Hi HN, I’m part of the DAIS Lab at Polytechnique Montréal, and we just released v0.7.0 of Flock, our open-source DuckDB extension. If you want semantic and analytical processing in one place, Flock lets you do it all in SQL. Instead of moving data out of your database into a Python script to run AI workflows, Flock allows you to execute LLM operators and RAG pipelines natively inside DuckDB.

In this new release, we've added a few major updates:

- Anthropic (Claude) & Multi-Provider Support: We now support OpenAI, Azure, Ollama, and Anthropic. You can define a model once with CREATE MODEL, then swap providers on the admin side without changing the downstream SQL queries.

- WASM Support: Flock now compiles and runs entirely inside DuckDB-WASM.

- LLM Metrics Tracking: We added end-to-end observability so you can track token usage, latency, and call counts for all LLM invocations directly within a query.

- Audio Transcription: We expanded our multimodal capabilities to include audio transcription, alongside our existing image support.

You can try it out via the community extension catalog: INSTALL flock FROM community;.

We'd love your feedback and contributions! Let us know what you think of the architecture or if you run into any edge cases.

Docs: https://dais-polymtl.github.io/flock/ Paper: https://dl.acm.org/doi/10.14778/3750601.3750685