Repolode
Submit
arc53

arc53/DocsGPT

Self-hosted platform for building AI agents and assistants that answer questions from your own documents, with citations and any model.

Open-source alternative to:ChatGPT EnterpriseNotion AI

View on GitHubWebsite
Stars
18,330
Forks
2,217
Open issues
109
Last commit
2 hours ago

What is it?

DocsGPT turns documents, websites and connected apps into AI agents that answer with sources. Each agent has its own prompt, knowledge base, tools and model, and can be published as a chat widget, an OpenAI-compatible API endpoint, or an MCP server.

The platform includes a visual workflow builder for chaining agents, conditions and sandboxed code, along with a deep research mode for multi-step answers. Knowledge can be ingested from PDFs, Office files, web pages and audio, and kept in sync with services such as Google Drive, SharePoint, Confluence, GitHub and S3. Model support covers OpenAI, Anthropic, Google, Groq and OpenRouter, as well as local models served through Ollama, vLLM or other OpenAI-compatible servers.

It runs in your own environment via a one-line installer, Docker Compose, pip, Kubernetes or an air-gapped setup. Team features include SSO (OIDC and SCIM), role-based access control, usage quotas and OpenTelemetry observability.

Key features

  • 01Agents with their own prompt, knowledge, tools and model
  • 02Visual drag-and-drop workflow builder with conditions and sandboxed code
  • 03Ingests PDFs, Office files, web pages and audio, synced from services like GitHub and Google Drive
  • 04Answers include citations to source documents
  • 05Works with cloud models or local models via Ollama and vLLM
  • 06Exposes an OpenAI-compatible API, embeddable widget and MCP server

Use cases

  • Building a support chatbot that answers from a company handbook or product docs
  • Giving internal teams a private search assistant over wikis and connected drives
  • Embedding a documentation assistant into a website or product page
  • Running an AI assistant fully on-premises with local models

Quick start

  1. 1

    Run the official install script from the README (requires Docker)

  2. 2

    Start DocsGPT and choose who can access it and which model to use

    docsgpt up
  3. 3

    Open the web app in your browser

Things to know

  • Self-hosting runs a multi-service stack (API, worker, Postgres, Redis and a vector store), which takes setup effort.
  • The one-command installer pipes a remote script into a shell, so review it before running it.

Topics

#rag#llm#ai-agents#document-qa#self-hosted#semantic-search#chatbot#vector-database#agent-builder#agents

FAQ

What is DocsGPT?
Self-hosted platform for building AI agents and assistants that answer questions from your own documents, with citations and any model. DocsGPT turns documents, websites and connected apps into AI agents that answer with sources. Each agent has its own prompt, knowledge base, tools and model, and can be published as a chat widget, an OpenAI-compatible API endpoint, or an MCP server.
Is DocsGPT free to use? What license does it have?
DocsGPT is open source under the MIT license. Check the license terms before using it in commercial products.
Is DocsGPT still maintained?
Actively maintained. The last commit to DocsGPT was on Oct 21, 2026.
What can I use DocsGPT for?
Building a support chatbot that answers from a company handbook or product docs Giving internal teams a private search assistant over wikis and connected drives Embedding a documentation assistant into a website or product page Running an AI assistant fully on-premises with local models
How do I install DocsGPT?
Run the official install script from the README (requires Docker) Start DocsGPT and choose who can access it and which model to use: docsgpt up Open the web app in your browser
Where can I get DocsGPT?
The source code and installation instructions are on GitHub: https://github.com/arc53/DocsGPT

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects