v0.0.3

What's Changed

• feat: Upgrade CI Milvus stack to v3.0 and pull MinIO from quay.io by @awinml in #12
• feat: Add hierarchical community detection by @awinml in #13
• feat: Isolate per-item failures in GraphStore bulk writes by @vrunm in #18
• feat: Track document identity and versioning in the graph by @vrunm in #19
• docs: Regenerate API reference and fix malformed docstring wraps by @awinml in #24
• feat: Add candidate search and non-destructive entity resolution by @awinml in #25
• ci: Pre-commit autoupdate by @pre-commit-ci[bot] in #10
• chore(deps): Bump actions/deploy-pages from 5.0.0 to 5.0.1 by @dependabot[bot] in #11
• chore(deps): Bump astral-sh/setup-uv from 10.0.1 to 10.1.0 by @dependabot[bot] in #14
• feat: Add schema-aware research loop with flexible graph tools by @awinml in #26
• feat: Resolve entity description conflicts via LLM by @vrunm in #27
• feat: Send neighbors and similarity to batched LLM verification by @vrunm in #28
• fix: Synthesize SimpleAgent answers with the LLM by @vrunm in #30
• feat: Crash-safe graph mutation with non-destructive resolution by @awinml in #29
• chore(deps): Bump codecov/codecov-action from 7.0.0 to 7.1.1 by @dependabot[bot] in #31
• test: Consolidate the test suite around E2E coverage by @awinml in #32
• feat: Return the run ledger from agent ainvoke by @awinml in #33
• feat: Add eval judge, median metric and CI evals by @awinml in #34
• feat: Trace agent runs using OpenInference callback by @awinml in #35
• feat: Add extraction quality evaluation by @awinml in #37
• feat: Add answer quality evaluation by @awinml in #38
• feat: Add entity resolution quality evaluation by @awinml in #39
• feat: Add verifier calibration eval by @awinml in #40
• ci: Move shared eval workflow steps into a shared workflow by @awinml in #41
• feat: Per-sentence citation report and noise-tolerant answer gate by @awinml in #42
• feat: Add agent trajectory evaluation by @vrunm in #43
• feat: Add per-document loader spans, retire traced decorator by @vrunm in #44
• refactor: Fix loaders/ingestion cycle, move StageFailure to data_models by @vrunm in #45
• feat: Trace the ingestion pipeline with OpenTelemetry spans by @vrunm in #46
• feat: Trace graph store, vector store and embedder with OpenTelemetry by @vrunm in #47
• feat: Trace every LLM call with per-request spans by @vrunm in #48
• refactor: Move rule files into project skills by @awinml in #49
• feat: Add OpenTelemetry tracing to retrieval and agent tools by @awinml in #50
• feat: Rework chunking with an explicit Chunker contract, rules and new strategies by @awinml in #51
• feat: Add benchmark harness by @awinml in #57
• chore: bump version to 0.0.3 by @vrunm in #58

New Contributors

• @pre-commit-ci[bot] made their first contribution in #10
• @dependabot[bot] made their first contribution in #11

Full Changelog: v0.0.2...v0.0.3