How do I enable persistent storage in Chroma DB for local machine learning projects?

All answers to this question.

To save your data, you need to switch from Client() to PersistentClient(path="./your_path"). This ensures that all collections you create are serialized to the specified directory using SQLite and Parquet files. I made this mistake early on in a document retrieval project and lost a whole day's worth of embedding generation. Once you set the path, you can just re-instantiate the client with the same path in your next session, and your collections will be right there. It’s incredibly useful for local development where you don’t want to keep paying for API-based re-embedding every time you restart your kernel.

   Answered 2025-05-15 by Kimberly Roberts

Does the persistent client handle automatic backups, or do we need to manually copy the directory if we want to migrate our vector store to a different cloud server?

   Answered 2025-05-16 by Jeffrey Davis

• Hey Jeffrey, it doesn't have a built-in "backup" button yet. Your best bet is to just zip the entire directory you specified in the persistent path. Since it's all local file-based storage, moving it to another server is as simple as unzipping it in the new environment and pointing your code to that folder.

     Commented 2025-05-17 by Brenda Miller

It’s much faster than Pinecone for local testing since there's zero network latency when the data is sitting on your SSD.

   Answered 2025-05-18 by Michael Garcia

• Totally agree, Michael. For datasets under 1 million vectors, the local performance is almost impossible to beat with a cloud-only provider.

     Commented 2025-05-19 by Kimberly Roberts