Happy to answer questions or go deeper on anything. A few notes that might help set expectations:
- Provider support is solid for Chroma, Qdrant, and Postgres/pgvector. Pinecone works for most read workflows but isn’t full parity yet.
- The tool is designed to be “forensic first”: surfacing metadata, provenance, and mismatches rather than hiding them behind abstractions.
- Visualization is intentionally minimal right now; clustering overlays and model-to-model comparison are in progress.
- I’m especially interested in how people think about creation workflows (re-embedding, mixed-model collections, reproducibility, etc.) since teams handle this very differently.
Just to set expectations: it’s basically been me running it so far. PyPI has been getting a lot of traffic, but real-world usage is still very small. I’m really curious how it behaves with other people’s data and workflows — that feedback is incredibly helpful at this stage.
If you hit anything confusing, missing, or surprising, I’d love to hear it. Real-world debugging stories are gold for shaping the next set of features.
Happy to answer questions or go deeper on anything. A few notes that might help set expectations:
- Provider support is solid for Chroma, Qdrant, and Postgres/pgvector. Pinecone works for most read workflows but isn’t full parity yet.
- The tool is designed to be “forensic first”: surfacing metadata, provenance, and mismatches rather than hiding them behind abstractions.
- Visualization is intentionally minimal right now; clustering overlays and model-to-model comparison are in progress.
- I’m especially interested in how people think about creation workflows (re-embedding, mixed-model collections, reproducibility, etc.) since teams handle this very differently.
Just to set expectations: it’s basically been me running it so far. PyPI has been getting a lot of traffic, but real-world usage is still very small. I’m really curious how it behaves with other people’s data and workflows — that feedback is incredibly helpful at this stage.
If you hit anything confusing, missing, or surprising, I’d love to hear it. Real-world debugging stories are gold for shaping the next set of features.