Alex a20402a6dc fix(graphrag): give each extraction thread its own LLM ...

Provider-reported usage is kept on the LLM instance (_last_usage) and claimed
by whichever call finishes next. With GRAPHRAG_EXTRACTION_WORKERS > 1 (the
default is 8) every extraction thread shared one instance, so a call could
claim another call's provider counts while its own fell back to the estimate:
token_usage rows, and the cost they bill, could be attributed to the wrong
call and summed wrong.

Each pool thread now builds its own extraction LLM on first use. The calling
thread's instance is still built up front, so a misconfigured model fails the
run before any chunk is touched.

2026-09-19 14:33:14 +01:00

..

__init__.py

feat(graphrag): config.graph sub-model + settings + pgvector availability gate

2026-06-23 00:39:27 +01:00

test_extraction.py

fix(graphrag): give each extraction thread its own LLM

2026-09-19 14:33:14 +01:00

test_graph_search_tool.py

feat(agents): let agents walk a source's graph when they can search it

2026-09-19 14:07:41 +01:00

test_graphrag_available.py

refactor: rename the application package to docsgpt

2026-09-07 10:20:43 +01:00

test_naming.py

fix(graphrag): fold an entity's singular and plural onto one key

2026-09-19 14:33:14 +01:00

test_retriever_default_path.py

feat(graphrag): seed the walk from what entities are, and rank with passages and vector hits

2026-09-19 14:07:41 +01:00

test_retriever_live.py

refactor: rename the application package to docsgpt

2026-09-07 10:20:43 +01:00

test_retriever_passages.py

feat(graphrag): seed the walk from what entities are, and rank with passages and vector hits

2026-09-19 14:07:41 +01:00

test_retriever_seeding.py

feat(graphrag): seed the walk from what entities are, and rank with passages and vector hits

2026-09-19 14:07:41 +01:00

test_store.py

feat(graphrag): seed the walk from what entities are, and rank with passages and vector hits

2026-09-19 14:07:41 +01:00