- Handling of titles in Pinecone database has been updated. Pinecone does not offer a simple way to retrieve metadata to get a list of titles. To overcome this, titles are encoded in record prefixes by ragcore. Titles can be case-sensitive, and can contain spaces now.
1.0.3
Added
- Pinecone uploads are now chunked to handle larger file sizes. The Pinecone recommendation is 100 records per batch [https://docs.pinecone.io/docs/upsert-data].
- Pinecone upload throughput was increased by using Pinecone's asynchronous upload option using `pool_threads`.
Removed
- Document titles have been removed from log outputs for increased privacy.
1.0.2
Added
- Pinecone database is now supported. To use it, select `pinecone` in the config file, add your Pinecone `base_url`, and done!
Changed
- Internally, a model is now used for the configuration of the app, replacing the dictionaries.
1.0.1
Fixed
- When trying to access a Chroma collection which does not exist, the collection is now automatically created.
1.0.0
Added
- New `query` response object `QueryResponse` contains information about the sources on which the response is based on. You can now extract the titles as well as the document chunks from the response. - The option to specify an identifier for a `user` with all app-level requests, to support multiple users. - A `get_titles()` method on the app level to get a string list of all titles owned by a user in the database.
Changed
- Changed the return type of the query method to return a `QueryResponse` object instead of an optional string. To get the response string as before, call `response.content` on the query response.