PrivateGPT launches a local AI application API layer that follows the Claude API spec

PrivateGPT 1.0 is rebuilt as an API layer for private AI applications. It doesn't run models itself; instead, via OPENAI_API_BASE, it connects to any OpenAI-compatible inference server (such as Ollama, llama.cpp, vLLM) implementing /v1/chat/completions and /v1/models. It offers a standard messages API (with streaming, async, token counting), file and artifact ingestion, retrieval with citations and agentic RAG, built-in tools aligned with the Claude API (web search, web scraping, code execution), custom tools and MCP connectors, plus structured access to databases and CSVs; the built-in /ui workbench is for demo and testing only. Its capability comparison with the Claude API shows Messages, streaming, token counting, file ingestion, retrieval with citations, embeddings, tool use, tools in streaming, built-in web search, MCP, structured output, extended thinking, and token auth are all supported; Skills is only basic, while prompt caching and OAuth/org management are not yet supported. For developers and ops teams building private AI in regulated environments, this means they can keep their existing local inference services and cut the cost of connecting upper-layer apps to near that of calling the Claude API; the integrations listed in the docs, such as Claude Code, Claude Desktop, OpenCode, n8n, and Microsoft 365 add-ins, also let these clients use local models directly as their backend.