Update the descriptor and form values
-      llmProviderDescriptor.model_configurations = updatedModelConfigs;
-
-      // Update selected model names to only include previously visible models that are available
-      const previouslySelectedModels = values.selected_model_names || [];
-      const stillAvailableSelectedModels = previouslySelectedModels.filter(
-        (modelName: string) => availableModels.includes(modelName)
-      );
-      setFieldValue("selected_model_names", stillAvailableSelectedModels);
-
-      // Set a default model if none is set
-      if (
-        (!values.default_model_name ||
-          !availableModels.includes(values.default_model_name)) &&
-        availableModels.length > 0
-      ) {
-        setFieldValue("default_model_name", availableModels[0]);
-      }
-
-      // Clear fast model if it's not in the new list
-      if (
-        values.fast_default_model_name &&
-        !availableModels.includes(values.fast_default_model_name)
-      ) {
-        setFieldValue("fast_default_model_name", null);
-      }
-
-      // Force a re-render by updating a timestamp or counter
-      setFieldValue("_modelListUpdated", Date.now());
-
-      setPopup?.({
-        message: `Successfully fetched ${availableModels.length} models for the selected region (including cross-region inference models).`,
-        type: "success",
-      });
-    } catch (error) {
-      const errorMessage =
-        error instanceof Error ? error.message : "Unknown error";
-      setFetchModelsError(errorMessage);
-      setPopup?.({
-        message: `Failed to fetch models: ${errorMessage}`,
-        type: "error",
-      });
-    } finally {
-      setIsFetchingModels(false);
-    }
-  };
-
-  const fetchOllamaModels = async (values: any, setFieldValue: any) => {
-    if (!values.api_base) {
-      setFetchModelsError("API Base is required to fetch Ollama models");
-      return;
-    }
-
-    setIsFetchingModels(true);
-    setFetchModelsError("");
-
-    try {
-      const response = await fetch("/api/admin/llm/ollama/available-models", {
-        method: "POST",
-        headers: {
-          "Content-Type": "application/json",
-        },
-        body: JSON.stringify({
-          api_base: values.api_base,
-          provider_name: existingLlmProvider?.name,
-        }),
-      });
-
-      if (!response.ok) {
-        let errorMessage = "Failed to fetch models";
-        try {
-          const errorData = await response.json();
-          errorMessage = errorData.detail || errorMessage;
-        } catch {
-          // ignore JSON parsing errors and use the fallback message
-        }
-
-        throw new Error(errorMessage);
-      }
-
-      const availableModels: string[] = await response.json();
-
-      const updatedModelConfigs = availableModels.map((modelName) => {
-        const existingConfig = llmProviderDescriptor.model_configurations.find(
-          (config) => config.name === modelName
-        );
-
-        return {
-          name: modelName,
-          is_visible: existingConfig?.is_visible ?? false,
-          max_input_tokens: null,
-          supports_image_input: false,
-        };
-      });
-
-      llmProviderDescriptor.model_configurations = updatedModelConfigs;
-
-      const previouslySelectedModels = values.selected_model_names || [];
-      const stillAvailableSelectedModels = previouslySelectedModels.filter(
-        (modelName: string) => availableModels.includes(modelName)
-      );
-
-      setFieldValue("selected_model_names", stillAvailableSelectedModels);
-
-      if (
-        (!values.default_model_name ||
-          !availableModels.includes(values.default_model_name)) &&
-        availableModels.length > 0
-      ) {
-        setFieldValue("default_model_name", availableModels[0]);
-      }
-
-      if (
-        values.fast_default_model_name &&
-        !availableModels.includes(values.fast_default_model_name)
-      ) {
-        setFieldValue("fast_default_model_name", null);
-      }
-
-      setFieldValue("_modelListUpdated", Date.now());
-
-      setPopup?.({
-        message: `Successfully fetched ${availableModels.length} models from Ollama.`,
-        type: "success",
-      });
-    } catch (error) {
-      const errorMessage =
-        error instanceof Error ? error.message : "Unknown error";
-      setFetchModelsError(errorMessage);
-      setPopup?.({
-        message: `Failed to fetch models: ${errorMessage}`,
-        type: "error",
-      });
-    } finally {
-      setIsFetchingModels(false);
-    }
-  };
-
   return (
     <Formik
       initialValues={initialValues}
@@ -407,11 +235,12 @@ export function LLMProviderUpdateForm({
           api_base: finalApiBase,
           api_version: finalApiVersion,
           api_key_changed: values.api_key !== initialValues.api_key,
-          model_configurations: llmProviderDescriptor.model_configurations.map(
-            (modelConfiguration): ModelConfigurationUpsertRequest => ({
+          model_configurations: getCurrentModelConfigurations(values).map(
+            (modelConfiguration): ModelConfiguration => ({
               name: modelConfiguration.name,
               is_visible: visibleModels.includes(modelConfiguration.name),
-              max_input_tokens: null,
+              max_input_tokens: modelConfiguration.max_input_tokens ?? null,
+              supports_image_input: modelConfiguration.supports_image_input,
             })
           ),
         };
@@ -515,211 +344,124 @@ export function LLMProviderUpdateForm({
         setSubmitting(false);
       }}
     >
-      {(formikProps) => (
-        <Form className="gap-y-4 items-stretch mt-6">
-          {!firstTimeConfiguration && (
-            <TextFormField
-              name="name"
-              label="Display Name"
-              subtext="A name which you can use to identify this provider when selecting it in the UI."
-              placeholder="Display Name"
-              disabled={existingLlmProvider ? true : false}
-            />
-          )}
-
-          {llmProviderDescriptor.api_key_required && (
-            <TextFormField
-              small={firstTimeConfiguration}
-              name="api_key"
-              label="API Key"
-              placeholder="API Key"
-              type="password"
-            />
-          )}
-
-          {llmProviderDescriptor.name === "azure" ? (
-            <TextFormField
-              small={firstTimeConfiguration}
-              name="target_uri"
-              label="Target URI"
-              placeholder="https://your-resource.cognitiveservices.azure.com/openai/deployments/deployment-name/chat/completions?api-version=2025-01-01-preview"
-              subtext="The complete Azure OpenAI endpoint URL including the API version as a query parameter"
-            />
-          ) : (
-            <>
-              {llmProviderDescriptor.api_base_required && (
-                <TextFormField
-                  small={firstTimeConfiguration}
-                  name="api_base"
-                  label="API Base"
-                  placeholder="API Base"
-                />
-              )}
+      {(formikProps) => {
+        // Get current model configurations for this render
+        const currentModelConfigurations = getCurrentModelConfigurations(
+          formikProps.values
+        );
 
-              {llmProviderDescriptor.api_version_required && (
-                <TextFormField
-                  small={firstTimeConfiguration}
-                  name="api_version"
-                  label="API Version"
-                  placeholder="API Version"
-                />
-              )}
-            </>
-          )}
+        return (
+          <Form className="gap-y-4 items-stretch mt-6">
+            {!firstTimeConfiguration && (
+              <TextFormField
+                name="name"
+                label="Display Name"
+                subtext="A name which you can use to identify this provider when selecting it in the UI."
+                placeholder="Display Name"
+                disabled={existingLlmProvider ? true : false}
+              />
+            )}
 
-          {llmProviderDescriptor.custom_config_keys?.map((customConfigKey) => {
-            if (customConfigKey.key_type === "text_input") {
-              return (
-                <div key={customConfigKey.name}>
+            {llmProviderDescriptor.api_key_required && (
+              <TextFormField
+                small={firstTimeConfiguration}
+                name="api_key"
+                label="API Key"
+                placeholder="API Key"
+                type="password"
+              />
+            )}
+
+            {llmProviderDescriptor.name === "azure" ? (
+              <TextFormField
+                small={firstTimeConfiguration}
+                name="target_uri"
+                label="Target URI"
+                placeholder="https://your-resource.cognitiveservices.azure.com/openai/deployments/deployment-name/chat/completions?api-version=2025-01-01-preview"
+                subtext="The complete Azure OpenAI endpoint URL including the API version as a query parameter"
+              />
+            ) : (
+              <>
+                {llmProviderDescriptor.api_base_required && (
                   <TextFormField
                     small={firstTimeConfiguration}
-                    name={`custom_config.${customConfigKey.name}`}
-                    optional={!customConfigKey.is_required}
-                    label={customConfigKey.display_name}
-                    subtext={
-                      <ReactMarkdown components={{ a: customLinkRenderer }}>
-                        {customConfigKey.description}
-                      </ReactMarkdown>
-                    }
-                    placeholder={customConfigKey.default_value || undefined}
-                    type={customConfigKey.is_secret ? "password" : "text"}
+                    name="api_base"
+                    label="API Base"
+                    placeholder="API Base"
                   />
-                </div>
-              );
-            } else if (customConfigKey.key_type === "file_input") {
-              return (
-                <FileUploadFormField
-                  key={customConfigKey.name}
-                  name={`custom_config.${customConfigKey.name}`}
-                  label={customConfigKey.display_name}
-                  subtext={customConfigKey.description || undefined}
-                />
-              );
-            } else {
-              throw new Error("Unreachable; there should only exist 2 options");
-            }
-          })}
-
-          {/* Bedrock-specific fetch models button */}
-          {llmProviderDescriptor.name === "bedrock" && (
-            <div className="flex flex-col gap-2">
-              <Button
-                type="button"
-                onClick={() =>
-                  fetchBedrockModels(
-                    formikProps.values,
-                    formikProps.setFieldValue
-                  )
-                }
-                disabled={
-                  isFetchingModels ||
-                  !formikProps.values.custom_config?.AWS_REGION_NAME
-                }
-                className="w-fit"
-              >
-                {isFetchingModels ? (
-                  <>
-                    <LoadingAnimation size="text-sm" />
-                    <span className="ml-2">Fetching Models...</span>
-                  </>
-                ) : (
-                  "Fetch Available Models for Region"
                 )}
-              </Button>
-
-              {fetchModelsError && (
-                <Text className="text-red-600 text-sm">{fetchModelsError}</Text>
-              )}
 
-              <Text className="text-sm text-gray-600">
-                Enter your AWS region, then click this button to fetch available
-                Bedrock models.
-                <br />
-                If you&apos;re updating your existing provider, you&apos;ll need
-                to click this button to fetch the latest models.
-              </Text>
-            </div>
-          )}
-
-          {llmProviderDescriptor.name === "ollama" && (
-            <div className="flex flex-col gap-2">
-              <Button
-                type="button"
-                onClick={() =>
-                  fetchOllamaModels(
-                    formikProps.values,
-                    formikProps.setFieldValue
-                  )
-                }
-                disabled={isFetchingModels || !formikProps.values.api_base}
-                className="w-fit"
-              >
-                {isFetchingModels ? (
-                  <>
-                    <LoadingAnimation size="text-sm" />
-                    <span className="ml-2">Fetching Models...</span>
-                  </>
-                ) : (
-                  "Fetch Available Ollama Models"
+                {llmProviderDescriptor.api_version_required && (
+                  <TextFormField
+                    small={firstTimeConfiguration}
+                    name="api_version"
+                    label="API Version"
+                    placeholder="API Version"
+                  />
                 )}
-              </Button>
+              </>
+            )}
 
-              {fetchModelsError && (
-                <Text className="text-red-600 text-sm">{fetchModelsError}</Text>
-              )}
+            {llmProviderDescriptor.custom_config_keys?.map(
+              (customConfigKey) => {
+                if (customConfigKey.key_type === "text_input") {
+                  return (
+                    <div key={customConfigKey.name}>
+                      <TextFormField
+                        small={firstTimeConfiguration}
+                        name={`custom_config.${customConfigKey.name}`}
+                        optional={!customConfigKey.is_required}
+                        label={customConfigKey.display_name}
+                        subtext={
+                          <ReactMarkdown components={{ a: customLinkRenderer }}>
+                            {customConfigKey.description}
+                          </ReactMarkdown>
+                        }
+                        placeholder={customConfigKey.default_value || undefined}
+                        type={customConfigKey.is_secret ? "password" : "text"}
+                      />
+                    </div>
+                  );
+                } else if (customConfigKey.key_type === "file_input") {
+                  return (
+                    <FileUploadFormField
+                      key={customConfigKey.name}
+                      name={`custom_config.${customConfigKey.name}`}
+                      label={customConfigKey.display_name}
+                      subtext={customConfigKey.description || undefined}
+                    />
+                  );
+                } else {
+                  throw new Error(
+                    "Unreachable; there should only exist 2 options"
+                  );
+                }
+              }
+            )}
 
-              <Text className="text-sm text-gray-600">
-                Ensure your Ollama server is accessible from Onyx and that the
-                requested models are pulled (e.g. via <code>ollama pull</code>).
-                Provide the server&apos;s base URL and optional API key (when
-                using Ollama Cloud) before fetching the available models.
-              </Text>
-            </div>
-          )}
-
-          {!firstTimeConfiguration && (
-            <>
-              <Separator />
-
-              {llmProviderDescriptor.model_configurations.length > 0 ? (
-                <SelectorFormField
-                  name="default_model_name"
-                  subtext="The model to use by default for this provider unless otherwise specified."
-                  label="Default Model"
-                  options={llmProviderDescriptor.model_configurations.map(
-                    (modelConfiguration) => ({
-                      // don't clean up names here to give admins descriptive names / handle duplicates
-                      // like us.anthropic.claude-3-7-sonnet-20250219-v1:0 and anthropic.claude-3-7-sonnet-20250219-v1:0
-                      name: modelConfiguration.name,
-                      value: modelConfiguration.name,
-                    })
-                  )}
-                  maxHeight="max-h-56"
-                />
-              ) : (
-                <TextFormField
-                  name="default_model_name"
-                  subtext="The model to use by default for this provider unless otherwise specified."
-                  label="Default Model"
-                  placeholder="E.g. gpt-4"
-                />
-              )}
+            {/* Fetch models button - automatically shows for supported providers */}
+            <FetchModelsButton
+              llmProviderDescriptor={llmProviderDescriptor}
+              existingLlmProvider={existingLlmProvider}
+              values={formikProps.values}
+              setFieldValue={formikProps.setFieldValue}
+              isFetchingModels={isFetchingModels}
+              setIsFetchingModels={setIsFetchingModels}
+              fetchModelsError={fetchModelsError}
+              setFetchModelsError={setFetchModelsError}
+              setPopup={setPopup}
+            />
 
-              {llmProviderDescriptor.deployment_name_required && (
-                <TextFormField
-                  name="deployment_name"
-                  label="Deployment Name"
-                  placeholder="Deployment Name"
-                />
-              )}
+            {!firstTimeConfiguration && (
+              <>
+                <Separator />
 
-              {!llmProviderDescriptor.single_model_supported &&
-                (llmProviderDescriptor.model_configurations.length > 0 ? (
+                {currentModelConfigurations.length > 0 ? (
                   <SelectorFormField
-                    name="fast_default_model_name"
-                    subtext="The model to use for lighter flows like `LLM Chunk Filter` for this provider. If not set, will use the Default Model configured above."
-                    label="[Optional] Fast Model"
-                    options={llmProviderDescriptor.model_configurations.map(
+                    name="default_model_name"
+                    subtext="The model to use by default for this provider unless otherwise specified."
+                    label="Default Model"
+                    options={currentModelConfigurations.map(
                       (modelConfiguration) => ({
                         // don't clean up names here to give admins descriptive names / handle duplicates
                         // like us.anthropic.claude-3-7-sonnet-20250219-v1:0 and anthropic.claude-3-7-sonnet-20250219-v1:0
@@ -727,127 +469,161 @@ export function LLMProviderUpdateForm({
                         value: modelConfiguration.name,
                       })
                     )}
-                    includeDefault
                     maxHeight="max-h-56"
                   />
                 ) : (
                   <TextFormField
-                    name="fast_default_model_name"
-                    subtext="The model to use for lighter flows like `LLM Chunk Filter` for this provider. If not set, will use the Default Model configured above."
-                    label="[Optional] Fast Model"
+                    name="default_model_name"
+                    subtext="The model to use by default for this provider unless otherwise specified."
+                    label="Default Model"
                     placeholder="E.g. gpt-4"
                   />
-                ))}
+                )}
 
-              <>
-                <Separator />
-                <AdvancedOptionsToggle
-                  showAdvancedOptions={showAdvancedOptions}
-                  setShowAdvancedOptions={setShowAdvancedOptions}
-                />
-                {showAdvancedOptions && (
-                  <>
-                    {llmProviderDescriptor.model_configurations.length > 0 && (
-                      <div className="w-full">
-                        <MultiSelectField
-                          selectedInitially={
-                            formikProps.values.selected_model_names ?? []
-                          }
-                          name="selected_model_names"
-                          label="Display Models"
-                          subtext="Select the models to make available to users. Unselected models will not be available."
-                          options={llmProviderDescriptor.model_configurations.map(
-                            (modelConfiguration) => ({
-                              value: modelConfiguration.name,
-                              // don't clean up names here to give admins descriptive names / handle duplicates
-                              // like us.anthropic.claude-3-7-sonnet-20250219-v1:0 and anthropic.claude-3-7-sonnet-20250219-v1:0
-                              label: modelConfiguration.name,
-                            })
-                          )}
-                          onChange={(selected) =>
-                            formikProps.setFieldValue(
-                              "selected_model_names",
-                              selected
-                            )
-                          }
-                        />
-                      </div>
-                    )}
-                    <IsPublicGroupSelector
-                      formikProps={formikProps}
-                      objectName="LLM Provider"
-                      publicToWhom="Users"
-                      enforceGroupSelection={true}
-                    />
-                  </>
+                {llmProviderDescriptor.deployment_name_required && (
+                  <TextFormField
+                    name="deployment_name"
+                    label="Deployment Name"
+                    placeholder="Deployment Name"
+                  />
                 )}
+
+                {!llmProviderDescriptor.single_model_supported &&
+                  (currentModelConfigurations.length > 0 ? (
+                    <SelectorFormField
+                      name="fast_default_model_name"
+                      subtext="The model to use for lighter flows like `LLM Chunk Filter` for this provider. If not set, will use the Default Model configured above."
+                      label="[Optional] Fast Model"
+                      options={currentModelConfigurations.map(
+                        (modelConfiguration) => ({
+                          // don't clean up names here to give admins descriptive names / handle duplicates
+                          // like us.anthropic.claude-3-7-sonnet-20250219-v1:0 and anthropic.claude-3-7-sonnet-20250219-v1:0
+                          name: modelConfiguration.name,
+                          value: modelConfiguration.name,
+                        })
+                      )}
+                      includeDefault
+                      maxHeight="max-h-56"
+                    />
+                  ) : (
+                    <TextFormField
+                      name="fast_default_model_name"
+                      subtext="The model to use for lighter flows like `LLM Chunk Filter` for this provider. If not set, will use the Default Model configured above."
+                      label="[Optional] Fast Model"
+                      placeholder="E.g. gpt-4"
+                    />
+                  ))}
+
+                <>
+                  <Separator />
+                  <AdvancedOptionsToggle
+                    showAdvancedOptions={showAdvancedOptions}
+                    setShowAdvancedOptions={setShowAdvancedOptions}
+                  />
+                  {showAdvancedOptions && (
+                    <>
+                      {currentModelConfigurations.length > 0 && (
+                        <div className="w-full">
+                          <MultiSelectField
+                            selectedInitially={
+                              formikProps.values.selected_model_names ?? []
+                            }
+                            name="selected_model_names"
+                            label="Display Models"
+                            subtext="Select the models to make available to users. Unselected models will not be available."
+                            options={currentModelConfigurations.map(
+                              (modelConfiguration) => ({
+                                value: modelConfiguration.name,
+                                // don't clean up names here to give admins descriptive names / handle duplicates
+                                // like us.anthropic.claude-3-7-sonnet-20250219-v1:0 and anthropic.claude-3-7-sonnet-20250219-v1:0
+                                label: modelConfiguration.name,
+                              })
+                            )}
+                            onChange={(selected) =>
+                              formikProps.setFieldValue(
+                                "selected_model_names",
+                                selected
+                              )
+                            }
+                          />
+                        </div>
+                      )}
+                      <IsPublicGroupSelector
+                        formikProps={formikProps}
+                        objectName="LLM Provider"
+                        publicToWhom="Users"
+                        enforceGroupSelection={true}
+                      />
+                    </>
+                  )}
+                </>
               </>
-            </>
-          )}
-
-          {/* NOTE: this is above the test button to make sure it's visible */}
-          {testError && <Text className="text-error mt-2">{testError}</Text>}
-
-          <div className="flex w-full mt-4">
-            <Button type="submit" variant="submit">
-              {isTesting ? (
-                <LoadingAnimation text="Testing" />
-              ) : existingLlmProvider ? (
-                "Update"
-              ) : (
-                "Enable"
-              )}
-            </Button>
-            {existingLlmProvider && (
-              <Button
-                type="button"
-                variant="destructive"
-                className="ml-3"
-                icon={FiTrash}
-                onClick={async () => {
-                  const response = await fetch(
-                    `${LLM_PROVIDERS_ADMIN_URL}/${existingLlmProvider.id}`,
-                    {
-                      method: "DELETE",
+            )}
+
+            {/* NOTE: this is above the test button to make sure it's visible */}
+            {testError && <Text className="text-error mt-2">{testError}</Text>}
+
+            <div className="flex w-full mt-4">
+              <Button type="submit" variant="submit">
+                {isTesting ? (
+                  <LoadingAnimation text="Testing" />
+                ) : existingLlmProvider ? (
+                  "Update"
+                ) : (
+                  "Enable"
+                )}
+              </Button>
+              {existingLlmProvider && (
+                <Button
+                  type="button"
+                  variant="destructive"
+                  className="ml-3"
+                  icon={FiTrash}
+                  onClick={async () => {
+                    const response = await fetch(
+                      `${LLM_PROVIDERS_ADMIN_URL}/${existingLlmProvider.id}`,
+                      {
+                        method: "DELETE",
+                      }
+                    );
+                    if (!response.ok) {
+                      const errorMsg = (await response.json()).detail;
+                      alert(`Failed to delete provider: ${errorMsg}`);
+                      return;
                     }
-                  );
-                  if (!response.ok) {
-                    const errorMsg = (await response.json()).detail;
-                    alert(`Failed to delete provider: ${errorMsg}`);
-                    return;
-                  }
-
-                  // If the deleted provider was the default, set the first remaining provider as default
-                  const remainingProvidersResponse = await fetch(
-                    LLM_PROVIDERS_ADMIN_URL
-                  );
-                  if (remainingProvidersResponse.ok) {
-                    const remainingProviders =
-                      await remainingProvidersResponse.json();
-
-                    if (remainingProviders.length > 0) {
-                      const setDefaultResponse = await fetch(
-                        `${LLM_PROVIDERS_ADMIN_URL}/${remainingProviders[0].id}/default`,
-                        {
-                          method: "POST",
+
+                    // If the deleted provider was the default, set the first remaining provider as default
+                    const remainingProvidersResponse = await fetch(
+                      LLM_PROVIDERS_ADMIN_URL
+                    );
+                    if (remainingProvidersResponse.ok) {
+                      const remainingProviders =
+                        await remainingProvidersResponse.json();
+
+                      if (remainingProviders.length > 0) {
+                        const setDefaultResponse = await fetch(
+                          `${LLM_PROVIDERS_ADMIN_URL}/${remainingProviders[0].id}/default`,
+                          {
+                            method: "POST",
+                          }
+                        );
+                        if (!setDefaultResponse.ok) {
+                          console.error("Failed to set new default provider");
                         }
-                      );
-                      if (!setDefaultResponse.ok) {
-                        console.error("Failed to set new default provider");
                       }
                     }
-                  }
 
-                  mutate(LLM_PROVIDERS_ADMIN_URL);
-                  onClose();
-                }}
-              >
-                Delete
-              </Button>
-            )}
-          </div>
-        </Form>
-      )}
+                    mutate(LLM_PROVIDERS_ADMIN_URL);
+                    onClose();
+                  }}
+                >
+                  Delete
+                </Button>
+              )}
+            </div>
+          </Form>
+        );
+      }}
     </Formik>
   );
 }
diff --git a/web/src/app/admin/configuration/llm/interfaces.ts b/web/src/app/admin/configuration/llm/interfaces.ts
index a19d0c8ce6c..613139ca626 100644
--- a/web/src/app/admin/configuration/llm/interfaces.ts
+++ b/web/src/app/admin/configuration/llm/interfaces.ts
@@ -10,13 +10,10 @@ export interface CustomConfigKey {
 
 export type CustomConfigKeyType = "text_input" | "file_input";
 
-export interface ModelConfigurationUpsertRequest {
+export interface ModelConfiguration {
   name: string;
   is_visible: boolean;
   max_input_tokens: number | null;
-}
-
-export interface ModelConfiguration extends ModelConfigurationUpsertRequest {
   supports_image_input: boolean;
 }
 
diff --git a/web/src/app/admin/configuration/llm/utils.ts b/web/src/app/admin/configuration/llm/utils.ts
index fc3e80b812d..377c93afa8a 100644
--- a/web/src/app/admin/configuration/llm/utils.ts
+++ b/web/src/app/admin/configuration/llm/utils.ts
@@ -11,6 +11,12 @@ import {
   OpenAISVG,
   QwenIcon,
 } from "@/components/icons/icons";
+import {
+  ModelConfiguration,
+  WellKnownLLMProviderDescriptor,
+  LLMProviderView,
+} from "./interfaces";
+import { PopupSpec } from "@/components/admin/connectors/Popup";
 
 export const getProviderIcon = (
   providerName: string,
@@ -62,3 +68,167 @@ export const getProviderIcon = (
 
 export const isAnthropic = (provider: string, modelName: string) =>
   provider === "anthropic" || modelName.toLowerCase().includes("claude");
+
+export const fetchModels = async (
+  llmProviderDescriptor: WellKnownLLMProviderDescriptor,
+  existingLlmProvider: LLMProviderView | undefined,
+  values: any,
+  setFieldValue: any,
+  setIsFetchingModels: (loading: boolean) => void,
+  setFetchModelsError: (error: string) => void,
+  setPopup?: (popup: PopupSpec) => void
+) => {
+  // Provider-specific configurations
+  const providerConfigs = {
+    bedrock: {
+      endpoint: "/api/admin/llm/bedrock/available-models",
+      validationCheck: () => !!values.custom_config?.AWS_REGION_NAME,
+      validationError: "AWS region is required to fetch Bedrock models",
+      requestBody: () => ({
+        aws_region_name: values.custom_config?.AWS_REGION_NAME,
+        aws_access_key_id: values.custom_config?.AWS_ACCESS_KEY_ID,
+        aws_secret_access_key: values.custom_config?.AWS_SECRET_ACCESS_KEY,
+        aws_bearer_token_bedrock:
+          values.custom_config?.AWS_BEARER_TOKEN_BEDROCK,
+        provider_name: existingLlmProvider?.name,
+      }),
+      processResponse: (data: string[]) =>
+        data.map((modelName) => {
+          const existingConfig =
+            llmProviderDescriptor.model_configurations.find(
+              (config) => config.name === modelName
+            );
+          return {
+            name: modelName,
+            is_visible: existingConfig?.is_visible ?? false,
+            max_input_tokens: null,
+            supports_image_input: existingConfig?.supports_image_input ?? null,
+          };
+        }),
+      getModelNames: (data: string[]) => data,
+      successMessage: (count: number) =>
+        `Successfully fetched ${count} models for the selected region (including cross-region inference models).`,
+    },
+    ollama: {
+      endpoint: "/api/admin/llm/ollama/available-models",
+      validationCheck: () => !!values.api_base,
+      validationError: "API Base is required to fetch Ollama models",
+      requestBody: () => ({
+        api_base: values.api_base,
+      }),
+      processResponse: (
+        data: {
+          name: string;
+          max_input_tokens: number;
+          supports_image_input: boolean;
+        }[]
+      ) =>
+        data.map((modelData) => {
+          const existingConfig =
+            llmProviderDescriptor.model_configurations.find(
+              (config) => config.name === modelData.name
+            );
+          return {
+            name: modelData.name,
+            is_visible: existingConfig?.is_visible ?? false,
+            max_input_tokens: modelData.max_input_tokens,
+            supports_image_input: modelData.supports_image_input,
+          };
+        }),
+      getModelNames: (
+        data: {
+          name: string;
+          max_input_tokens: number;
+          supports_image_input: boolean;
+        }[]
+      ) => data.map((model) => model.name),
+      successMessage: (count: number) =>
+        `Successfully fetched ${count} models from Ollama.`,
+    },
+  };
+
+  const config =
+    providerConfigs[llmProviderDescriptor.name as keyof typeof providerConfigs];
+  if (!config) {
+    return;
+  }
+
+  // Validation check
+  if (!config.validationCheck()) {
+    setFetchModelsError(config.validationError);
+    return;
+  }
+
+  setIsFetchingModels(true);
+  setFetchModelsError("");
+
+  try {
+    const response = await fetch(config.endpoint, {
+      method: "POST",
+      headers: {
+        "Content-Type": "application/json",
+      },
+      body: JSON.stringify(config.requestBody()),
+    });
+
+    if (!response.ok) {
+      let errorMessage = "Failed to fetch models";
+      try {
+        const errorData = await response.json();
+        errorMessage = errorData.detail || errorMessage;
+      } catch {
+        // ignore JSON parsing errors and use the fallback message
+      }
+      throw new Error(errorMessage);
+    }
+
+    const availableModels = await response.json();
+    const updatedModelConfigs = config.processResponse(availableModels);
+    const availableModelNames = config.getModelNames(availableModels);
+
+    // Store the updated model configurations in form state instead of mutating props
+    setFieldValue("fetched_model_configurations", updatedModelConfigs);
+
+    // Update selected model names to only include previously visible models that are available
+    const previouslySelectedModels = values.selected_model_names || [];
+    const stillAvailableSelectedModels = previouslySelectedModels.filter(
+      (modelName: string) => availableModelNames.includes(modelName)
+    );
+    setFieldValue("selected_model_names", stillAvailableSelectedModels);
+
+    // Set a default model if none is set
+    if (
+      (!values.default_model_name ||
+        !availableModelNames.includes(values.default_model_name)) &&
+      availableModelNames.length > 0
+    ) {
+      setFieldValue("default_model_name", availableModelNames[0]);
+    }
+
+    // Clear fast model if it's not in the new list
+    if (
+      values.fast_default_model_name &&
+      !availableModelNames.includes(values.fast_default_model_name)
+    ) {
+      setFieldValue("fast_default_model_name", null);
+    }
+
+    // Force a re-render by updating a timestamp or counter
+    setFieldValue("_modelListUpdated", Date.now());
+
+    setPopup?.({
+      message: config.successMessage(availableModelNames.length),
+      type: "success",
+    });
+  } catch (error) {
+    const errorMessage =
+      error instanceof Error ? error.message : "Unknown error";
+    setFetchModelsError(errorMessage);
+    setPopup?.({
+      message: `Failed to fetch models: ${errorMessage}`,
+      type: "error",
+    });
+  } finally {
+    setIsFetchingModels(false);
+  }
+};
diff --git a/web/src/lib/hooks.ts b/web/src/lib/hooks.ts
index 4ab0c8f9a91..46a2714bb76 100644
--- a/web/src/lib/hooks.ts
+++ b/web/src/lib/hooks.ts
@@ -960,6 +960,7 @@ const MODEL_DISPLAY_NAMES: { [key: string]: string } = {
   "gpt-oss:20b-cloud": "gpt-oss 20B Cloud",
   "gpt-oss:120b-cloud": "gpt-oss 120B Cloud",
   "deepseek-v3.1:671b-cloud": "DeepSeek-v3.1 671B Cloud",
+  "kimi-k2:1t": "Kimi K2 1T Cloud",
   "qwen3-coder:480b-cloud": "Qwen3-Coder 480B Cloud",
 
   // Ollama models in litellm map (disjoint from ollama's supported model list)

From 63fd86fc4945fa2299deb33abc553a68e7d05954 Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Sun, 28 Sep 2025 17:40:27 -0700
Subject: [PATCH 04/14] .

---
 .../64bd5677aeb6_add_image_input_support_to_model_config.py     | 2 --
 1 file changed, 2 deletions(-)

diff --git a/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py b/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py
index 02a8c9a2a43..7625268400d 100644
--- a/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py
+++ b/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py
@@ -25,6 +25,4 @@ def upgrade() -> None:
 
 
 def downgrade() -> None:
-    # ### commands auto generated by Alembic - please adjust! ###
     op.drop_column("model_configuration", "supports_image_input")
-    # ### end Alembic commands ###

From 8a053c9083a4daf0d8c08b953e4293eee028dda7 Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Sun, 28 Sep 2025 18:08:59 -0700
Subject: [PATCH 05/14] cleanup ollama support

---
 backend/onyx/db/models.py                     |  1 -
 backend/onyx/llm/factory.py                   |  2 +-
 backend/onyx/llm/llm_provider_options.py      |  2 +-
 backend/onyx/llm/utils.py                     |  2 --
 backend/onyx/server/manage/llm/api.py         | 12 ++++-----
 backend/onyx/server/manage/llm/models.py      |  2 +-
 .../configuration/llm/FetchModelsButton.tsx   | 27 ++++---------------
 .../app/admin/configuration/llm/interfaces.ts | 21 +++++++++++++++
 web/src/app/admin/configuration/llm/utils.ts  |  6 +----
 9 files changed, 36 insertions(+), 39 deletions(-)

diff --git a/backend/onyx/db/models.py b/backend/onyx/db/models.py
index 89d9b2c513e..2d39d486695 100644
--- a/backend/onyx/db/models.py
+++ b/backend/onyx/db/models.py
@@ -2353,7 +2353,6 @@ class ModelConfiguration(Base):
     # - The end-user is configuring a model and chooses not to set a max-input-tokens limit.
     max_input_tokens: Mapped[int | None] = mapped_column(Integer, nullable=True)
 
-    # Whether this model supports image input
     supports_image_input: Mapped[bool | None] = mapped_column(Boolean, nullable=True)
 
     llm_provider: Mapped["LLMProvider"] = relationship(
diff --git a/backend/onyx/llm/factory.py b/backend/onyx/llm/factory.py
index 5059f44c814..aeba171aa46 100644
--- a/backend/onyx/llm/factory.py
+++ b/backend/onyx/llm/factory.py
@@ -285,7 +285,7 @@ def get_llm(
 
     # NOTE: this is needed since Ollama API key is optional
     # User may access Ollama cloud via locally hosted instance (logged in)
-    # or just via the cloud API (not logged in)
+    # or just via the cloud API (not logged in, using API key)
     provider_extra_headers = _build_provider_extra_headers(provider, custom_config)
     if provider_extra_headers:
         extra_headers.update(provider_extra_headers)
diff --git a/backend/onyx/llm/llm_provider_options.py b/backend/onyx/llm/llm_provider_options.py
index 844e6740023..32e4899af39 100644
--- a/backend/onyx/llm/llm_provider_options.py
+++ b/backend/onyx/llm/llm_provider_options.py
@@ -200,7 +200,7 @@ def fetch_available_well_known_llms() -> list[WellKnownLLMProviderDescriptor]:
                 CustomConfigKey(
                     name=OLLAMA_API_KEY_CONFIG_KEY,
                     display_name="Ollama API Key",
-                    description="Optional API key used when connecting to Ollama Cloud (API base must be https://ollama.com).",
+                    description="Optional API key used when connecting to Ollama Cloud (i.e. API base is https://ollama.com).",
                     is_required=False,
                     is_secret=True,
                 )
diff --git a/backend/onyx/llm/utils.py b/backend/onyx/llm/utils.py
index b0d1d3419b2..cfdfb2f9d18 100644
--- a/backend/onyx/llm/utils.py
+++ b/backend/onyx/llm/utils.py
@@ -40,8 +40,6 @@
 from onyx.utils.logger import setup_logger
 from shared_configs.configs import LOG_LEVEL
 
-# Database imports for OLLAMA provider check
-
 
 if TYPE_CHECKING:
     from onyx.server.manage.llm.models import LLMProviderView
diff --git a/backend/onyx/server/manage/llm/api.py b/backend/onyx/server/manage/llm/api.py
index 9809a7aa659..cf401f49f14 100644
--- a/backend/onyx/server/manage/llm/api.py
+++ b/backend/onyx/server/manage/llm/api.py
@@ -43,8 +43,8 @@
 from onyx.server.manage.llm.models import LLMProviderUpsertRequest
 from onyx.server.manage.llm.models import LLMProviderView
 from onyx.server.manage.llm.models import ModelConfigurationUpsertRequest
+from onyx.server.manage.llm.models import OllamaFinalModelResponse
 from onyx.server.manage.llm.models import OllamaModelDetails
-from onyx.server.manage.llm.models import OllamaModelResponse
 from onyx.server.manage.llm.models import OllamaModelsRequest
 from onyx.server.manage.llm.models import TestLLMRequest
 from onyx.server.manage.llm.models import VisionProviderResponse
@@ -503,7 +503,7 @@ def _get_ollama_available_model_names(api_base: str) -> set[str]:
 def get_ollama_available_models(
     request: OllamaModelsRequest,
     _: User | None = Depends(current_admin_user),
-) -> list[OllamaModelResponse]:
+) -> list[OllamaFinalModelResponse]:
     """Fetch the list of available models from an Ollama server."""
 
     cleaned_api_base = request.api_base.strip().rstrip("/")
@@ -519,7 +519,7 @@ def get_ollama_available_models(
             detail="No models found from your Ollama server",
         )
 
-    models_with_context_size: list[OllamaModelResponse] = []
+    all_models_with_context_size_and_vision: list[OllamaFinalModelResponse] = []
     show_url = f"{cleaned_api_base}/api/show"
 
     for model_name in model_names:
@@ -567,12 +567,12 @@ def get_ollama_available_models(
         if not supports_image_input:
             supports_image_input = False
 
-        models_with_context_size.append(
-            OllamaModelResponse(
+        all_models_with_context_size_and_vision.append(
+            OllamaFinalModelResponse(
                 name=model_name,
                 max_input_tokens=context_limit,
                 supports_image_input=supports_image_input,
             )
         )
 
-    return models_with_context_size
+    return all_models_with_context_size_and_vision
diff --git a/backend/onyx/server/manage/llm/models.py b/backend/onyx/server/manage/llm/models.py
index c63141c6107..5449eab1c83 100644
--- a/backend/onyx/server/manage/llm/models.py
+++ b/backend/onyx/server/manage/llm/models.py
@@ -205,7 +205,7 @@ class OllamaModelsRequest(BaseModel):
     api_base: str
 
 
-class OllamaModelResponse(BaseModel):
+class OllamaFinalModelResponse(BaseModel):
     name: str
     max_input_tokens: int
     supports_image_input: bool
diff --git a/web/src/app/admin/configuration/llm/FetchModelsButton.tsx b/web/src/app/admin/configuration/llm/FetchModelsButton.tsx
index 1c1ea6a3b85..a77421202e3 100644
--- a/web/src/app/admin/configuration/llm/FetchModelsButton.tsx
+++ b/web/src/app/admin/configuration/llm/FetchModelsButton.tsx
@@ -2,29 +2,12 @@ import { Button } from "@/components/ui/button";
 import { LoadingAnimation } from "@/components/Loading";
 import Text from "@/components/ui/text";
 import { fetchModels } from "./utils";
-import { WellKnownLLMProviderDescriptor, LLMProviderView } from "./interfaces";
-import { PopupSpec } from "@/components/admin/connectors/Popup";
+import {
+  ProviderFetchModelsConfig,
+  FetchModelsButtonProps,
+} from "./interfaces";
 
-interface FetchModelsButtonProps {
-  llmProviderDescriptor: WellKnownLLMProviderDescriptor;
-  existingLlmProvider?: LLMProviderView;
-  values: any;
-  setFieldValue: any;
-  isFetchingModels: boolean;
-  setIsFetchingModels: (loading: boolean) => void;
-  fetchModelsError: string;
-  setFetchModelsError: (error: string) => void;
-  setPopup?: (popup: PopupSpec) => void;
-}
-
-interface ProviderConfig {
-  buttonText: string;
-  loadingText: string;
-  helperText: string | React.ReactNode;
-  isDisabled: (values: any) => boolean;
-}
-
-const providerConfigs: Record<string, ProviderConfig> = {
+const providerConfigs: Record<string, ProviderFetchModelsConfig> = {
   bedrock: {
     buttonText: "Fetch Available Models for Region",
     loadingText: "Fetching Models...",
diff --git a/web/src/app/admin/configuration/llm/interfaces.ts b/web/src/app/admin/configuration/llm/interfaces.ts
index 613139ca626..b20fc2c4918 100644
--- a/web/src/app/admin/configuration/llm/interfaces.ts
+++ b/web/src/app/admin/configuration/llm/interfaces.ts
@@ -1,3 +1,5 @@
+import { PopupSpec } from "@/components/admin/connectors/Popup";
+
 export interface CustomConfigKey {
   name: string;
   display_name: string;
@@ -78,3 +80,22 @@ export interface LLMProviderDescriptor {
   groups: number[];
   model_configurations: ModelConfiguration[];
 }
+
+export interface ProviderFetchModelsConfig {
+  buttonText: string;
+  loadingText: string;
+  helperText: string | React.ReactNode;
+  isDisabled: (values: any) => boolean;
+}
+
+export interface FetchModelsButtonProps {
+  llmProviderDescriptor: WellKnownLLMProviderDescriptor;
+  existingLlmProvider?: LLMProviderView;
+  values: any;
+  setFieldValue: any;
+  isFetchingModels: boolean;
+  setIsFetchingModels: (loading: boolean) => void;
+  fetchModelsError: string;
+  setFetchModelsError: (error: string) => void;
+  setPopup?: (popup: PopupSpec) => void;
+}
diff --git a/web/src/app/admin/configuration/llm/utils.ts b/web/src/app/admin/configuration/llm/utils.ts
index 377c93afa8a..7225dac651b 100644
--- a/web/src/app/admin/configuration/llm/utils.ts
+++ b/web/src/app/admin/configuration/llm/utils.ts
@@ -11,11 +11,7 @@ import {
   OpenAISVG,
   QwenIcon,
 } from "@/components/icons/icons";
-import {
-  ModelConfiguration,
-  WellKnownLLMProviderDescriptor,
-  LLMProviderView,
-} from "./interfaces";
+import { WellKnownLLMProviderDescriptor, LLMProviderView } from "./interfaces";
 import { PopupSpec } from "@/components/admin/connectors/Popup";
 
 export const getProviderIcon = (

From 6a8d57efa0fb6ee22271b2b53efb75fac3a8962a Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Sun, 28 Sep 2025 18:28:34 -0700
Subject: [PATCH 06/14] cleanup ollama support

---
 backend/onyx/llm/llm_provider_options.py      |  2 ++
 .../configuration/llm/FetchModelsButton.tsx   |  4 +--
 .../llm/LLMProviderUpdateForm.tsx             |  3 +-
 .../app/admin/configuration/llm/interfaces.ts | 24 +++++++++++++--
 web/src/app/admin/configuration/llm/utils.ts  | 29 ++++++++-----------
 web/src/lib/hooks.ts                          | 24 ---------------
 6 files changed, 40 insertions(+), 46 deletions(-)

diff --git a/backend/onyx/llm/llm_provider_options.py b/backend/onyx/llm/llm_provider_options.py
index 32e4899af39..2e896ac3eee 100644
--- a/backend/onyx/llm/llm_provider_options.py
+++ b/backend/onyx/llm/llm_provider_options.py
@@ -39,6 +39,7 @@ class WellKnownLLMProviderDescriptor(BaseModel):
     model_configurations: list[ModelConfigurationView]
     default_model: str | None = None
     default_fast_model: str | None = None
+    default_api_base: str | None = None
     # set for providers like Azure, which require a deployment name.
     deployment_name_required: bool = False
     # set for providers like Azure, which support a single model per deployment.
@@ -210,6 +211,7 @@ def fetch_available_well_known_llms() -> list[WellKnownLLMProviderDescriptor]:
             ),
             default_model=None,
             default_fast_model=None,
+            default_api_base="http://127.0.0.1:11434",
         ),
         WellKnownLLMProviderDescriptor(
             name=ANTHROPIC_PROVIDER_NAME,
diff --git a/web/src/app/admin/configuration/llm/FetchModelsButton.tsx b/web/src/app/admin/configuration/llm/FetchModelsButton.tsx
index a77421202e3..93aaa62e0ec 100644
--- a/web/src/app/admin/configuration/llm/FetchModelsButton.tsx
+++ b/web/src/app/admin/configuration/llm/FetchModelsButton.tsx
@@ -3,11 +3,11 @@ import { LoadingAnimation } from "@/components/Loading";
 import Text from "@/components/ui/text";
 import { fetchModels } from "./utils";
 import {
-  ProviderFetchModelsConfig,
+  ProviderFetchModelsButtonConfig,
   FetchModelsButtonProps,
 } from "./interfaces";
 
-const providerConfigs: Record<string, ProviderFetchModelsConfig> = {
+const providerConfigs: Record<string, ProviderFetchModelsButtonConfig> = {
   bedrock: {
     buttonText: "Fetch Available Models for Region",
     loadingText: "Fetching Models...",
diff --git a/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx b/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx
index f9c107f7db0..aa1fbc5d21b 100644
--- a/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx
+++ b/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx
@@ -68,7 +68,8 @@ export function LLMProviderUpdateForm({
     api_key: existingLlmProvider?.api_key ?? "",
     api_base:
       existingLlmProvider?.api_base ??
-      (llmProviderDescriptor.name === "ollama" ? "http://127.0.0.1:11434" : ""),
+      llmProviderDescriptor.default_api_base ??
+      "",
     api_version: existingLlmProvider?.api_version ?? "",
     // For Azure OpenAI, combine api_base and api_version into target_uri
     target_uri:
diff --git a/web/src/app/admin/configuration/llm/interfaces.ts b/web/src/app/admin/configuration/llm/interfaces.ts
index b20fc2c4918..951ddba8357 100644
--- a/web/src/app/admin/configuration/llm/interfaces.ts
+++ b/web/src/app/admin/configuration/llm/interfaces.ts
@@ -16,7 +16,7 @@ export interface ModelConfiguration {
   name: string;
   is_visible: boolean;
   max_input_tokens: number | null;
-  supports_image_input: boolean;
+  supports_image_input: boolean | null;
 }
 
 export interface WellKnownLLMProviderDescriptor {
@@ -33,6 +33,7 @@ export interface WellKnownLLMProviderDescriptor {
   model_configurations: ModelConfiguration[];
   default_model: string | null;
   default_fast_model: string | null;
+  default_api_base: string | null;
   is_public: boolean;
   groups: number[];
 }
@@ -81,7 +82,7 @@ export interface LLMProviderDescriptor {
   model_configurations: ModelConfiguration[];
 }
 
-export interface ProviderFetchModelsConfig {
+export interface ProviderFetchModelsButtonConfig {
   buttonText: string;
   loadingText: string;
   helperText: string | React.ReactNode;
@@ -99,3 +100,22 @@ export interface FetchModelsButtonProps {
   setFetchModelsError: (error: string) => void;
   setPopup?: (popup: PopupSpec) => void;
 }
+
+export interface OllamaModelResponse {
+  name: string;
+  max_input_tokens: number;
+  supports_image_input: boolean;
+}
+
+export interface FetchModelsConfig<
+  TApiResponse = any,
+  TProcessedResponse = ModelConfiguration,
+> {
+  endpoint: string;
+  validationCheck: () => boolean;
+  validationError: string;
+  requestBody: () => Record<string, any>;
+  processResponse: (data: TApiResponse) => TProcessedResponse[];
+  getModelNames: (data: TApiResponse) => string[];
+  successMessage: (count: number) => string;
+}
diff --git a/web/src/app/admin/configuration/llm/utils.ts b/web/src/app/admin/configuration/llm/utils.ts
index 7225dac651b..13bf78385d6 100644
--- a/web/src/app/admin/configuration/llm/utils.ts
+++ b/web/src/app/admin/configuration/llm/utils.ts
@@ -11,7 +11,13 @@ import {
   OpenAISVG,
   QwenIcon,
 } from "@/components/icons/icons";
-import { WellKnownLLMProviderDescriptor, LLMProviderView } from "./interfaces";
+import {
+  WellKnownLLMProviderDescriptor,
+  LLMProviderView,
+  FetchModelsConfig,
+  OllamaModelResponse,
+  ModelConfiguration,
+} from "./interfaces";
 import { PopupSpec } from "@/components/admin/connectors/Popup";
 
 export const getProviderIcon = (
@@ -75,7 +81,7 @@ export const fetchModels = async (
   setPopup?: (popup: PopupSpec) => void
 ) => {
   // Provider-specific configurations
-  const providerConfigs = {
+  const providerConfigs: Record<string, FetchModelsConfig> = {
     bedrock: {
       endpoint: "/api/admin/llm/bedrock/available-models",
       validationCheck: () => !!values.custom_config?.AWS_REGION_NAME,
@@ -88,7 +94,7 @@ export const fetchModels = async (
           values.custom_config?.AWS_BEARER_TOKEN_BEDROCK,
         provider_name: existingLlmProvider?.name,
       }),
-      processResponse: (data: string[]) =>
+      processResponse: (data: string[]): ModelConfiguration[] =>
         data.map((modelName) => {
           const existingConfig =
             llmProviderDescriptor.model_configurations.find(
@@ -112,13 +118,7 @@ export const fetchModels = async (
       requestBody: () => ({
         api_base: values.api_base,
       }),
-      processResponse: (
-        data: {
-          name: string;
-          max_input_tokens: number;
-          supports_image_input: boolean;
-        }[]
-      ) =>
+      processResponse: (data: OllamaModelResponse[]): ModelConfiguration[] =>
         data.map((modelData) => {
           const existingConfig =
             llmProviderDescriptor.model_configurations.find(
@@ -131,13 +131,8 @@ export const fetchModels = async (
             supports_image_input: modelData.supports_image_input,
           };
         }),
-      getModelNames: (
-        data: {
-          name: string;
-          max_input_tokens: number;
-          supports_image_input: boolean;
-        }[]
-      ) => data.map((model) => model.name),
+      getModelNames: (data: OllamaModelResponse[]) =>
+        data.map((model) => model.name),
       successMessage: (count: number) =>
         `Successfully fetched ${count} models from Ollama.`,
     },
diff --git a/web/src/lib/hooks.ts b/web/src/lib/hooks.ts
index 46a2714bb76..1514ac1b94e 100644
--- a/web/src/lib/hooks.ts
+++ b/web/src/lib/hooks.ts
@@ -1006,30 +1006,6 @@ export function getDisplayNameForModel(modelName: string): string {
   return MODEL_DISPLAY_NAMES[modelName] || modelName;
 }
 
-export const defaultModelsByProvider: { [name: string]: string[] } = {
-  openai: [
-    "gpt-4",
-    "gpt-4o",
-    "gpt-4o-mini",
-    "gpt-4.1",
-    "o3-mini",
-    "o1-mini",
-    "o1",
-    "o4-mini",
-    "o3",
-  ],
-  bedrock: [
-    "meta.llama3-1-70b-instruct-v1:0",
-    "meta.llama3-1-8b-instruct-v1:0",
-    "anthropic.claude-3-opus-20240229-v1:0",
-    "mistral.mistral-large-2402-v1:0",
-    "anthropic.claude-3-5-sonnet-20241022-v2:0",
-    "anthropic.claude-3-7-sonnet-20250219-v1:0",
-  ],
-  anthropic: ["claude-3-opus-20240229", "claude-3-5-sonnet-20241022"],
-  ollama: ["llama3.2", "phi3:mini", "mistral-small"],
-};
-
 // Get source metadata for configured sources - deduplicated by source type
 function getConfiguredSources(
   availableSources: ValidSources[]

From 6990821aaf0e26cc354d7386e67a5616294f706f Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Sun, 28 Sep 2025 18:45:04 -0700
Subject: [PATCH 07/14] cleanup ollama support

---
 backend/onyx/llm/utils.py | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)

diff --git a/backend/onyx/llm/utils.py b/backend/onyx/llm/utils.py
index cfdfb2f9d18..f8396a5c73d 100644
--- a/backend/onyx/llm/utils.py
+++ b/backend/onyx/llm/utils.py
@@ -28,6 +28,7 @@
 from onyx.configs.model_configs import GEN_AI_MODEL_FALLBACK_MAX_TOKENS
 from onyx.configs.model_configs import GEN_AI_NUM_RESERVED_OUTPUT_TOKENS
 from onyx.db.engine.sql_engine import get_session_with_current_tenant
+from onyx.db.models import LLMProvider
 from onyx.db.models import ModelConfiguration
 from onyx.file_store.models import ChatFileType
 from onyx.file_store.models import InMemoryChatFile
@@ -645,22 +646,26 @@ def get_max_input_tokens_from_llm_provider(
 def model_supports_image_input(model_name: str, model_provider: str) -> bool:
     # TODO: Add support to check model config for any provider
     # TODO: Circular import means OLLAMA_PROVIDER_NAME is not available here
-    # NOTE: Theoretically, if user has model configured through both Ollama and another provider,
-    # this may erroenously return False because this checks just one database entry
-    # and it may be the other provider's entry, which would be None
+
     if model_provider == "ollama":
         try:
             with get_session_with_current_tenant() as db_session:
                 model_config = db_session.scalar(
-                    select(ModelConfiguration).where(
-                        ModelConfiguration.name == model_name
+                    select(ModelConfiguration)
+                    .join(
+                        LLMProvider,
+                        ModelConfiguration.llm_provider_id == LLMProvider.id,
+                    )
+                    .where(
+                        ModelConfiguration.name == model_name,
+                        LLMProvider.provider == model_provider,
                     )
                 )
                 if model_config and model_config.supports_image_input is not None:
                     return model_config.supports_image_input
         except Exception as e:
             logger.warning(
-                f"Failed to query database for Ollama model {model_name} image support: {e}"
+                f"Failed to query database for {model_provider} model {model_name} image support: {e}"
             )
 
     model_map = get_model_map()

From 12eb3b4c1b50566e47ccdce120ae398b8b179030 Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Sun, 28 Sep 2025 19:14:14 -0700
Subject: [PATCH 08/14] show ollama card on provider update

---
 ...add_image_input_support_to_model_config.py |  9 +++++++++
 backend/onyx/llm/llm_provider_options.py      |  5 ++---
 backend/onyx/server/manage/llm/models.py      |  4 ++--
 .../llm/ConfiguredLLMProviderDisplay.tsx      | 20 +++++++++----------
 web/src/app/admin/configuration/llm/utils.ts  |  2 +-
 5 files changed, 24 insertions(+), 16 deletions(-)

diff --git a/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py b/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py
index 7625268400d..8b56c8e692b 100644
--- a/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py
+++ b/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py
@@ -23,6 +23,15 @@ def upgrade() -> None:
         sa.Column("supports_image_input", sa.Boolean(), nullable=True),
     )
 
+    # Seems to be left over from when model visibility was introduced and a nullable field.
+    # Set any null is_visible values to False
+    connection = op.get_bind()
+    connection.execute(
+        sa.text(
+            "UPDATE model_configuration SET is_visible = false WHERE is_visible IS NULL"
+        )
+    )
+
 
 def downgrade() -> None:
     op.drop_column("model_configuration", "supports_image_input")
diff --git a/backend/onyx/llm/llm_provider_options.py b/backend/onyx/llm/llm_provider_options.py
index 2e896ac3eee..e2e56b7fcfd 100644
--- a/backend/onyx/llm/llm_provider_options.py
+++ b/backend/onyx/llm/llm_provider_options.py
@@ -96,7 +96,6 @@ class WellKnownLLMProviderDescriptor(BaseModel):
     for model in list(litellm.bedrock_models.union(litellm.bedrock_converse_models))
     if "/" not in model and "embed" not in model
 ][::-1]
-BEDROCK_DEFAULT_MODEL = "anthropic.claude-3-5-sonnet-20241022-v2:0"
 
 OLLAMA_PROVIDER_NAME = "ollama"
 OLLAMA_API_KEY_CONFIG_KEY = "OLLAMA_API_KEY"
@@ -169,7 +168,7 @@ class WellKnownLLMProviderDescriptor(BaseModel):
 
 _PROVIDER_TO_VISIBLE_MODELS_MAP = {
     OPENAI_PROVIDER_NAME: OPEN_AI_VISIBLE_MODEL_NAMES,
-    BEDROCK_PROVIDER_NAME: [BEDROCK_DEFAULT_MODEL],
+    BEDROCK_PROVIDER_NAME: [],
     ANTHROPIC_PROVIDER_NAME: ANTHROPIC_VISIBLE_MODEL_NAMES,
     VERTEXAI_PROVIDER_NAME: VERTEXAI_VISIBLE_MODEL_NAMES,
     OLLAMA_PROVIDER_NAME: [],
@@ -276,7 +275,7 @@ def fetch_available_well_known_llms() -> list[WellKnownLLMProviderDescriptor]:
             model_configurations=fetch_model_configurations_for_provider(
                 BEDROCK_PROVIDER_NAME
             ),
-            default_model=BEDROCK_DEFAULT_MODEL,
+            default_model=None,
             default_fast_model=None,
         ),
         WellKnownLLMProviderDescriptor(
diff --git a/backend/onyx/server/manage/llm/models.py b/backend/onyx/server/manage/llm/models.py
index 5449eab1c83..4074cb06db1 100644
--- a/backend/onyx/server/manage/llm/models.py
+++ b/backend/onyx/server/manage/llm/models.py
@@ -139,7 +139,7 @@ def from_model(
 
 class ModelConfigurationUpsertRequest(BaseModel):
     name: str
-    is_visible: bool | None = False
+    is_visible: bool
     max_input_tokens: int | None = None
     supports_image_input: bool | None = None
 
@@ -157,7 +157,7 @@ def from_model(
 
 class ModelConfigurationView(BaseModel):
     name: str
-    is_visible: bool | None = False
+    is_visible: bool
     max_input_tokens: int | None = None
     supports_image_input: bool
 
diff --git a/web/src/app/admin/configuration/llm/ConfiguredLLMProviderDisplay.tsx b/web/src/app/admin/configuration/llm/ConfiguredLLMProviderDisplay.tsx
index 92e2f8ea858..cc3f52fa682 100644
--- a/web/src/app/admin/configuration/llm/ConfiguredLLMProviderDisplay.tsx
+++ b/web/src/app/admin/configuration/llm/ConfiguredLLMProviderDisplay.tsx
@@ -176,16 +176,16 @@ export function ConfiguredLLMProviderDisplay({
             // then the provider is custom - don't use the default
             // provider descriptor
             llmProviderDescriptor={
-              isSubset(
-                defaultProviderDesciptor
-                  ? defaultProviderDesciptor.model_configurations.map(
-                      (model_configuration) => model_configuration.name
-                    )
-                  : [],
-                provider.model_configurations.map(
-                  (model_configuration) => model_configuration.name
-                )
-              )
+              defaultProviderDesciptor &&
+              (defaultProviderDesciptor.model_configurations.length === 0 ||
+                isSubset(
+                  defaultProviderDesciptor.model_configurations.map(
+                    (model_configuration) => model_configuration.name
+                  ),
+                  provider.model_configurations.map(
+                    (model_configuration) => model_configuration.name
+                  )
+                ))
                 ? defaultProviderDesciptor
                 : null
             }
diff --git a/web/src/app/admin/configuration/llm/utils.ts b/web/src/app/admin/configuration/llm/utils.ts
index 13bf78385d6..e062da6dc5f 100644
--- a/web/src/app/admin/configuration/llm/utils.ts
+++ b/web/src/app/admin/configuration/llm/utils.ts
@@ -126,7 +126,7 @@ export const fetchModels = async (
             );
           return {
             name: modelData.name,
-            is_visible: existingConfig?.is_visible ?? false,
+            is_visible: existingConfig?.is_visible ?? true,
             max_input_tokens: modelData.max_input_tokens,
             supports_image_input: modelData.supports_image_input,
           };

From 3b2eafd011ce7d2b2498a7d3e0e6008761b4c9d7 Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Sun, 28 Sep 2025 19:29:49 -0700
Subject: [PATCH 09/14] fix: non-image models shouldn't be hidden

---
 web/src/app/chat/components/input/ChatInputBar.tsx | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/web/src/app/chat/components/input/ChatInputBar.tsx b/web/src/app/chat/components/input/ChatInputBar.tsx
index dda968de675..70429bad932 100644
--- a/web/src/app/chat/components/input/ChatInputBar.tsx
+++ b/web/src/app/chat/components/input/ChatInputBar.tsx
@@ -660,7 +660,7 @@ export const ChatInputBar = React.memo(function ChatInputBar({
                   <LLMPopover
                     llmProviders={llmProviders}
                     llmManager={llmManager}
-                    requiresImageGeneration={true}
+                    requiresImageGeneration={llmManager.imageFilesPresent}
                     currentAssistant={selectedAssistant}
                   />
 

From 7b2ed309ef0b5e6b7a32590627eaa504bd403775 Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Mon, 29 Sep 2025 10:24:53 -0700
Subject: [PATCH 10/14] mypy

---
 backend/onyx/server/manage/llm/api.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/backend/onyx/server/manage/llm/api.py b/backend/onyx/server/manage/llm/api.py
index cf401f49f14..ab444dda4ef 100644
--- a/backend/onyx/server/manage/llm/api.py
+++ b/backend/onyx/server/manage/llm/api.py
@@ -524,6 +524,7 @@ def get_ollama_available_models(
 
     for model_name in model_names:
         context_limit: int | None = None
+        supports_image_input: bool | None = None
         try:
             show_response = httpx.post(
                 show_url,

From 35385ce22715190910197c2fde41c4487e292fb8 Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Mon, 29 Sep 2025 10:46:31 -0700
Subject: [PATCH 11/14] update integration tests

---
 .../tests/llm_provider/test_llm_provider.py   | 27 +++++++++----------
 1 file changed, 12 insertions(+), 15 deletions(-)

diff --git a/backend/tests/integration/tests/llm_provider/test_llm_provider.py b/backend/tests/integration/tests/llm_provider/test_llm_provider.py
index 85d46ef2fe3..c6736aa319c 100644
--- a/backend/tests/integration/tests/llm_provider/test_llm_provider.py
+++ b/backend/tests/integration/tests/llm_provider/test_llm_provider.py
@@ -92,31 +92,22 @@ def fill_max_input_tokens_and_supports_image_input(
                 )
             ],
         ),
-        # Test the case in which the basic model-configuration is passed, but its visibility is not
-        # specified (and thus defaulted to False).
-        # In this case, since the one model-configuration is also the default-model-name, its
-        # visibility should be overriden to True.
-        (
-            "gpt-4",
-            [ModelConfigurationUpsertRequest(name="gpt-4")],
-            [ModelConfigurationUpsertRequest(name="gpt-4", is_visible=True)],
-        ),
         # Test the case in which multiple model-configuration are passed.
         (
             "gpt-4",
             [
-                ModelConfigurationUpsertRequest(name="gpt-4"),
-                ModelConfigurationUpsertRequest(name="gpt-4o"),
+                ModelConfigurationUpsertRequest(name="gpt-4", is_visible=True),
+                ModelConfigurationUpsertRequest(name="gpt-4o", is_visible=True),
             ],
             [
                 ModelConfigurationUpsertRequest(name="gpt-4", is_visible=True),
-                ModelConfigurationUpsertRequest(name="gpt-4o"),
+                ModelConfigurationUpsertRequest(name="gpt-4o", is_visible=True),
             ],
         ),
         # Test the case in which duplicate model-configuration are passed.
         (
             "gpt-4",
-            [ModelConfigurationUpsertRequest(name="gpt-4")] * 4,
+            [ModelConfigurationUpsertRequest(name="gpt-4", is_visible=True)] * 4,
             [ModelConfigurationUpsertRequest(name="gpt-4", is_visible=True)],
         ),
         # Test the case in which no model-configurations are passed.
@@ -132,10 +123,16 @@ def fill_max_input_tokens_and_supports_image_input(
         # (`ModelConfiguration(name="gpt-4", is_visible=True, max_input_tokens=None)`).
         (
             "gpt-4",
-            [ModelConfigurationUpsertRequest(name="gpt-4o", max_input_tokens=4096)],
+            [
+                ModelConfigurationUpsertRequest(
+                    name="gpt-4o", is_visible=True, max_input_tokens=4096
+                )
+            ],
             [
                 ModelConfigurationUpsertRequest(name="gpt-4", is_visible=True),
-                ModelConfigurationUpsertRequest(name="gpt-4o", max_input_tokens=4096),
+                ModelConfigurationUpsertRequest(
+                    name="gpt-4o", is_visible=True, max_input_tokens=4096
+                ),
             ],
         ),
     ],

From ffcc0dceefa91390f4eb1de3ae4e2922eecb79fb Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Mon, 29 Sep 2025 13:31:51 -0700
Subject: [PATCH 12/14] fix more integration tests

---
 .../tests/llm_provider/test_llm_provider.py   | 26 +++++++++++++------
 1 file changed, 18 insertions(+), 8 deletions(-)

diff --git a/backend/tests/integration/tests/llm_provider/test_llm_provider.py b/backend/tests/integration/tests/llm_provider/test_llm_provider.py
index c6736aa319c..7680e571870 100644
--- a/backend/tests/integration/tests/llm_provider/test_llm_provider.py
+++ b/backend/tests/integration/tests/llm_provider/test_llm_provider.py
@@ -179,7 +179,11 @@ def test_create_llm_provider(
         (
             (
                 "gpt-4",
-                [ModelConfigurationUpsertRequest(name="gpt-4", max_input_tokens=4096)],
+                [
+                    ModelConfigurationUpsertRequest(
+                        name="gpt-4", is_visible=True, max_input_tokens=4096
+                    )
+                ],
             ),
             [
                 ModelConfigurationUpsertRequest(
@@ -188,7 +192,7 @@ def test_create_llm_provider(
             ],
             (
                 "gpt-4",
-                [ModelConfigurationUpsertRequest(name="gpt-4")],
+                [ModelConfigurationUpsertRequest(name="gpt-4", is_visible=True)],
             ),
             [ModelConfigurationUpsertRequest(name="gpt-4", is_visible=True)],
         ),
@@ -198,19 +202,25 @@ def test_create_llm_provider(
             (
                 "gpt-4",
                 [
-                    ModelConfigurationUpsertRequest(name="gpt-4"),
+                    ModelConfigurationUpsertRequest(name="gpt-4", is_visible=True),
                     ModelConfigurationUpsertRequest(
-                        name="gpt-4o", max_input_tokens=4096
+                        name="gpt-4o", is_visible=True, max_input_tokens=4096
                     ),
                 ],
             ),
             [
                 ModelConfigurationUpsertRequest(name="gpt-4", is_visible=True),
-                ModelConfigurationUpsertRequest(name="gpt-4o", max_input_tokens=4096),
+                ModelConfigurationUpsertRequest(
+                    name="gpt-4o", is_visible=True, max_input_tokens=4096
+                ),
             ],
             (
                 "gpt-4",
-                [ModelConfigurationUpsertRequest(name="gpt-4", max_input_tokens=4096)],
+                [
+                    ModelConfigurationUpsertRequest(
+                        name="gpt-4", is_visible=True, max_input_tokens=4096
+                    )
+                ],
             ),
             [
                 ModelConfigurationUpsertRequest(
@@ -325,8 +335,8 @@ def test_update_model_configurations(
         (
             "gpt-4",
             [
-                ModelConfigurationUpsertRequest(name="gpt-4o"),
-                ModelConfigurationUpsertRequest(name="gpt-4"),
+                ModelConfigurationUpsertRequest(name="gpt-4o", is_visible=True),
+                ModelConfigurationUpsertRequest(name="gpt-4", is_visible=True),
             ],
         ),
     ],

From e6b71ba86f445e1933d133d6354c2be95fb85c2c Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Wed, 1 Oct 2025 14:23:18 -0700
Subject: [PATCH 13/14] update migration

---
 .../64bd5677aeb6_add_image_input_support_to_model_config.py   | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py b/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py
index 8b56c8e692b..b373114302a 100644
--- a/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py
+++ b/backend/alembic/versions/64bd5677aeb6_add_image_input_support_to_model_config.py
@@ -1,7 +1,7 @@
 """Add image input support to model config
 
 Revision ID: 64bd5677aeb6
-Revises: 2b75d0a8ffcb
+Revises: b30353be4eec
 Create Date: 2025-09-28 15:48:12.003612
 
 """
@@ -12,7 +12,7 @@
 
 # revision identifiers, used by Alembic.
 revision = "64bd5677aeb6"
-down_revision = "2b75d0a8ffcb"
+down_revision = "b30353be4eec"
 branch_labels = None
 depends_on = None
 

From a9184d81026acc856cecde5ec1b159dec0868145 Mon Sep 17 00:00:00 2001
From: Wenxi Onyx <wenxi@onyx.app>
Date: Wed, 1 Oct 2025 16:07:47 -0700
Subject: [PATCH 14/14] cleanup ollama support

---
 .../configuration/llm/FetchModelsButton.tsx   |  95 -------------
 .../llm/LLMProviderUpdateForm.tsx             | 130 +++++++++++++++--
 .../app/admin/configuration/llm/interfaces.ts |  35 ++---
 web/src/app/admin/configuration/llm/utils.ts  | 134 +++++++++---------
 4 files changed, 195 insertions(+), 199 deletions(-)
 delete mode 100644 web/src/app/admin/configuration/llm/FetchModelsButton.tsx

diff --git a/web/src/app/admin/configuration/llm/FetchModelsButton.tsx b/web/src/app/admin/configuration/llm/FetchModelsButton.tsx
deleted file mode 100644
index 93aaa62e0ec..00000000000
--- a/web/src/app/admin/configuration/llm/FetchModelsButton.tsx
+++ /dev/null
@@ -1,95 +0,0 @@
-import { Button } from "@/components/ui/button";
-import { LoadingAnimation } from "@/components/Loading";
-import Text from "@/components/ui/text";
-import { fetchModels } from "./utils";
-import {
-  ProviderFetchModelsButtonConfig,
-  FetchModelsButtonProps,
-} from "./interfaces";
-
-const providerConfigs: Record<string, ProviderFetchModelsButtonConfig> = {
-  bedrock: {
-    buttonText: "Fetch Available Models for Region",
-    loadingText: "Fetching Models...",
-    helperText: (
-      <>
-        Enter your AWS region, then click this button to fetch available Bedrock
-        models.
-        <br />
-        If you&apos;re updating your existing provider, you&apos;ll need to
-        click this button to fetch the latest models.
-      </>
-    ),
-    isDisabled: (values) => !values.custom_config?.AWS_REGION_NAME,
-  },
-  ollama: {
-    buttonText: "Fetch Available Ollama Models",
-    loadingText: "Fetching Models...",
-    helperText: (
-      <>
-        Ensure your Ollama server is accessible from Onyx and that the requested
-        models are pulled (e.g. via <code>ollama pull</code>). Provide the
-        server&apos;s base URL and optional API key (when using Ollama Cloud)
-        before fetching the available models.
-      </>
-    ),
-    isDisabled: (values) => !values.api_base,
-  },
-};
-
-export function FetchModelsButton({
-  llmProviderDescriptor,
-  existingLlmProvider,
-  values,
-  setFieldValue,
-  isFetchingModels,
-  setIsFetchingModels,
-  fetchModelsError,
-  setFetchModelsError,
-  setPopup,
-}: FetchModelsButtonProps) {
-  const config = providerConfigs[llmProviderDescriptor.name];
-
-  // Only render if the provider supports model fetching
-  if (!config) {
-    return null;
-  }
-
-  const handleFetchModels = () => {
-    fetchModels(
-      llmProviderDescriptor,
-      existingLlmProvider,
-      values,
-      setFieldValue,
-      setIsFetchingModels,
-      setFetchModelsError,
-      setPopup
-    );
-  };
-
-  return (
-    <div className="flex flex-col gap-2">
-      <Button
-        type="button"
-        onClick={handleFetchModels}
-        disabled={isFetchingModels || config.isDisabled(values)}
-        className="w-fit"
-      >
-        {isFetchingModels ? (
-          <>
-            <LoadingAnimation size="text-sm" />
-            <span className="ml-2">{config.loadingText}</span>
-          </>
-        ) : (
-          config.buttonText
-        )}
-      </Button>
-
-      {fetchModelsError && (
-        <Text className="text-red-600 text-sm">{fetchModelsError}</Text>
-      )}
-
-      <Text className="text-sm text-gray-600">{config.helperText}</Text>
-    </div>
-  );
-}
diff --git a/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx b/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx
index aa1fbc5d21b..38513f7ce91 100644
--- a/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx
+++ b/web/src/app/admin/configuration/llm/LLMProviderUpdateForm.tsx
@@ -5,6 +5,7 @@ import Text from "@/components/ui/text";
 import { Separator } from "@/components/ui/separator";
 import { Button } from "@/components/ui/button";
 import { Form, Formik } from "formik";
+import type { FormikProps } from "formik";
 import { FiTrash } from "react-icons/fi";
 import { LLM_PROVIDERS_ADMIN_URL } from "./constants";
 import {
@@ -13,19 +14,79 @@ import {
   MultiSelectField,
   FileUploadFormField,
 } from "@/components/Field";
-import { useState } from "react";
+import { useEffect, useRef, useState } from "react";
 import { useSWRConfig } from "swr";
 import {
   LLMProviderView,
   ModelConfiguration,
   WellKnownLLMProviderDescriptor,
 } from "./interfaces";
-import { FetchModelsButton } from "./FetchModelsButton";
+import { dynamicProviderConfigs, fetchModels } from "./utils";
 import { PopupSpec } from "@/components/admin/connectors/Popup";
 import * as Yup from "yup";
 import isEqual from "lodash/isEqual";
 import { IsPublicGroupSelector } from "@/components/IsPublicGroupSelector";
 
+function AutoFetchModelsOnEdit({
+  llmProviderDescriptor,
+  existingLlmProvider,
+  values,
+  setFieldValue,
+  setIsFetchingModels,
+  setFetchModelsError,
+  setPopup,
+}: {
+  llmProviderDescriptor: WellKnownLLMProviderDescriptor;
+  existingLlmProvider?: LLMProviderView;
+  values: any;
+  setFieldValue: FormikProps<any>["setFieldValue"];
+  setIsFetchingModels: (loading: boolean) => void;
+  setFetchModelsError: (error: string) => void;
+  setPopup?: (popup: PopupSpec) => void;
+}) {
+  const hasAutoFetchedRef = useRef(false);
+
+  useEffect(() => {
+    if (!existingLlmProvider) {
+      return;
+    }
+
+    const config = dynamicProviderConfigs[llmProviderDescriptor.name];
+    if (!config) {
+      return;
+    }
+
+    if (hasAutoFetchedRef.current) {
+      return;
+    }
+
+    if (config.isDisabled(values)) {
+      return;
+    }
+
+    hasAutoFetchedRef.current = true;
+    fetchModels(
+      llmProviderDescriptor,
+      existingLlmProvider,
+      values,
+      setFieldValue,
+      setIsFetchingModels,
+      setFetchModelsError,
+      setPopup
+    );
+  }, [
+    existingLlmProvider,
+    llmProviderDescriptor,
+    setFieldValue,
+    setFetchModelsError,
+    setIsFetchingModels,
+    setPopup,
+    values,
+  ]);
+
+  return null;
+}
+
 export function LLMProviderUpdateForm({
   llmProviderDescriptor,
   onClose,
@@ -350,9 +411,20 @@ export function LLMProviderUpdateForm({
         const currentModelConfigurations = getCurrentModelConfigurations(
           formikProps.values
         );
+        const dynamicConfig =
+          dynamicProviderConfigs[llmProviderDescriptor.name];
 
         return (
           <Form className="gap-y-4 items-stretch mt-6">
+            <AutoFetchModelsOnEdit
+              llmProviderDescriptor={llmProviderDescriptor}
+              existingLlmProvider={existingLlmProvider}
+              values={formikProps.values}
+              setFieldValue={formikProps.setFieldValue}
+              setIsFetchingModels={setIsFetchingModels}
+              setFetchModelsError={setFetchModelsError}
+              setPopup={setPopup}
+            />
             {!firstTimeConfiguration && (
               <TextFormField
                 name="name"
@@ -439,19 +511,49 @@ export function LLMProviderUpdateForm({
                 }
               }
             )}
-
             {/* Fetch models button - automatically shows for supported providers */}
-            <FetchModelsButton
-              llmProviderDescriptor={llmProviderDescriptor}
-              existingLlmProvider={existingLlmProvider}
-              values={formikProps.values}
-              setFieldValue={formikProps.setFieldValue}
-              isFetchingModels={isFetchingModels}
-              setIsFetchingModels={setIsFetchingModels}
-              fetchModelsError={fetchModelsError}
-              setFetchModelsError={setFetchModelsError}
-              setPopup={setPopup}
-            />
+            {dynamicConfig && (
+              <div className="flex flex-col gap-2">
+                <Button
+                  type="button"
+                  onClick={() =>
+                    fetchModels(
+                      llmProviderDescriptor,
+                      existingLlmProvider,
+                      formikProps.values,
+                      formikProps.setFieldValue,
+                      setIsFetchingModels,
+                      setFetchModelsError,
+                      setPopup
+                    )
+                  }
+                  disabled={
+                    isFetchingModels ||
+                    dynamicConfig.isDisabled(formikProps.values)
+                  }
+                  className="w-fit"
+                >
+                  {isFetchingModels ? (
+                    <>
+                      <LoadingAnimation size="text-sm" />
+                      <span className="ml-2">Fetching models...</span>
+                    </>
+                  ) : (
+                    "Fetch Available Models"
+                  )}
+                </Button>
+
+                {fetchModelsError && (
+                  <Text className="text-red-600 text-sm">
+                    {fetchModelsError}
+                  </Text>
+                )}
+
+                <Text className="text-sm text-gray-600">
+                  Retrieve the latest available models for this provider.
+                </Text>
+              </div>
+            )}
 
             {!firstTimeConfiguration && (
               <>
diff --git a/web/src/app/admin/configuration/llm/interfaces.ts b/web/src/app/admin/configuration/llm/interfaces.ts
index 951ddba8357..6c32ba96bd8 100644
--- a/web/src/app/admin/configuration/llm/interfaces.ts
+++ b/web/src/app/admin/configuration/llm/interfaces.ts
@@ -82,40 +82,27 @@ export interface LLMProviderDescriptor {
   model_configurations: ModelConfiguration[];
 }
 
-export interface ProviderFetchModelsButtonConfig {
-  buttonText: string;
-  loadingText: string;
-  helperText: string | React.ReactNode;
-  isDisabled: (values: any) => boolean;
-}
-
-export interface FetchModelsButtonProps {
-  llmProviderDescriptor: WellKnownLLMProviderDescriptor;
-  existingLlmProvider?: LLMProviderView;
-  values: any;
-  setFieldValue: any;
-  isFetchingModels: boolean;
-  setIsFetchingModels: (loading: boolean) => void;
-  fetchModelsError: string;
-  setFetchModelsError: (error: string) => void;
-  setPopup?: (popup: PopupSpec) => void;
-}
-
 export interface OllamaModelResponse {
   name: string;
   max_input_tokens: number;
   supports_image_input: boolean;
 }
 
-export interface FetchModelsConfig<
+export interface DynamicProviderConfig<
   TApiResponse = any,
   TProcessedResponse = ModelConfiguration,
 > {
   endpoint: string;
-  validationCheck: () => boolean;
-  validationError: string;
-  requestBody: () => Record<string, any>;
-  processResponse: (data: TApiResponse) => TProcessedResponse[];
+  isDisabled: (values: any) => boolean;
+  disabledReason: string;
+  buildRequestBody: (args: {
+    values: any;
+    existingLlmProvider?: LLMProviderView;
+  }) => Record<string, any>;
+  processResponse: (
+    data: TApiResponse,
+    llmProviderDescriptor: WellKnownLLMProviderDescriptor
+  ) => TProcessedResponse[];
   getModelNames: (data: TApiResponse) => string[];
   successMessage: (count: number) => string;
 }
diff --git a/web/src/app/admin/configuration/llm/utils.ts b/web/src/app/admin/configuration/llm/utils.ts
index e062da6dc5f..aa59f7c4b6a 100644
--- a/web/src/app/admin/configuration/llm/utils.ts
+++ b/web/src/app/admin/configuration/llm/utils.ts
@@ -14,7 +14,7 @@ import {
 import {
   WellKnownLLMProviderDescriptor,
   LLMProviderView,
-  FetchModelsConfig,
+  DynamicProviderConfig,
   OllamaModelResponse,
   ModelConfiguration,
 } from "./interfaces";
@@ -71,6 +71,63 @@ export const getProviderIcon = (
 export const isAnthropic = (provider: string, modelName: string) =>
   provider === "anthropic" || modelName.toLowerCase().includes("claude");
 
+export const dynamicProviderConfigs: Record<
+  string,
+  DynamicProviderConfig<any, ModelConfiguration>
+> = {
+  bedrock: {
+    endpoint: "/api/admin/llm/bedrock/available-models",
+    isDisabled: (values) => !values.custom_config?.AWS_REGION_NAME,
+    disabledReason: "AWS region is required to fetch Bedrock models",
+    buildRequestBody: ({ values, existingLlmProvider }) => ({
+      aws_region_name: values.custom_config?.AWS_REGION_NAME,
+      aws_access_key_id: values.custom_config?.AWS_ACCESS_KEY_ID,
+      aws_secret_access_key: values.custom_config?.AWS_SECRET_ACCESS_KEY,
+      aws_bearer_token_bedrock: values.custom_config?.AWS_BEARER_TOKEN_BEDROCK,
+      provider_name: existingLlmProvider?.name,
+    }),
+    processResponse: (data: string[], llmProviderDescriptor) =>
+      data.map((modelName) => {
+        const existingConfig = llmProviderDescriptor.model_configurations.find(
+          (config) => config.name === modelName
+        );
+        return {
+          name: modelName,
+          is_visible: existingConfig?.is_visible ?? false,
+          max_input_tokens: null,
+          supports_image_input: existingConfig?.supports_image_input ?? null,
+        };
+      }),
+    getModelNames: (data: string[]) => data,
+    successMessage: (count: number) =>
+      `Successfully fetched ${count} models for the selected region (including cross-region inference models).`,
+  },
+  ollama: {
+    endpoint: "/api/admin/llm/ollama/available-models",
+    isDisabled: (values) => !values.api_base,
+    disabledReason: "API Base is required to fetch Ollama models",
+    buildRequestBody: ({ values }) => ({
+      api_base: values.api_base,
+    }),
+    processResponse: (data: OllamaModelResponse[], llmProviderDescriptor) =>
+      data.map((modelData) => {
+        const existingConfig = llmProviderDescriptor.model_configurations.find(
+          (config) => config.name === modelData.name
+        );
+        return {
+          name: modelData.name,
+          is_visible: existingConfig?.is_visible ?? true,
+          max_input_tokens: modelData.max_input_tokens,
+          supports_image_input: modelData.supports_image_input,
+        };
+      }),
+    getModelNames: (data: OllamaModelResponse[]) =>
+      data.map((model) => model.name),
+    successMessage: (count: number) =>
+      `Successfully fetched ${count} models from Ollama.`,
+  },
+};
+
 export const fetchModels = async (
   llmProviderDescriptor: WellKnownLLMProviderDescriptor,
   existingLlmProvider: LLMProviderView | undefined,
@@ -80,73 +137,13 @@ export const fetchModels = async (
   setFetchModelsError: (error: string) => void,
   setPopup?: (popup: PopupSpec) => void
 ) => {
-  // Provider-specific configurations
-  const providerConfigs: Record<string, FetchModelsConfig> = {
-    bedrock: {
-      endpoint: "/api/admin/llm/bedrock/available-models",
-      validationCheck: () => !!values.custom_config?.AWS_REGION_NAME,
-      validationError: "AWS region is required to fetch Bedrock models",
-      requestBody: () => ({
-        aws_region_name: values.custom_config?.AWS_REGION_NAME,
-        aws_access_key_id: values.custom_config?.AWS_ACCESS_KEY_ID,
-        aws_secret_access_key: values.custom_config?.AWS_SECRET_ACCESS_KEY,
-        aws_bearer_token_bedrock:
-          values.custom_config?.AWS_BEARER_TOKEN_BEDROCK,
-        provider_name: existingLlmProvider?.name,
-      }),
-      processResponse: (data: string[]): ModelConfiguration[] =>
-        data.map((modelName) => {
-          const existingConfig =
-            llmProviderDescriptor.model_configurations.find(
-              (config) => config.name === modelName
-            );
-          return {
-            name: modelName,
-            is_visible: existingConfig?.is_visible ?? false,
-            max_input_tokens: null,
-            supports_image_input: existingConfig?.supports_image_input ?? null,
-          };
-        }),
-      getModelNames: (data: string[]) => data,
-      successMessage: (count: number) =>
-        `Successfully fetched ${count} models for the selected region (including cross-region inference models).`,
-    },
-    ollama: {
-      endpoint: "/api/admin/llm/ollama/available-models",
-      validationCheck: () => !!values.api_base,
-      validationError: "API Base is required to fetch Ollama models",
-      requestBody: () => ({
-        api_base: values.api_base,
-      }),
-      processResponse: (data: OllamaModelResponse[]): ModelConfiguration[] =>
-        data.map((modelData) => {
-          const existingConfig =
-            llmProviderDescriptor.model_configurations.find(
-              (config) => config.name === modelData.name
-            );
-          return {
-            name: modelData.name,
-            is_visible: existingConfig?.is_visible ?? true,
-            max_input_tokens: modelData.max_input_tokens,
-            supports_image_input: modelData.supports_image_input,
-          };
-        }),
-      getModelNames: (data: OllamaModelResponse[]) =>
-        data.map((model) => model.name),
-      successMessage: (count: number) =>
-        `Successfully fetched ${count} models from Ollama.`,
-    },
-  };
-
-  const config =
-    providerConfigs[llmProviderDescriptor.name as keyof typeof providerConfigs];
+  const config = dynamicProviderConfigs[llmProviderDescriptor.name];
   if (!config) {
     return;
   }
 
-  // Validation check
-  if (!config.validationCheck()) {
-    setFetchModelsError(config.validationError);
+  if (config.isDisabled(values)) {
+    setFetchModelsError(config.disabledReason);
     return;
   }
 
@@ -159,7 +156,9 @@ export const fetchModels = async (
       headers: {
         "Content-Type": "application/json",
       },
-      body: JSON.stringify(config.requestBody()),
+      body: JSON.stringify(
+        config.buildRequestBody({ values, existingLlmProvider })
+      ),
     });
 
     if (!response.ok) {
@@ -174,7 +173,10 @@ export const fetchModels = async (
     }
 
     const availableModels = await response.json();
-    const updatedModelConfigs = config.processResponse(availableModels);
+    const updatedModelConfigs = config.processResponse(
+      availableModels,
+      llmProviderDescriptor
+    );
     const availableModelNames = config.getModelNames(availableModels);
 
     // Store the updated model configurations in form state instead of mutating props