zylon-ai/private-gpt

PrivateGPT wraps a complete private AI stack — RAG, skills, tools, MCP integration, text-to-SQL — into a single deployable API that speaks OpenAI-compatible interfaces. The pitch is straightforward: any model server you already run locally becomes a full-featured AI backend without touching a cloud API. What separates it from generic local-model wrappers is the breadth of the included primitives. Text-to-SQL alone is a feature most teams build as a weekend project and then maintain forever; having it in the same package as RAG and tool-calling is a real time save. The honest reservation is that local inference performance is on the team to sort out — PrivateGPT does not solve GPU provisioning, and at document scale the RAG pipeline quality depends heavily on the embedding model the team chooses. Worth evaluating before rolling a bespoke solution. -> Best for: SaaS team of 2-5 with compliance or data-residency requirements shipping internal AI tooling.