RAG framework with a UI, an API, and a retirement notice

What it does Cognita is a structural organizer—essentially well-architected glue—for LangChain and LlamaIndex RAG code. It forces notebook experiments into a modular, API-driven layout with a FastAPI backend, a Postgres-backed metadata store, and a no-code frontend for document upload and Q&A. The system supports incremental indexing, vector databases like Qdrant and SingleStore, and an LLM gateway that normalizes calls across OpenAI, Ollama, and TrueFoundry providers.

The interesting bit Instead of replacing your favorite abstractions, it treats them as implementation details and wraps them in a rigid architecture where data sources, indexing jobs, query controllers, and a model gateway all communicate through APIs. It also supports multimodal parsing—audio, video, and vision—extending the standard text-only RAG pipeline.

Key highlights

• Ships with a no-code UI for document upload and Q&A, plus a fully API-driven backend for integration.
• Supports incremental indexing that tracks already-ingested documents to avoid redundant compute.
• Uses an LLM Gateway to proxy embedding and generation requests across OpenAI, Ollama, and TrueFoundry providers.
• Vector database support includes Qdrant and SingleStore.
• Includes audio and video parsers alongside standard document chunking.

Caveats

• The project is no longer actively maintained, as stated at the top of the README.
• Vector database support is currently limited to Qdrant and SingleStore.
• The README implies smoother deployment when using TrueFoundry’s own platform components.

Verdict Good for teams who want a reference architecture for production RAG and are comfortable with fork-and-own maintenance. Skip it if you need active upstream support or a framework that fully replaces LangChain and LlamaIndex.