The open source project CocoIndex just released V1 with a complete engine rewrite. The tool turns data sources—codebases, documents, meeting notes, PDFs—into structured context for AI agents. The killer feature is incremental processing: if you edit a paragraph in a document, only that section gets re-indexed; everything else is skipped. Delete the source file, and the corresponding database records vanish automatically.
V1 brings two major changes for users. First, the project’s own configuration language is gone. You now write the entire data pipeline in plain Python functions, and you can set breakpoints and debug them normally. The engine automatically tracks each function’s inputs and code changes in the background, only re-running what actually changed. Second, the hard dependency on Postgres is gone. After pip install cocoindex, the tool runs immediately—no database setup required. The core engine is still written in Rust, and it’s released under the Apache 2.0 license. That’s a similar licensing choice to Tencent Cloud’s recent open-source project Cube Sandbox, also Apache 2.0.
Data can be exported to more than a dozen targets, including Postgres, LanceDB, Neo4j, SQLite, Kafka, and S3. The GitHub repo comes with over 20 ready-made examples covering text vectorization, PDF conversion, code indexing, and turning conversations into knowledge graphs.