cocoindex-io/cocoindex ? reverse-engineered prompt

Build me an open source toolkit for AI agents that keeps their knowledge fresh automatically. I want developers to point it at things like a codebase, docs, notes, inbox exports, Slack data, PDFs, or videos, then have it continuously turn that content into searchable context for an LLM app.

The important part is that it should not rebuild everything every time. It should notice what changed and only process the delta, so big collections stay fast and cheap. Make the main user experience a simple Python workflow where someone can describe sources, transformations, and outputs without writing a lot of glue code. It should support RAG style use cases, semantic search, and storing processed results in target systems that an agent can query later.

Please include clear examples, a quickstart, tests, and enough docs that a developer can get a working fresh context pipeline running quickly. Use a fast core where it matters, but keep the public API friendly and Python first.