From Pinecone: Inactive-index policy for Auto-GPT indexes

[gkogan]

At the beginning of each session, Auto-GPT creates an index inside the user’s Pinecone account and loads it with a small number of embeddings. At the end of the session, it deletes the data but inadvertently leaves the index running. As a result, there are many thousands of accounts with empty indexes that were unintentionally left running.

While we’re committed to investing in infrastructure to support a rapidly growing number of developers exploring vector databases on our free plan, keeping discarded indexes running is not beneficial to anyone.

As a result, we took the following actions:

• We voiced our support to @Torantulino and the community for Implement Local Cache and Redis Memory backend #372.
• We are actively contacting customers on paid plans who are using Auto-GPT to ensure they understand their usage and expected costs. If you recently upgraded your Pinecone account and use Auto-GPT, please check your usage report (Organization > Usage in the Pinecone console) for unexpected spend. If you need assistance with billing, please contact support.
• We began archiving indexes created by Auto-GPT after 24 hours of inactivity. This is an exception to our standard policy of archiving indexes after seven days of inactivity. By design, Auto-GPT expects indexes it creates not to exist once the application completes, so we do not expect this change to impact customers. As usual with our archiving process, the data in the index is safely stored in a collection that can be restored to a new index in a few minutes.

We’re amazed by the Auto-GPT project and the community that rapidly formed around it. It’s a testament to the ingenuity of the creator, the spirit of the community, the magic of GPT, and the power and ease of use of Pinecone. The future for Auto-GPT is bright!

— Greg