Bring your existing data
Map IDs, documents, metadata, embeddings, and media URIs to a LanceDB schema. Keep the same embedding model and dimensions when reusing vectors.
December 7, 2026: Chroma Cloud becomes read-only. Move write traffic before this date.
December 22, 2026: Reads end. Complete your export and migration before this date.
Keep your collections, embeddings, and retrieval quality in focus. Work with our engineers to map your data, validate the results, and move production traffic before Chroma Cloud shuts down.
Map IDs, documents, metadata, embeddings, and media URIs to a LanceDB schema. Keep the same embedding model and dimensions when reusing vectors.
Translate filters and ranking logic. Check relevance, row counts, updates, deletes, warm and cold latency, and concurrent traffic against your requirements.
Backfill data, reconcile writes, and test a rollback path. Agree on the deployment and operating plan before routing production traffic.
collection.get() omits embeddings by default. Request include=["embeddings", "documents", "metadatas", "uris"] and paginate with limit and offset. Record IDs are always returned. Get reference ↗chroma copy --from-cloud --to-local --all with your active Cloud profile and a local Chroma server (default: localhost:8000). This creates a local Chroma copy; importing it into LanceDB is a separate step. CLI copy reference ↗collection.search() cannot fall back unchanged to local or single-node Chroma. Preserve your filters and ranking logic, then translate and validate them on your target database. Search API availability ↗THE LANCEDB ENTERPRISE ADVANTAGE
ONE DATA FOUNDATION
Store images, video, audio, documents, embeddings, and metadata in open Lance tables. Give retrieval and data processing a shared foundation.
Explore LanceDB Enterprise ↗MANAGED FEATURE ENGINEERING
Create feature columns with Python UDFs, then run backfills through LanceDB Functions' managed distributed jobs. Move from notebook experiments to production data workflows.
Explore feature engineering ↗COMPUTE THAT FITS THE WORK
Separate durable object storage from query serving and background processing. Add capacity where your workload needs it.
Explore the architecture ↗Compare what your AI stack needs beyond retrieval: source data, feature pipelines, and the compute to process them. LanceDB Enterprise brings those capabilities together.
| What your AI stack needs | LanceDB Enterprise | Chroma Cloud (shutting down) | Chroma Open Source |
|---|---|---|---|
| Raw multimodal data | Raw media, embeddings, and typed metadata together in open Lance tables. | Documents, embeddings, and metadata in Chroma. Original images and other non-text media stay external, accessed through URIs and loaders. | Multimodal embeddings supported. Original images and other non-text media stay external. |
| Feature engineering | Python UDFs become feature columns, with backfills and managed distributed execution through LanceDB Functions. | Embedding functions support retrieval. Plan for broader preprocessing and feature pipelines alongside the database. | Embedding functions integrate with your application. Your team runs preprocessing and feature jobs. |
| Search | Vector and full-text search with reranking and SQL-style filters, alongside the source data in your tables. | Vector and hybrid search, rank fusion, metadata filters, and custom ranking through the Cloud Search API. | Vector search, metadata filters, and document-content filters. The Cloud Search API is not yet available locally. |
| Storage & scaling | Open Lance tables in object storage, with separate capacity for query serving, caching, and background work. | Distributed services with object storage and SSD caches. Compute and storage are separated. | Single-node capacity is bounded by the host; positioned for small to medium workloads. |
| Deployment | Private cloud and bring-your-own-cloud (BYOC) options for your Enterprise data platform. | Shutting down: read-only December 7, 2026; reads end December 22, 2026. Move to another deployment or database before these deadlines. | Embedded library or self-hosted single server. Your team operates the infrastructure. |
| Operating model | Managed query serving, caching, and table maintenance, plus feature engineering in the same platform. | Database operations managed by Chroma. Support, isolation, and SLA options depend on the plan. | Your team owns monitoring, backups, upgrades, and capacity planning. |
| Cost planning | Size compute for serving and feature workloads separately. Scope the full data workflow with LanceDB. | Plan fees plus metered writes, storage, queried data, and returned data. Enterprise pricing is custom. | Apache 2.0 software, plus the infrastructure and engineering time to operate it. |
FROM RAW DATA TO AI FEATURES
Retrieve a vector, inspect its source, and use that data for the next processing step. LanceDB keeps raw media, vectors, and metadata in a shared table model.
Enterprise overview ↗Derive new columns with Python UDFs and backfill them through managed jobs. Build enrichment workflows on the tables your application already uses.
Feature engineering ↗LanceDB’s architecture separates serving from background work. Keep durable tables in object storage and size compute for the work it performs.
Architecture ↗CrewAI consolidated a two-system memory architecture into LanceDB Open Source, reducing dependencies and bringing scoped retrieval into one table. Its experience shows how a unified data model can simplify an AI application.
Read the CrewAI case study ↗CodeRabbit uses LanceDB to retrieve context from code, past pull requests, and issue trackers. The team reports sub-second semantic search across millions of code interactions, with cloud and on-prem deployment flexibility.
Read the CodeRabbit case study ↗Yes. Chroma Cloud becomes read-only on December 7, 2026 and stops serving reads on December 22, 2026. This shutdown applies to the hosted service; open-source Chroma continues under Apache 2.0. If you are evaluating a replacement, LanceDB Enterprise combines retrieval with raw multimodal storage, open Lance tables, and managed feature engineering. Chroma’s official wind-down timeline ↗
Yes. Combine vector and full-text search with reranking and metadata filters. Chroma Cloud also offers hybrid search; LanceDB’s advantage is keeping retrieval alongside raw data and feature workflows. Chroma’s Cloud Search API is not currently available in local single-node deployments. Search API availability ↗
Existing vectors can be reused when the model, dimensions, distance metric, and preprocessing remain compatible. Map your collections into LanceDB tables, validate retrieval quality, and add new features as your application evolves.
Separate durable object storage from query serving, caching, and background work so you can size compute for each workload. Actual cost depends on data volume, traffic, and feature jobs. Our team can help evaluate your Chroma deployment and the pipelines around it.
Bring your collection sizes, embedding models, and search requirements. Work with a LanceDB engineer to scope your migration and production cutover.