PrivateGPT 1.0 is out 🚀

[imartinez]

In 2023 we built the first offline RAG implementation. It hit top project on GitHub. Then we forked our own repo and went heads down for two years.

Here is what we were building.

PrivateGPT 1.0 is a full local AI application backend. It sits above any OpenAI-compatible inference server (Ollama, vLLM, llama.cpp, LM Studio) and gives you everything you need to build real AI products on your own hardware:

• Claude-compatible messages API with streaming and multi-turn
• Agentic RAG with citations and smart retrieval
• Tool use, custom tools, MCP connectors
• Database querying and CSV analysis
• Code execution, web search, web extraction
• Structured outputs, token counting, embeddings
• Async workflows

The architecture is simple:

Your app / agent / workflow / UI
|
PrivateGPT 1.0 API
|
Your inference server (Ollama, vLLM, etc.)

The part that surprises most people: because PrivateGPT implements the full Claude API spec, Anthropic's first-party tools run natively on top of it. Claude Code, Cowork, the Word, Excel, and PowerPoint add-ins, n8n, OpenCode. All of them, fully on-premise, zero data leaving your environment. No other local inference project does this today.

👀 One more thing:
Zylon.ai, our commercial product, now runs on this open source codebase.
We closed our private fork.
We build in public from here on.

Two years of enterprise deployments in defense, banking, healthcare, and government. All of the learnings back in the open.

Can’t wait to build together.
We are glad to be back.

Docs: https://docs.privategpt.dev/
Blog post: https://www.zylon.ai/resources/blog/introducing-privategpt-1-0-the-open-source-application-backend-for-private-ai