PrivateGPT Release 1.0.0: Messaging API, Agentic RAG with References, and MCP Connector
The major release of PrivateGPT 1.0.0 brings standard messaging APIs, Agentic RAG with references, built-in and custom tools, MCP connectors and code execution, and a switch to connecting to external inference services instead of running models yourself.
PrivateGPT 1.0.0: From "local running model" to "private agent platform"
PrivateGPT releases 1.0.0 major version. This is not an ordinary iteration - Li has redefined the role of PrivateGPT: it no longer runs the model itself, but hands over reasoning to external services, and focuses on messaging APIs, Agentic RAG with references, and tool orchestration.
The most critical turning point
Prior to 1.0, PrivateGPT ran models by itself; starting from 1.0, it instead connects to external OpenAI-compatible inference services (such as Ollama, llama.cpp, vLLM). Inference and orchestration are thus decoupled - you are free to choose a self-hosted backend and individually weigh cost and privacy, while PrivateGPT is responsible for the upper-layer Agent logic.
1.0 What can be done
| Capabilities | Description |
|---|---|
| Message API | Standard messages API |
| Ingestion | Document and artifact ingestion |
| Search | Search with citations and Agentic RAG |
| Tools | Built-in tools (mapped to Claude API) + custom tools |
| Connections | MCP Connector |
| Data | Database/CSV structured access, Web search and extraction |
| Execution | Code execution, Token counting, embedding and orchestration |
The standard messaging API plus tool calls and MCP connectors make PrivateGPT essentially closer to a general private agent platform.
Breaking changes and project trends
1.0 is not forward compatible with previous versions and is a destructive major version. The official documentation provides migration guidelines. It is worth mentioning that project governance is also changing: Zylon has closed its private branch and will continue to develop based on this open source repository.
Landing
- Before upgrading, read the official migration guide and evaluate the incompatibilities with existing integrations one by one.
- Prepare an OpenAI compatible inference service (Ollama/llama.cpp/vLLM) as the backend.
- For the complete capability list and migration details, please refer to the official release notes and documents.
Reviews