7 Chroma Alternatives as Chroma Cloud Shuts Down

The right Chroma alternatives depend on which Chroma you were using. If you were on Chroma Cloud, move to Qdrant Cloud, Pinecone or Zilliz Cloud, or run open-source Chroma yourself. If you want an embedded store with no server, pick LanceDB or sqlite-vec. If you want one codebase from laptop to cluster, pick Qdrant or Milvus Lite. If your data is already in Postgres, use pgvector.

There's a deadline behind this search. Chroma Cloud stops accepting writes on December 7, 2026 and stops serving reads on December 22, 2026. Open-source Chroma isn't going anywhere, so plenty of readers don't need to leave at all. This guide covers what changed, who actually has to move, and where each kind of Chroma user fits best.

Quick Comparison

What Changed at Chroma

On October 23, 2026, Chroma announced it had signed a definitive agreement to join Replit. The same post says Chroma is winding down its existing cloud service over the following 60 days.

Chroma's Cloud wind-down timeline gives two dates:

• December 7, 2026: writes disabled. Chroma Cloud becomes read-only.
• December 22, 2026: reads disabled. Chroma Cloud stops serving read requests.

The timeline tells customers to stop or redirect write workloads and move to "another Chroma deployment or database" before the first date. Before the second, finish exporting data, check that nothing still calls Chroma Cloud endpoints, and remove or rotate stored Chroma Cloud credentials.

The open-source project continues. The timeline page says the change applies only to the hosted service and that open-source ChromaDB remains available under the Apache 2.0 licence. The announcement adds three things: the Chroma team will keep working on ChromaDB at Replit, Chroma has committed additional code from its private repositories to the Apache 2.0 repo, and Replit will keep contributing to the project.

One practical warning. Parts of Chroma's README and docs still invite you to sign up for Chroma Cloud. The wind-down timeline is the page to trust.

Where Chroma is still the right call

If you never used Chroma Cloud, nothing forces you to move. The Chroma repo is Apache-2.0, and its README still describes the same simple loop: create a client, make a collection, add documents, query. Chroma tokenizes, embeds and indexes for you, or accepts your own embeddings. It runs in memory for prototyping, persists to disk when you ask, and has a client-server mode through chroma run.

Stay on Chroma if:

• Your app uses the in-memory or persistent client on one machine.
• You already run a Chroma server and are happy operating it.
• You value the built-in embedding step that gets a prototype working with no API key.

Leave if you depended on someone else running the database for you, or if the ownership change makes your team want a vendor whose main business is the database. For a head-to-head on the managed side, see Pinecone vs Chroma.

Where Chroma Cloud Users Can Go

Self-hosted Chroma: the smallest code change

Best for: teams that want to keep their query code and are willing to run a server.

Chroma's Docker guide runs the official chromadb/chroma image with a mounted data volume and port 8000 exposed. Your app then connects over HTTP instead of pointing at Chroma Cloud.

The trade-off: backups, upgrades, scaling and security are now your job. Also check that every feature you used on Cloud exists in the open-source build before you plan a like-for-like move. Test your real queries against a self-hosted instance first.

Qdrant: open source with a managed cloud

Best for: teams that want a hosted service today and the option to self-host later.

Qdrant is an Apache-2.0 vector similarity search engine and vector database written in Rust. It combines multiple vectors in one query and merges results with Reciprocal Rank Fusion or Distribution-Based Score Fusion, which covers hybrid search.

For former Chroma Cloud users, the appeal is the range of ways to run it. Qdrant's pricing page lists a free single-node cluster for testing and prototypes, usage-based managed clusters billed on compute, memory and storage, a hybrid cloud that runs managed clusters on your own infrastructure, and a private cloud for isolated deployments.

Qdrant also suits local work. Its Python client has a local mode: QdrantClient(":memory:") or QdrantClient(path="path/to/db") runs the same code with no server. The README calls this useful for development, prototyping and testing, and says to switch to server mode when you need to scale.

The trade-off: local mode is a development tool, not a production embedded database. And Chroma's "pass in text, get results" default takes a little more setup here, through the optional FastEmbed extra.

Pinecone: nothing to run

Best for: teams that picked Chroma Cloud because they didn't want to operate anything.

Pinecone describes its database as a search engine for AI agents and applications. Integrated embedding lets you upsert and query with text while Pinecone generates the vectors on its servers, which is close to how Chroma handles embedding for you. One index can serve semantic, full-text and hybrid search, and there's a bring-your-own-cloud option.

The trade-off: there's no open-source version. You can't prototype offline, and you can't take the engine with you if you leave. Our Pinecone alternatives guide covers that side.

Zilliz Cloud and turbopuffer

Two more hosted options fit specific cases.

Zilliz Cloud is the managed service from the main contributor to Milvus. Its pricing page lists serverless and dedicated clusters plus a bring-your-own-cloud deployment. It makes the most sense if you expect very large collections, or if you like the Milvus Lite path described below.

turbopuffer describes itself as vector and full-text search built on object storage, with a memory and SSD cache in front. Billing is usage-based across storage, writes and queries, with a monthly minimum on its entry plan. It fits apps with lots of data that is queried only occasionally. It's hosted only, so there's no local mode.

Embedded and Local Chroma Alternatives

Many people use Chroma for one reason: pip install, two lines of Python, and you have a vector store. If that's you, these are the closest matches.

LanceDB: the nearest embedded match

Best for: notebooks, desktop apps and data that mixes text with images or video.

LanceDB is Apache-2.0 and built on the Lance columnar format. LanceDB Embedded runs inside your application's process, with data on local disk or in object storage. The README lists vector similarity search, full-text search and SQL, automatic data versioning, and SDKs for Python, Node.js/TypeScript and Rust. A commercial product, LanceDB Enterprise, handles distributed production-scale search.

The trade-off: embedded means your app handles concurrency and serving. And you'll bring your own embedding step or wire one in, where Chroma has a default.

sqlite-vec: vectors inside SQLite

Best for: small local apps, browser and mobile projects, and anyone who already ships a SQLite file.

sqlite-vec is a vector search extension for SQLite, dual-licensed under Apache-2.0 and MIT. Its README calls it extremely small and "fast enough". It's written in pure C with no dependencies. It stores float, int8 and binary vectors in vec0 virtual tables, and runs on Linux, macOS and Windows, in the browser through WASM, and on devices like Raspberry Pis.

Your vectors live in the same file as the rest of your app data, and you query them with SQL.

The trade-off: the README says plainly that sqlite-vec is pre-v1, so expect breaking changes. It's a search extension, not a database with collections, embedding functions or a server. You write more of the plumbing.

Milvus Lite: local now, cluster later

Best for: prototypes that might grow far beyond one machine.

Milvus Lite is an Apache-2.0 lightweight version of Milvus that exposes the PyMilvus API through a local file, aimed at prototyping and small local workloads such as notebooks and tests. When you outgrow it, the README describes exporting a collection with milvus-lite dump and importing it into self-hosted Milvus or Zilliz Cloud.

The trade-off: the README lists real limits. Only one process can own a database directory at a time, authentication and TLS aren't supported, and some vector types and index types are missing. Treat it as a development environment for Milvus, not a place to run production.

pgvector: no new database at all

Best for: apps that already store their data in Postgres.

pgvector is open-source vector similarity search for Postgres. It supports exact and approximate nearest-neighbour search, with HNSW and IVFFlat indexes. For hybrid search, the README suggests combining it with Postgres full-text search and merging results with Reciprocal Rank Fusion or a cross-encoder.

The trade-off: it isn't embedded, because you need a running Postgres. And the standard vector type can only be indexed up to 2,000 dimensions, so check your embedding model's output size first.

How to Choose a Chroma Alternative

If you're migrating off Chroma Cloud, work backwards from the dates:

1. Export early. After December 7 you can read but not write. After December 22 you can't read. Pull your documents, metadata and embeddings out, and validate the export while the data is still reachable.
2. Keep the embedding model the same. If Chroma's default embedding function created your vectors, either move the vectors as they are or re-embed everything with one model. Mixing models breaks similarity scores.
3. Re-test filters and keyword search. Each engine has its own filter syntax and its own way of scoring keyword matches. Run your real queries before you switch traffic.
4. Clean up credentials. Chroma's timeline asks you to remove or rotate Chroma Cloud keys stored in your apps and infrastructure.

We haven't quoted benchmarks. Vendor numbers are measured on the vendor's terms, and a test on your own data tells you more. For a wider view of the open-source field, our Weaviate alternatives guide compares more engines, and the explainer on what RAG is covers when a vector store earns its place at all.

Where Taku Fits

Taku isn't a vector database, and it doesn't replace Chroma or anything on this list. It helps with a different problem: you've seen someone share a working RAG app or document-search workflow built on one of these stores, and you can't get it running on your own machine.

That's where Taku comes in. Taku Apps and Sites, built by the best AI minds. Find one, run it, and keep it in your Mac Dock. Taku is in Beta, and the Mac app is available now.

FAQ

Is Chroma shutting down?

No. Chroma Cloud, the hosted service, is shutting down. It stops accepting writes on December 7, 2026 and stops serving reads on December 22, 2026. Open-source ChromaDB remains available under the Apache 2.0 licence, and Chroma says its team will keep working on it at Replit.

What is the best Chroma Cloud alternative?

It depends on how much you want to run. Self-hosted Chroma needs the least code change. Qdrant gives you a managed cloud with an open-source engine behind it. Pinecone is the most hands-off.

What is the best open source Chroma alternative for local use?

LanceDB is the closest match for an embedded, in-process store. sqlite-vec fits if your app already uses SQLite. Qdrant's Python client local mode and Milvus Lite both work well for development when you expect to move to a server later.

Is ChromaDB still free?

Yes. The open-source database is Apache-2.0, so you can run it yourself, commercially, with no licence fee. You pay for the machine it runs on and the time to operate it.

Do I need to migrate if I only use Chroma locally?

No. The wind-down applies only to the Chroma Cloud service. Local and self-hosted Chroma keep working.

Can I move my Chroma embeddings to another database?

Yes, as long as you export the vectors with their documents and metadata. Keep using the same embedding model for new data, or re-embed everything, so old and new vectors stay comparable.