29ba806fc6
* feat(knowledge): add independent Wiki and entity relation retrieval tools * fix(knowledge): address review diagnostics and CI catalog coverage * fix(knowledge): deduplicate graph evidence and update catalog e2e * fix(knowledge): clarify bounded evidence and pin browser CI runners * Update e2e-tests to run on ubuntu-latest * Update workflow to use ubuntu-latest * test(ci): stop pinning maintainer-controlled browser runner images --------- Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
1.2 KiB
LightRAG retrieval extension
Keep HTTP calls, connection bindings and response mapping in this package. Use the core contextual tool dispatcher and unchanged source envelope. Entity/edge summaries are not document excerpts. Join source_id to returned chunk_id only; never infer edges or multi-hop paths, or use file_path alone as evidence. Reject selected scope, missing bindings, HTTP errors and redirects without retrying with broader credentials. Offload SSL client construction. Tests: backend/tests/test_lightrag_relations_extension.py and the shared three-tool suite. Their responses are synthetic. Separate opt-in checks in backend/tests/test_knowledge_phase1_live.py require existing indexed fixtures; the GLM loop requires KNOWLEDGE_RUN_MODEL_TEST=1. See VALIDATION.md for live results and limits. Never let live tests upload data or silently enable API usage.
Shared graph evidence chunks are emitted once per returned chunk ID, before the first referencing edge. Every edge names its evidence result positions; prefix truncation therefore cannot leave a retained edge pointing to omitted evidence.
At the hit limit, distinguish omitted evidence from absent retrieved evidence.