Zed AI: OpenRouter Support

[daun-io]

Hi, I'm from OpenRouter and a fan of Zed and would love to see OpenRouter integration as a new assistant provider.
OpenRouter supports dozens of the best coding models and hundreds of different models with one API key and an OpenAI-style schema.
I saw that some Zed community users have requested this feature 1 2.
Is there anything I can do to help? Thanks! 💙

[deanrie]

It would be nice to have this support.

[mintisan]

It would be nice to have this support.

[minato7namikazi]

it would be awesome : )

[tm17-abcgen]

Would love to have this too. ZedAI currently works with Anthropic only. Since OpenRouter has access to Anthropic, i hope that ZedAI can be used with OpenRouter too, to make use of the prompt caching also!

[dkphhh]

we need this！

[hrefhref]

isn't this already supported : #13276

[drivencompassion]

Awaiting OpenRouter config?

[gliptic]

OpenRouter's API is OpenAI compatible, so just e.g.:

  "language_models": {
    "openai": {
      "version": "1",
      "api_url": "https://openrouter.ai/api/v1",
      "available_models": [
        {
          "display_name": "Qwen2.5-72B",
          "name": "qwen/qwen-2.5-72b-instruct",
          "max_tokens": 131072
        }
      ]
...

[drivencompassion]

Is there an example of configuring multiple models like openai/chatgpt-4o-latest, o1-mini & o1-preview?

[drivencompassion]

[drivencompassion]

[drivencompassion]

In addition to the LLM providers (e.g., the OpenAI API Key option), may there be an option for OpenRouter?

Run assistant (shift+command+p): show configuration

[bfulop]

I just had the same issue:

I had to add my Openrouter API key in the export OPENAI_API_KEY = env var and restart zed

[drivencompassion]

Where else is env var stored? I get an invalid settings error when trying to add "export OPENAI_API_KEY": "" to settings.json.

[bfulop]

In my case it's in .zshrc file in my home directory.

[bfulop]

If you don't have this set in you shell, then you can set it via the assistant: show configuration and replace the OpenAI key with your openrouter key. In my case because it's set via the shell I cannot edit it.

[azdanov]

Does it work for anyone else?

[knealking]

It worked for me. But I had to restart the app.

[phil65]

Seems as if right now we cant "mix" multiple OpenAI(-compatible) APIs because they all would have to share the same API key. Would be great if Zed would get rid of these restrictions so that someone use multiple different OpenAi-compatible APIs simultaniously. :)

[netapy]

+1, is the Zed team on this ? It would massively open up the ability to use any models at token level pricing

[TYTheBeast]

Waiting on this as well, I just discovered Zed and this missing integration is a killer, I have an OpenRouter API key but no Anthropic API key.

[pconerly]

+1 for this. I want to throw deepseek-r1 at my problems!

[bfulop]

I'm using this. Not perfect (it overrides openai) but works:

  "language_models": {
    "openai": {
      "api_url": "https://openrouter.ai/api/v1",
      "available_models": [
        {
          "name": "deepseek/deepseek-r1",
          "display_name": "DeepSeek R1",
          "max_tokens": 64000
        },
        {
          "name": "deepseek/deepseek-chat",
          "display_name": "DeepSeek V3",
          "max_tokens": 64000
        }
      ],
      "version": "1"
    }
  },

Then in my .zshrc I replaced the OPENAI_API_KEY with the openrouter key.

[pconerly]

Thanks! That worked. FYI @drivencompassion , i put these settings into ~/.config/zed/settings.json, and then configured the 'openai' env variable via the Zed UI.

screenshot:

[drivencompassion]

After configuring this method in settings, and in .zshrc replacing key listed in OPENAI_API_KEY with the openrouter key, while keeping OPENAI_API_KEY=sk-or-v1- … of openrouter key, and testing a prompt, going to openrouter activity shows OpenAI: o1-mini model was selected, not DeepSeek R1 Anyone else try this and see different activity?

…

On Jan 21, 2025, at 4:47 AM, Bálint Fülöp ***@***.***> wrote: I'm using this. Not perfect (it overrides openai) but works: "language_models": { "openai": { "api_url": "https://openrouter.ai/api/v1", "available_models": [ { "name": "deepseek/deepseek-r1", "display_name": "DeepSeek R1", "max_tokens": 64000 }, { "name": "deepseek/deepseek-chat", "display_name": "DeepSeek V3", "max_tokens": 64000 } ], "version": "1" } }, Then in my .zshrc I replaced the OPENAI_API_KEY with the openrouter key. — Reply to this email directly, view it on GitHub <#16576 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAFZMFDWHAJA6WSVKXHBKXL2LYJUXAVCNFSM6AAAAABM3LHOWOVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCOJQGAZTCMA>. You are receiving this because you commented.

[drivencompassion]

~/.config/zed/settings.json, updated … how did you configure the 'openai' env variable via the Zed UI, as DeepSeek R1 is not an option under my config options like your screenshot.

…

---- On Jan 21, 2025 at 12:41:25 PM, Peter Conerly wrote: Thanks! That worked. FYI @drivencompassion , i put these settings into ~/.config/zed/settings.json, and then configured the 'openai' env variable via the Zed UI. screenshot: Screenshot.2025-01-21.at.9.30.36.AM.png (view on web) — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

[pconerly]

In the context window (or the "llm chat window") there's a 3-dot button to open options, and one of those options is "Configure". I also gave Zed a restart but I don't know if that was necessary

[drivencompassion]

Tried rebooting and restarting app, but 3-dot options does not have an option for DeepSeek R1, although settings does have: "language_models": { "openai": { "api_url": "https://openrouter.ai/api/v1", "available_models": [ { "name": "deepseek/deepseek-r1", "display_name": "DeepSeek R1", "max_tokens": 64000 }, { "name": "deepseek/deepseek-chat", "display_name": "DeepSeek V3", "max_tokens": 64000 } ], "version": "1" } }

…

---- On Jan 21, 2025 at 1:40:15 PM, Peter Conerly wrote: In the context window there's a 3-dot button to open options, and one of those options is "Configure". I also gave Zed a restart but I don't know if that was necessary Screenshot.2025-01-21.at.10.38.43.AM.png (view on web) — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

[amin-abouee]

I've tried both OpenRouter and the experimental Gemini model, but neither worked for me.

"language_models": {
"openai": {
"api_url": "https://openrouter.ai/api/v1",
"available_models": [
{
"name": "deepseek/deepseek-r1",
"display_name": "DeepSeek R1",
"max_tokens": 64000
},
{
"name": "deepseek/deepseek-chat",
"display_name": "DeepSeek V3",
"max_tokens": 64000
}
],
"version": "1"
},
"google": {
"available_models": [
{
"name": "gemini-exp-1206",
"display_name": "Gemini 1206",
"max_tokens": 1000000
},
{
"name": "gemini-2.0-flash-exp",
"display_name": "Gemini 2.0 Flash",
"max_tokens": 1000000
}
]
}
},

[realmorrisliu]

Hey folks, just wanted to chime in — I recently put together a Rust SDK for OpenRouter, and I’ve also started working on adding OpenRouter support to Zed. I’ve been using OpenRouter quite a bit, and figured it’d be great if Zed could support it out of the box.

It seems no one else is actively working on this, so I’ve decided to take it on. Happy to collaborate if anyone’s interested!

[imumesh18]

Hey @sambarnes , needed your help verify something we are sending below headers but on the openrouter dashboard it's not getting tagged properly. It is getting tagged as unknown. Can you point out are we doing something wrong here.

        .header("HTTP-Referer", "zed.dev")
        .header("X-Title", "Zed Editor");

[imumesh18]

Never mind fixed it. It seems that openrouter requires url in this format: https://zed.dev.

[sambarnes]

@imumesh18 yo just waking up, best guess is put https:// in front of the url

but assuming y'all tried that, once I'm at my computer I can try to take a closer look and repro -- there's some debt in this part of our system so could be on us here

[imumesh18]

Hey thanks @sambarnes for responding you are correct need to prefix https:// before the url for it to get tagged.

[sambarnes]

right on! can't wait to see zed climb the leaderboard (:

[jedisct1]

Pretty crazy that Openrouter is still not supported.

[imumesh18]

Hey there is open pr: #29496. The best way you can speed up getting this landed is by testing it and reporting issues/feedback.

[mishaal79]

hey any update on support for openrouter ?
can we build this and make a pull request ?

[imumesh18]

Hey there is open pr: #29496. The best way you can speed up getting this landed is by testing it and reporting issues/feedback.

[vavans]

The configuration

"language_models": {
    "openai": {
      "api_url": "https://openrouter.ai/api/v1",
      "version": "1"
    }

is working for some open ai models (gpt-4.1 for instance) but i got an error for o4-mini: