Cognita is an open-source framework designed to productionize Retrieval Augmented Generation (RAG) applications by providing modular, API-driven components and a no-code UI. It addresses the challenges of scaling RAG systems beyond prototyping, offering a structured approach for developers and enabling non-technical users to interact with and customize RAG pipelines.

How It Works

Cognita leverages libraries like Langchain and LlamaIndex under the hood but abstracts them into a modular, API-driven architecture. It separates concerns into distinct components: Data Sources, Metadata Store (Postgres), LLM Gateway (for unified model access), Vector DB (Qdrant, SingleStore supported), Indexing Jobs, and an API Server for question-answering. This design promotes code organization, scalability, and extensibility, allowing for easy customization of data loaders, embedders, parsers, and vector databases.

Quick Start & Requirements

• Local Deployment: Recommended via docker-compose --env-file compose.env up.
• Prerequisites: Docker and Docker Compose.
• Configuration: Requires models_config.yaml for model providers (e.g., OpenAI API key, Ollama, Infinity Server).
• Resources: Starts Postgres, Qdrant, backend, and frontend services. Additional profiles for Ollama and Infinity Server can be enabled.
• Docs: https://cognita.truefoundry.com

Highlighted Details

• Supports multiple document retrievers (Similarity Search, Query Decomposition, Reranking).
• Integrates with SOTA open-source embeddings and rerankers (e.g., Mixedbread-ai).
• Offers incremental indexing by default.
• Features a no-code UI for data ingestion and querying.
• Supports multimodal vision parsing.

Maintenance & Community

• Actively developed by TrueFoundry.
• Recent updates include Audio/Video parsers, Pydantic v2 migration, Model Gateway, and a Metadatastore powered by Prisma/Postgres.
• Contribution guide available.

Licensing & Compatibility

• The repository does not explicitly state a license in the provided README.

Limitations & Caveats

• Future developments list includes support for additional vector databases (Chroma, Weaviate) and RAG evaluation, suggesting current support is limited.
• Deployment with TrueFoundry requires registration and setup on their platform.