arc53/DocsGPT

DocsGPT – Private AI for agents, assistants and enterprise search

What it is – DocsGPT is an open‑source platform that lets you turn collections of documents (PDFs, Word files, spreadsheets, web pages, audio recordings, etc.) into searchable knowledge bases that can be queried by large language models. It ships a Flask backend, a React/Vite web UI, and a set of ready‑made integrations (chat widgets, Discord/Telegram bots, API connectors). The system can run entirely on‑premises or in the cloud and works with any LLM you prefer – OpenAI, Anthropic, Google, or self‑hosted models via Ollama or llama_cpp.

Key capabilities

• Broad file support – PDFs, DOCX, CSV, XLSX, EPUB, Markdown, HTML, JSON, PPTX, images and audio (MP3/WAV/etc.).
• Speech workflow – Record voice, have it transcribed, and add the resulting text to the knowledge base.
• Web & data ingestion – Pull content from URLs, sitemaps, Reddit, GitHub, or run a crawler.
• Hallucination‑reduced answers – Responses are accompanied by citations that point back to the original source document.
• Actionable tooling – API keys tie a user’s settings, documents and model choice together, and the platform can call external APIs from within a chat (e.g., trigger a webhook, run a database query).
• Pre‑built integrations – HTML/React chat widgets, searchable UI, Discord & Telegram bots, and more.
• Model flexibility – Plug‑in OpenAI, Anthropic, Google models, or bring your own model (BYOM) via Ollama or llama_cpp.
• Enterprise‑ready deployment – Docker‑compose, Kubernetes manifests, RBAC, OIDC/SSO, observability via OpenTelemetry, and scalable PostgreSQL storage.

Typical use cases

• Building a company‑wide knowledge‑assistant that answers employee questions from internal docs, policies, and meeting recordings.
• Creating a research‑assistant that ingests academic papers, web articles, and code repositories, then provides source‑cited answers.
• Deploying a customer‑support chatbot that can look up product manuals, troubleshooting guides, and support tickets.
• Enabling voice‑driven assistants that understand spoken queries and can retrieve information from recorded meetings.

Getting started

1. Prerequisite – Install Docker.
2. Clone the repo and run the provided setup script (./setup.sh on macOS/Linux or setup.ps1 on Windows). The script walks you through five deployment options (public API, local model, cloud provider, etc.) and creates a .env file.
3. After setup, start the stack with Docker Compose and open http://localhost:5173 in a browser.
4. Use the UI to upload documents, configure an LLM provider, and begin chatting.

Architecture at a glance

• Backend – Flask app (docsgpt package) handling document ingestion, vector store indexing, and LLM request routing.
• Frontend – React UI built with Vite for a responsive chat and admin interface.
• Extensions – Optional widgets and integrations (e.g., Chatwoot, Discord bot).
• Deployment – Docker images, Kubernetes manifests, and scripts for local development.

License – MIT (permissive, commercial‑friendly).

Community & support – Public Discord, documentation site, quick‑start guide, and a “Lighthouse Program” for enterprises that want direct help and roadmap influence.

---

All details above are taken directly from the repository’s README.