feat: Dr v0 - initial work to unify flows and set up for Deep Research #5010
Conversation
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎
|
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
There was a problem hiding this comment.
PR Summary
Initial Deep Research (DR) implementation that unifies knowledge graph and basic search flows, introducing a new decision-based routing system for query processing.
- Added new
TimeBudgetenum inbackend/onyx/agents/agent_search/models.pyto configure search depth with FAST/SHALLOW/DEEP options, laying groundwork for more sophisticated research capabilities - Implemented decision routing in
backend/onyx/agents/agent_search/dr/nodes/dr_a1_orchestrator.pythat dynamically chooses between search or knowledge graph paths, but lacks proper error handling for non-FAST time budgets - Enhanced search result handling in
backend/onyx/agents/agent_search/basic/models.pywith newBasicSearchProcessedStreamResultsfor better structure of answers and citations - Changed KB search in
backend/onyx/agents/agent_search/kb_search/nodes/d1_generate_answer.pyto properly capture final answers, but some paths lack test coverage - Replaced KB graph with DR graph in
backend/onyx/chat/answer.py, using temporaryTMP_DRALPHA_PERSONA_NAMEconstant suggesting transitional implementation
23 files reviewed, 32 comments
Edit PR Review Bot Settings | Greptile
| return BasicSearchProcessedStreamResults( | ||
| ai_message_chunk=cast(AIMessageChunk, tool_call_chunk), | ||
| full_answer=full_answer, | ||
| cited_references=[], | ||
| retrieved_documents=[], | ||
| ) |
There was a problem hiding this comment.
logic: cited_references and retrieved_documents are always empty lists. Consider populating from final_search_results if available
There was a problem hiding this comment.
I agree with greptile, either these should be populated or the fields should be removed (if you never use them later)
| return BasicSearchProcessedStreamResults( | ||
| ai_message_chunk=cast(AIMessageChunk, tool_call_chunk), | ||
| full_answer=full_answer, | ||
| cited_references=[], | ||
| retrieved_documents=[], | ||
| ) |
There was a problem hiding this comment.
I agree with greptile, either these should be populated or the fields should be removed (if you never use them later)
backend/onyx/agents/agent_search/orchestration/nodes/use_tool_response.py
Outdated
Show resolved
Hide resolved
| Here are the entity types that are available in the knowledge graph: | ||
| {SEPARATOR_LINE} | ||
| ---possible_entities--- | ||
| {SEPARATOR_LINE} |
There was a problem hiding this comment.
since you're doing two distinct rounds of substitution, I do think it's fair to do something like {{possible_entities}} instead of the raw string replace
Orbital-Web
changed the title
Description
How Has This Been Tested?
locally in multi-tenant environment
Backporting (check the box to trigger backport action)
Note: You have to check that the action passes, otherwise resolve the conflicts manually and tag the patches.