Verified project record

osaurus-ai/osaurus

Osaurus is a native macOS application that provides on-device AI agents, layered persistent memory, autonomous execution, and cryptographic identity across interchangeable models. It targets users who need local ownership of their AI context but requires Apple Silicon hardware and macOS 15.5 or later.

Project overview

Project type: MCP · AI Agent · Model Runtime
Use cases: Knowledge Q&A · Coding & Development · Automation
Deployment: Refer to project documentation
License: MIT

Best for

Developers on macOS who require autonomous code execution and native tool integration via a model-agnostic local harness.
Users who need to keep AI context, identity, and memory on local hardware rather than external servers.

Key capabilities

Agents operate with custom prompts, memory, visual themes, and tools to execute autonomously, write todo lists, and verify summaries.
Three layers of memory comprising identity, pinned facts, and per-session episodes that distill conversations, score salience, and inject compact context slices.
An isolated Linux VM running Alpine via the Apple Containerization framework provides shell, Python, Node.js, and compilers without risk to the host.
An on-device classifier detects PII and custom patterns before sending data to cloud models, replacing detected information with stable placeholders and blocking sends upon leak detection.
Provides Secp256k1 cryptographic addresses for participants, a verifiable chain of trust from a master key, and portable scoped access keys.
Exposes Osaurus tools to any MCP-compatible client through a command-based stdio bridge.
Aggregates tools from URL-based remote MCP providers with one-tap connect and OAuth 2.1.
20+ native plugins including Mail, Calendar, Vision, Browser, and Git, with support for creating new Swift plugins using hot reload.

Limitations and risks

The application requires macOS 15.5 or later and Apple Silicon hardware.
The isolated Linux VM code sandbox specifically requires macOS 26+ (Tahoe).
The application includes anonymous usage analytics via Aptabase and crash reporting via Sentry. Crash reporting is opt-out for standard installs.

Getting started

Install using `brew install --cask osaurus` or by downloading the `.dmg` graphical installer. Launch the application from Spotlight or the CLI using `osaurus ui`.

Evidence and sources

GitHub project description: The native macOS harness for AI agents -- any model, persistent memory, autonomous execution, cryptographic identity. Built in Swift. Fully offline. Open source.
README: Others keep that layer on their servers. Osaurus keeps it on your machine.
README: Native Swift on Apple Silicon. No Electron. No compromises. MIT licensed.
Release: Osaurus 0.22.3
README: Agents are the core of Osaurus. Each one gets its own prompts, memory, and visual theme -- a research assistant, a coding partner, a file organizer, whatever you need.

Reviewed at: 2026-07-15T14:43:27Z