feat(connectors): add the Qdrant connector (embeds via external service) [roadmap:rac-connectors]#3
Merged
Conversation
Adds a `category` field to each connector page and groups the generated README Connectors section under kind headings (Memory & RAG, Knowledge graph). Grouping is by kind, not direction, so a provider that does both ingest and export (e.g. a future Atlassian suite) sits in one kind group. Re-applies the grouping that did not reach main.
…ce) [roadmap:rac-connectors] Adds a Qdrant vector-store connector and the generic external-embedding seam it needs. Qdrant stores vectors but does not produce them, so the connector embeds each record's text via a configured OpenAI-compatible endpoint (a LiteLLM gateway the reference deployment) before upserting; RAC bundles no model — the Embedder seam uses stdlib HTTP. Idempotent on the canonical id (point id = uuid5(id)), one collection per corpus source. Records the decision as ADR-009 in the connectors corpus: vector-store connectors embed via an external service, a scoped exception to the model-free invariant (rac-core ADR-002/ADR-066 unchanged). Adds the qdrant extra, the rac-connect qdrant CLI subcommand, fake-driven tests, and docs/connectors/qdrant.md.
…[roadmap:rac-connectors] Records the exact steps to validate the Qdrant connector end to end against a real Qdrant + embeddings endpoint (the live path is drafted/unproven), and the verification to flip the page to shipped.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds a Qdrant vector-store connector to rac-connectors, plus the generic external-embedding seam it needs, and records the design as ADR-009.
Why Qdrant is different
The existing connectors (Supermemory, Mem0, Zep) embed server-side, so they stay model-free. Qdrant stores vectors but doesn't produce them — self-hosted OSS has no server-side embedding — so the connector must embed first. To keep RAC AI-optional (rac-core ADR-002/ADR-066) and bundle no model, it embeds via a configured external OpenAI-compatible
/embeddingsendpoint (a LiteLLM gateway is the reference deployment, mirroring the grounding benchmark's*_BASE_URLpattern). Model + keys live in that endpoint, never in RAC.What's in it
src/rac_connectors/embedding.py— a smallEmbedderProtocol + a stdlib-HTTPExternalEmbedder(no SDK/dependency); env configRAC_EMBED_BASE_URL/RAC_EMBED_MODEL/RAC_EMBED_API_KEY.src/rac_connectors/qdrant/—QdrantConnectoron the existingConnector.pushseam: embeds each record, upserts a point. Idempotent on the canonical id (point id =uuid5(id)); one collection per corpussource; payload{rac_id, type, status, title, text, …metadata}.QdrantClientProtocol + lazily-importedSdkQdrantClient.qdrantextra (qdrant-client),rac-connect qdrantCLI subcommand, fake-driven tests (connector, CLI, embedder), anddocs/connectors/qdrant.md(category Memory & RAG).rac/decisions/): vector-store connectors embed via an external service — a scoped exception to the model-free invariant; engine unchanged. Generic, so future vector stores follow one rule.Also: re-lands the kind-grouping
Heads-up — the kind-grouping from PR #2 didn't actually reach
main(only the CHANGELOG commit landed). This PR re-applies it: acategoryfield per connector page and grouped README headings (Memory & RAG / Knowledge graph), with Qdrant under Memory & RAG. Worth checking how PR #2's merge dropped that commit.Verification (local)
ruff check+ruff format --check(src/tests/scripts) clean;mypy src/clean; 142 tests pass;sync_readme.py --checkin sync (7 connectors).rac validate rac/→ 11/11 valid;rac relationships rac/ --validate→ 0 issues.Status is
drafted (live run pending)— no live Qdrant/embedding run has been exercised yet. Merge with a merge commit.Execution tracking: asdecided/core#228 (
rac-connectors).