feat: Add Supabase integration with PgVector document store and embedding retriever

[pontuselmrin]

Related Issues

• fixes SupabasePgVectorDocumentStore and SupabasePgvectorEmbeddingRetriever #3081

Proposed Changes:

• Adds a new supabase integration (integrations/supabase/) that provides SupabasePgVectorDocumentStore and SupabasePgvectorEmbeddingRetriever as thin wrappers around the existing pgvector-haystack integration, pre-configured for Supabase's connection model.
• They subclass the PgvectorDocumentStore and PgvectorEmbeddingRetriever, respectively

How did you test it?

The testing functionality is done using a Supabase instance but because the feature is wrapping the pgvector integration the tests are essentially lifted from there.

• Unit tests covering init (defaults, custom params, invalid store), serialization (to_dict/from_dict), run, and run_async
• All existing integration tests continue to pass
• integrated a full CI workflow via supabase.yml

Notes for the reviewer

Tested on Mac OS and Supabase CLI run via Docker locally.

Checklist

• I have read the contributors guidelines and the code of conduct
• I have updated the related issue with new insights and changes
• I added unit tests and updated the docstrings
• I've used one of the conventional commit types for my PR title: fix:, feat:, build:, chore:, ci:, docs:, style:, refactor:, perf:, test:.

[CLAassistant]

All committers have signed the CLA.

[srini047]

Do we have to rename the file from document_store.py to pg_document_store.py the reason for the same is this issue makes use of SupabaseVectorBucketDocumentStore. Hence the question.
cc @davidsbatista

[pontuselmrin]

I think its a good idea to wait for this one to be merged. I already scaffolded the integration so I think it could simplify the workflow for you a bit. @srini047

[srini047]

Yes, that's what even I think orelse the effort gets duplicated here. So only my question on file naming remains.

[davidsbatista]

let's first merge this one - it's already in a very good state, just a few things need being addressed

[davidsbatista]

@pontuselmrin I did a first review and pushed some adjustments and fixed a few issues. But, overall, the PR looks pretty good!

Just some small improvements to complete it, and it should be ready to go:

• We can also add a SupabasePgvectorKeywordRetriever as with the embedding retriever inheriting from the PgvectorEmbeddingRetriever
• No retriever integration test, all the tests are still unit tests, we should add some integration tests that run against a docker instance
• The README is missing setup instructions, env var, connection modes, and examples

Let me know if you have any questions/doubts

[pontuselmrin]

@davidsbatista Thanks for the review David! I'm working on addressing your points.

[davidsbatista]

@pontuselmrin thanks for the corrections/additions and adjustments. I worked a bit on your branch, namely:

• Dropped create_extension=True from test fixtures — Supabase pre-enables pgvector, so the override was incorrect; CI workflow and local docker-compose.yml now pre-create the extension instead
• Replaced test_retrieval.py + test_retriever.py with one file per retriever, matching the Elasticsearch/OpenSearch convention
• Removed test classes in favour of module-level functions; @pytest.mark.integration is now applied per function, also following what Elasticsearch/OpenSearch integrations do
• Added missing tests: test_from_dict_without_filter_policy, test_run_with_filters, test_run_async_with_filters, and split test_init_default into focused single-concern tests

[davidsbatista]

Ot looks good - green - I will give it another look tomorrow morning and then merge.

In the mean time, @pontuselmrin do you want to open a PR here: https://github.com/deepset-ai/haystack-integrations so that this integration is listed ou our website?

[pontuselmrin]

@davidsbatista Thanks for going over it again, I'm opening the PR at the integrations page now :)

[davidsbatista]

thanks for this @pontuselmrin - we appreciated it! :)