Make chat routing mandatory and document retrieval opt-in

[KostasCherv]

Summary

• make chat action routing mandatory for workspace and agent chat
• default the router to the registered main LLM while preserving router-specific overrides
• retrieve document collections only after an explicit answer_from_rag decision
• gate RAG citations on that decision and keep explicit session attachments scoped separately
• return structured router_error failures instead of silently retrieving all documents
• update configuration docs and deterministic regression coverage

Root cause

When the router was disabled or unavailable, should_use_workspace_resources(None) defaulted to retrieving every ready workspace document. Citation selection then treated retrieved context as supporting evidence even for direct messages such as hi, so the backend emitted an unrelated citation that the UI rendered.

User impact

Greetings and other direct/tool/clarification turns no longer query or cite workspace or agent document collections. Document retrieval remains available for explicit RAG routes, and router failures terminate without retrieval.

Validation

• uv run pytest -q tests/test_api.py tests/test_rag_chat_helpers.py tests/test_citations.py tests/test_llm_factory.py tests/test_ai_regression_gate.py — 166 passed
• uv run ruff check src tests — passed
• full suite after rebase — 574 passed, 1 unrelated concurrency timing test failed once while reaching Supabase; the exact failing test passed immediately on rerun
• graphify update . — completed

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
cortex	Ready	Preview, Comment	Jul 14, 2026 8:58am