osaurus-ai/osaurus

Own your AI. The native macOS harness for AI agents -- any model, persistent memory, autonomous execution, cryptographic identity. Built in Swift. Fully offline. Open source.

What it solves

Osaurus is a local AI harness for macOS that allows users to own their AI identity, memory, and tools. It prevents the lock-in of personal context and data on cloud servers by keeping the "layer around the model"—including agents, memory, and identity—entirely on the user's machine, while still allowing flexible connection to either local or cloud-based LLMs.

How it works

Built natively in Swift for Apple Silicon, Osaurus acts as an intermediary between the user and various AI models. It manages a system of specialized agents, each with its own prompts and memory. These agents can execute code in an isolated Linux VM sandbox (or a macOS Seatbelt sandbox on older versions) to ensure system safety. It uses a three-layer memory system (identity, pinned facts, and session episodes) to maintain continuity without bloating context windows. For privacy, it includes an on-device classifier to scrub sensitive PII before data is sent to cloud providers.

Who it’s for

Mac users who want a private, offline-capable AI assistant system that can execute code, browse the web, and integrate with their local files and apps without sacrificing privacy or relying on a cloud-only ecosystem.

Highlights

  • Local-First Architecture: Fully offline operation with local models (via MLX) or optional cloud connectivity.
  • Secure Sandboxing: Code execution in an isolated Linux VM powered by Apple's Containerization framework.
  • Agentic Capabilities: Support for subagents, autonomous loops with todo lists, and a built-in Orchestrator for delegation.
  • Privacy Filter: On-device PII detection and redaction for cloud-model requests.
  • MCP Integration: Acts as both an MCP server and client, aggregating tools from remote providers.
  • Cryptographic Identity: End-to-end encrypted communication between agents using secp256k1 addresses.
  • Native Performance: Written in Swift for Apple Silicon, avoiding Electron for better efficiency.

Related

  • Project
    Sentient-OS-Labs/sentient-os

    An on-device AI operating system for macOS that builds a private, local knowledge base of your digital life to power proactive agents and computer-use automation.

  • Project
    webcoyote/sandvault

    A macOS tool that runs AI agents and shell commands in a sandboxed user account to provide isolation without the overhead of virtual machines.

  • Project
    ThinkInAIXYZ/clawdhome

    A multi-agent security console for macOS that runs isolated AI agent instances in separate macOS user accounts to ensure data privacy and system security.

  • Project
    open-software-network/os-clovy

    A private AI workspace for macOS that provides chat, dictation, and meeting notes while keeping data local and routing inference through a verifiable, privacy-preserving API.

  • Project
    andrew-waters/orchard

    A native macOS application for managing Apple containers, Linux VMs, and local Kubernetes clusters, featuring specialized sandboxes for running local AI agents.