osaurus-ai / osaurus

项目摘要

Osaurus is a native macOS AI agent harness built in Swift that enables fully offline, local AI agent execution with persistent memory, autonomous tool use, and cryptographic identity. It supports multiple model providers including OpenAI, Anthropic, Ollama, and Apple Foundation Models, while exposing an OpenAI-compatible API and MCP server.

为什么值得关注

Osaurus addresses the growing concern of AI privacy and ownership by keeping the entire agent stack — memory, tools, identity, and execution — on the user's local machine. Its native Swift implementation leverages Apple Neural Engine for efficient inference, and its rapid release cadence (100 releases in 6 months) signals strong momentum in the local AI agent space.

优势

• Fully offline, privacy-first AI agent execution on macOS with Apple Neural Engine optimization
• Broad model provider compatibility (OpenAI, Anthropic, Ollama, Apple Foundation Models) with OpenAI-compatible API
• Persistent agent memory, autonomous execution, and MCP server support for extensibility
• Very active development with 100 releases in 6 months and 31 contributors

局限性

• No automated tests in the repository despite high release velocity
• No examples or Docker support, limiting onboarding and cross-platform accessibility
• macOS Apple Silicon only, excluding Windows and Linux users

使用场景

• Running privacy-sensitive AI agents locally without cloud dependency
• Building persistent agent workflows with memory across sessions across multiple LLM backends
• Serving as a local MCP server for AI tool integration on macOS