DocsGPT
Self-hosted AI document search and agents with source citations. Run local models through Ollama, vLLM or llama.cpp, including fully offline deployments.
DocsGPT is an MIT-licensed, open-source platform for teams that want AI search, assistants and agents over their own documents. It can run on your servers with local models, including fully air-gapped deployments where documents and questions stay inside your network. Answers include the source title and page number so readers can check the evidence.
It reads PDFs, Office files, web content, images and audio. OCR makes scanned PDFs searchable, while audio transcription lets teams search meeting recordings and voice notes. Connectors bring in documents from Google Drive, SharePoint, Confluence and S3; it can also ingest websites and sitemaps.
Beyond answering questions, agents can conduct research and call tools through MCP or connected APIs. A visual workflow builder combines agents with conditions and code for document jobs, with execution history and human approval gates. Teams can embed chat in a website or connect assistants to Discord and Telegram.
Deployment determines where processing happens. Docker-based self-hosting works on macOS, Windows and Linux, with local inference through Ollama, vLLM or llama.cpp. Cloud model options include OpenAI, Google and Anthropic. A hybrid deployment keeps the platform on premises but sends model requests to a cloud provider; Arc53 also operates hosted and dedicated managed services.
For organisations with access and audit requirements, DocsGPT supports single sign-on, user provisioning and role-based permissions. Its append-only audit log records questions, answers and actions, and guardrails can flag, redact or block inputs and outputs.
