feat: add Dakera document store integration

[ferhimedamine]

Related Issues

New integration — no prior issue. Happy to open a tracking issue if the team prefers.

Proposed Changes

This PR adds a new integration, dakera-haystack (integrations/dakera/), that lets Haystack pipelines store and retrieve documents from Dakera — a self-hosted memory server that provides persistent, decay-weighted vector recall.

Components

• DakeraDocumentStore (sync + async) — implements the standard document-store surface (count_documents, write_documents, filter_documents, delete_documents, plus _embedding_retrieval) and their _async counterparts. It drives Dakera's raw vector-namespace API (upsert / query / delete / get_namespace / configure_namespace), so embeddings are supplied by any Haystack embedder while Dakera handles storage and similarity search — the same model as the other cloud/self-hosted vector-store integrations.
• DakeraEmbeddingRetriever — a @component for dense retrieval with run / run_async and full FilterPolicy support.

Design notes

• Connection is self-hosted: url (default http://localhost:3000, the port used by the dakera-deploy docker-compose stack) plus an api_key Secret from DAKERA_API_KEY. No metered/hosted secret.
• Namespace lifecycle: on first use the store adopts the dimension of an existing namespace, or creates it via configure_namespace (upsert semantics) when absent.
• Filters: filters.py converts Haystack's filter syntax into Dakera's MongoDB-style DSL ({"field": {"$eq": ...}}, $and/$or, $in/$nin, comparison operators), mirroring the approach used by the Pinecone/Qdrant document stores.
• Metadata-only filtering reuses the established dummy-vector query pattern, and content is stored as a content metadata field.
• Registered in .github/labeler.yml, .github/workflows/CI_coverage_comment.yml, the root README.md inventory table, and a per-integration .github/workflows/dakera.yml.

How did you test it?

• 48 unit tests (mocked DakeraClient) covering init/lazy-init, namespace adoption vs. creation, to_dict/from_dict round-trips, metadata validation, document↔vector conversion, dummy-vector handling, the full filter converter, and the retriever. All green.
• Integration tests (test_integration.py) exercise a real write → count → embedding-retrieval → filter → delete round-trip against a live server; they auto-skip unless DAKERA_URL is set (start a server with dakera-deploy).
• Local run against haystack-ai 2.30.2 and dakera 0.12.8: hatch run test:unit → 48 passed, hatch run test:types → clean, ruff check/ruff format --check → clean.

Notes for the reviewer

• The integration folder is dakera (no hyphen) per the format check; the PyPI package name is dakera-haystack.
• Integration tests need the Dakera server + MinIO (via dakera-deploy), so the dakera.yml integration step is a no-op skip in CI until a DAKERA_URL is provided — unit tests, lint, and types run on every PR.

Checklist

• I have read the contributors guidelines and the code of conduct
• I added unit tests and updated the docstrings
• I've used a conventional commit type for my PR title (feat:)

[CLAassistant]

All committers have signed the CLA.

[ferhimedamine]

I have read the CLA Document and I hereby sign the CLA

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	pypi/​dakera@​0.12.8

View full report