Search operation is not implemented for local executor: the Chroma error and the fix

Chroma 1.5.9 answers Search operation is not implemented for local executor when you call the Search API on a single-node server. Chroma runs the Search API only in its distributed form, the one Chroma Cloud is built on. VecturaDB, a single-node server built from Chroma with the same API, runs the same code.

VecturaDB is an independent project, not affiliated with Chroma.

The error

The Search API is collection.search(), with Knn, Rrf, select and group_by. This call raises it:

results = collection.search(Search().rank(Knn(query=[0, 1, 0])).limit(2))

Fix: run the same code on VecturaDB

Start VecturaDB:

docker run -d --name vecturadb -p 127.0.0.1:8000:8000 -v vecturadb-data:/data ghcr.io/vecturadb/vecturadb:1.0.0-preview.8

Point your client at it. The rest of the code does not change:

client = chromadb.HttpClient(host="127.0.0.1", port=8000)

VecturaDB opens the data folder of Chroma 1.5.9 as it is. To keep your data, mount that folder instead of the vecturadb-data volume. Copy the folder first, and stop Chroma before you copy it.

Without a server, in Python: pip uninstall -y chromadb, then pip install VecturaDB. The package keeps the import name chromadb, and chromadb.PersistentClient runs the Search API.

The binary for Windows, Linux and macOS is in the releases. Examples with BM25 and hybrid search: Chroma BM25 search on your own server, Hybrid search on your own server.

Fix: dense search with query()

If you need only dense search with metadata filters, collection.query() runs on Chroma 1.5.9:

results = collection.query(query_embeddings=[[0, 1, 0]], n_results=2, where={"topic": {"$ne": "tech"}})

It has no sparse vectors, BM25 or RRF.

Fix: distributed Chroma on Kubernetes

Distributed Chroma runs the Search API on several nodes: see Migrate from Chroma Cloud to self-hosted distributed Chroma on Kubernetes.

Other errors of the single node

Chroma 1.5.9 refuses these calls on a single node too. VecturaDB runs them:

The full list: What VecturaDB has.