Semantic kernel memory connector#69

[tikhandesanket]

Description

Adds a Semantic Kernel VectorStore adapter backed by Dynavec, allowing Semantic Kernel applications to use Dynavec for vector storage, retrieval, filtering, and namespaces.

Related issue

Fixes #69

Changes

• Added DynavecStore and DynavecCollection Semantic Kernel VectorStore integration.
• Added support for upsert, get, delete, vector search, metadata filtering, and namespace-based collections.
• Added optional semantic-kernel>=1.44 dependency and included it in the all extra.
• Added unit tests covering CRUD operations, vector search, metadata filters, collection creation, and collection listing.
• Added a runnable Semantic Kernel vector store example.

Testing

• Tests pass locally
• Ruff checks pass
• Documentation updated, if applicable

Test results

• Semantic Kernel tests: 8 passed
• Ruff: Passed
• git diff --cached --check: Passed
• Pre-commit hooks: Passed

Checklist

• My changes are focused and relevant to this pull request.
• I have added or updated tests where appropriate.
• I have reviewed my changes for unrelated modifications.
• I have updated documentation where necessary.

[Kaap10]

LGTM!

[codeforstartups]

Really nice Semantic Kernel connector, @tikhandesanket — DynavecCollection / DynavecStore with a clean example. One blocker before merge, and it is not visible on this branch: strict mypy landed on development via #207, and it is now a required CI job. This PR predates it, so the typecheck job never ran here — but I merged current development in locally and ran it, and it fails. Two things to fix:

1. Add semantic-kernel to the typecheck extra in pyproject.toml (alongside langchain-core / llama-index-core / haystack-ai), or CI mypy cannot resolve the import at all.

2. Make src/dynavec/integrations/semantic_kernel.py strict-mypy clean — 14 errors right now. Grouped:

Mechanical:

• data = {...} dicts inferred as dict[str, str] then assigned vectors/None → annotate data: dict[str, Any] = {} (lines ~210/216/218).
• _inner_get missing return type + param annotations (line 71).
• DynavecCollection used without type args (line 298) → DynavecCollection[Any, Any] (or the real key/model params).
• _get_score_from_result returns Any where float | None is declared (159) → cast.

Need your SK-API knowledge (v1.44):

• _serialize_dicts_to_store_models / _deserialize_store_models_to_dicts violate the base signatures (Liskov, lines 162/199) — match SK’s exact Sequence[...] / TModel signatures.
• VectorStoreCollection.__init__ definition / collection_name argument types (53/54) and _inner_upsert returning list[str] vs Sequence[TKey] (69).
• filter= passed to Dynavec.search is SK’s union type (129/137) — narrow/cast to the dict dynavec expects.

Run make run-ci locally after uv pip install -e ".[dev,ingest,typecheck,semantic-kernel]" to reproduce. If a couple of the override mismatches are genuinely irreconcilable between SK’s generics and dynavec’s Document, a targeted # type: ignore[override] on just those two methods is acceptable — but let’s keep it to those.

I did not merge because it would turn development’s required mypy job red. Ping me once it’s green and I’ll merge right away. Thanks! 🙌