# Build a knowledge assistant over your docs

> What should I use to build a RAG knowledge assistant over my documents?

Chat over your own documents — the most-built agent app there is. The open-source field offers a visual platform route and a code-first pipeline route; pick by who's going to maintain it.

A hand-picked shortlist from the Agent Search Engine index. Every pick is a real listing; the adoption, pricing, and activity figures are the index's live numbers. Browse on the web: https://agentsearchengine.app/use/build-rag-knowledge-assistant

## Recommended picks

### Visual platform: dify
- Open source · Platform · 153k★ · TypeScript · active 2 days ago
- **Setup:** Self-host — Docker
- **Why:** Visual builder, RAG pipelines, and agent workflows in one self-hostable platform — one of the most adopted projects in the index, strong when non-engineers need to iterate on prompts and flows.
- [listing](https://agentsearchengine.app/agents/dify) · [official](https://dify.ai)

### Code-first pipelines: haystack
- Open source · Framework · 26k★ · Python · Apache-2.0 · active 3 days ago
- **Setup:** Developer — Python
- **Why:** The mature Python framework for retrieval pipelines — full control over chunking, retrieval, and ranking when quality tuning matters more than a visual editor.
- [listing](https://agentsearchengine.app/agents/haystack) · [official](https://haystack.deepset.ai)

### Knowledge-base focus: FastGPT
- Open source · Platform · 29k★ · TypeScript · active 2 days ago
- **Setup:** Self-host — Docker
- **Why:** A knowledge-base-first platform with strong out-of-the-box document Q&A — the quickest of the three to a working assistant on a folder of docs.
- [listing](https://agentsearchengine.app/agents/fastgpt) · [official](https://fastgpt.io)

## What to watch for
RAG quality is a data problem before it's a model problem: bad chunking and stale documents produce confident wrong answers. Build a small eval set of real questions first, and re-index on a schedule — an assistant over last quarter's docs is worse than none.

---
Source: [Agent Search Engine](https://agentsearchengine.app) — more task-first guides: https://agentsearchengine.app/use