Skip to content

R2R V3
Compare
Choose a tag to compare
@emrgnt-cmplxty emrgnt-cmplxty released this 23 Aug 23:33
· 802 commits to main since this release
v0.3.0
00d6c19
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

We are very excited to introduce R2R V3. We have been working very intensely as a team over this past month to comprehensively address all the feedback we received after recent explosive growth in R2R usage.

In addition to modifying 420 files and tens of thousands of LOC in the core repository, we have refactored the documentation here - https://r2r-docs.sciphi.ai/introduction.

NOTE - This release has many breaking changes w.r.t to R2R v2 and so we are working on a bespoke migration script to help those already in production.

Highlights

  • We've added full GraphRAG support as per the Microsoft GraphRAG paper. This implementation is built alongside our integration with Neo4j inside R2R so that you can build a non-ephemeral user-facing system. The system supports dynamic tuning of a number of query parameters so that you can optimize it for your given application, you can read more here!
  • Added full group permissions over vector search, meaning you can create unique dynamic groupings of documents which are permissioned at the user level. Support is included for granular actions like create_group, update_group, add_user_to_group, add_document_to_group, add_user_to_group, and more. Get started with the cookbook here.
  • We've doubled down on hybrid search by improving the full-text search component and its integration with vector search. The new system employs advanced techniques including text stemming, synonym expansion, phrase matching, and partial matching, all combined with a sophisticated ranking algorithm. Additionally, we've implemented runtime configurability through the VectorSearchSettings object. You can read more here.
  • We've implemented a powerful RAG Agent that enables developers to add custom tools and behaviors tailored to specific use cases. This allows R2R developers to trivially go from RAG to Agentic RAG, allowing 1-click construction of full chat intelligence. You can read more here.
  • Bug fixes and new focus on production R2R has matured significantly over the last quarter and has now answered over 1 million questions in production. As the system matures we are focusing more on supporting developers in production and carrying out our active development in a way that is minimally disruptive. This new release polishes a number of features that were rough / experimental in V2, while removing some features that were not seeing enough use, such as continuous evaluation.
  • Slim CLI and Python SDK to reduce bloat required to launch / interact with R2R through docker.

What's Changed

New Contributors

Full Changelog: v0.2.85...v3.0.0*

Contributors

manuelciosici, tkellogg, and 5 other contributors
Assets 2
Loading