Cognita for RAG

Cognita is TrueFoundry’s open-source modular RAG framework (Apache-2.0) that wraps LangChain and LlamaIndex components behind API-driven loaders, parsers, embedders, rerankers, vector-database adapters, and a web UI. As of July 2026, the GitHub repository is archived (13 March 2026) and marked no longer actively maintained, with about 4,400 stars in the live capture. This page states what Cognita shipped for production RAG and why the archive status matters for new builds.

What is Cognita, and what part of a RAG system does it own?

Cognita organizes production RAG into modular services and a UI instead of notebook glue code (TrueFoundry GitHub README, 2026). It targets the gap between LangChain/LlamaIndex prototyping and production needs: chunking-and-embedding jobs abstracted from query APIs, embedding and LLM calls hosted as services, and vector databases deployed outside in-memory tests.

Cognita uses LangChain and LlamaIndex under the hood while adding registration hooks for loaders, parsers, embedders, rerankers, and vector DB adapters. That means it standardizes how documents are parsed, chunked, embedded, stored, and retrieved — but parser quality, chunk policy, and embedding choice still set what retrieval can find. Ingestion mechanics sit at document ingestion; the broader framework cluster is at RAG frameworks.

How does Cognita structure a modular RAG pipeline?

Cognita breaks RAG into seven swappable modules registered in Python and exposed through a FastAPI backend plus frontend (TrueFoundry blog, April 2026; GitHub architecture section, 2026):

• Data loaders — local directories, web URLs, GitHub repos, and TrueFoundry artifacts in the shipped set.
• Parsers — normalize PDF, Markdown, and text (plus audio and video parsers added in the 2024 changelog).
• Embedders — OpenAI, Cohere, mixedbread-ai, Infinity Server, Ollama, and TrueFoundry embedding paths per the README changelog.
• Rerankers — mixedbread-ai reranker support documented in the April 2026 blog post.
• Vector DBs — adapter layer with shipped implementations for Qdrant and SingleStore.
• Metadata store — Postgres with Prisma for collections and data sources (2024 changelog).
• Query controllers — FastAPI QnA endpoints that orchestrate retrieval and generation.

The recommended local stack uses Docker Compose to start Postgres metadata, a Qdrant server, the backend on http://localhost:8000, and the frontend on http://localhost:5001 (GitHub quickstart, 2026). Step-by-step commands belong on building a RAG pipeline; this section only names the shape.

Which vector stores and model providers does Cognita support?

Cognita’s shipped vector-database adapters, in the archived codebase, cover Qdrant and SingleStore (GitHub README vector DB section, 2026). TrueFoundry’s product page also lists Weaviate, Qdrant, SingleStore, Chroma, and others as integration categories in the broader ecosystem — but the README’s future-development list still treated Weaviate and Chroma as additions, not finished adapters, at archive time.

On models, the README changelog documents OpenAI and local paths through Ollama and Infinity Server for embeddings and reranking, plus mixedbread-ai embedder and reranker support called out in the April 2026 blog post. The real limit is operational: you choose and configure whichever backend you plug in, and adding a new vector store means implementing the BaseVectorDB interface and registering it. Choosing among Weaviate, Qdrant, Pinecone, and the rest is a separate decision at which vector database should you use; embedding trade-offs sit at embeddings.

What are Cognita’s capabilities and limits, side by side?

Cognita was strongest as a modular production template with a UI and shared component library across teams. Its main limit today is maintenance status: the repository is frozen.

The archive row is the one that changes the decision today. A team evaluating Cognita in 2024 for a modular internal RAG factory got a different risk profile than a team starting in July 2026. How reranking changes which chunks survive to the prompt is covered at reranking, not duplicated here.

When should you choose Cognita over LangChain, LlamaIndex, or Haystack?

Cognita made sense when you wanted a shared modular RAG factory — reusable components, a UI for non-technical upload and QnA, incremental indexing, and API-first integration across multiple internal teams (GitHub advantages list, 2026). Against LangChain or LlamaIndex alone, Cognita added production structure and a UI on top of those libraries rather than replacing them.

Against Haystack, Cognita was closer to a batteries-included TrueFoundry product template with a fixed module layout, while Haystack stays a general Python orchestration framework with typed pipelines. Do not choose Cognita for a new greenfield production build in July 2026 unless you explicitly accept maintaining a fork of archived code — the GitHub banner states the project is no longer actively maintained. For maintained alternatives, compare LangChain, LlamaIndex, and Haystack on choosing a RAG framework, with sibling profiles at LangChain for RAG, LlamaIndex for RAG, and Haystack for RAG.

What are the most common Cognita questions?

These answers reflect the July 2026 query harvest and the live GitHub capture, including the archive status that most listicles omit.