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.
Last verified:
What is Osaurus?
Osaurus is a native macOS app for AI agents that runs entirely on your Mac, giving you complete ownership of your AI experience. It sits between you and any AI model—local or cloud—and provides persistent memory, autonomous execution, and the ability to run real code safely. The app is built in native Swift for Apple Silicon (M1/M2/M3+), requires no Electron, and is MIT-licensed open source. Nothing leaves your device unless you explicitly choose a cloud provider.
Key features include custom AI agents that remember your conversations, execute tools autonomously, and evolve with you; a chat overlay accessible anywhere via ⌘; persistent memory that distills past conversations into compact facts; working folders that give agents safe file/search/git tools scoped to a directory; a Linux Sandbox (macOS 26+) for running code in an isolated VM; voice input with wake-word detection; schedules and folder watchers for automation; and cryptographic identity for you and each agent. It supports 20+ native plugins for Mail, Calendar, Vision, Browser, Git, and more.
Osaurus supports any model: local MLX models (Gemma 4, Qwen3.6, Llama, DeepSeek V4, MiniMax M2.5), Apple Foundation Models on macOS 26+, and cloud providers including OpenAI, Anthropic, Gemini, xAI/Grok, Venice AI, OpenRouter, Ollama, and LM Studio. It speaks OpenAI, Anthropic, Open Responses, and Ollama APIs at the same port, making it compatible with any SDK you already use. It's also a full MCP server and client, integrat ing with Cursor, Claude Desktop, and other MCP harnesses.
The tool is designed for developers, tinkerers, and privacy-conscious users who want personal AI with complete data control. It's ideal for coding partners, research assistants, file organizers, and anyone building custom AI workflows. With 129.3k downloads and 5.5k GitHub stars, it's an indie project built in public with active community support via Discord and GitHub.
Osaurus pricing
Pricing model: Freemium
Osaurus is free and open source, MIT licensed forever. There is no paid tier or subscription. You can install it via Homebrew (brew install osaurus) or download directly. The only costs come from cloud AI providers if you choose to use them (OpenAI, Anthropic, etc.)—local MLX models run completely free on your Mac. All features are unlocked with no paywalls.
Osaurus pros
- Fully offline capability with local MLX models
- Native Swift built for Apple Silicon, no Electron
- MIT licensed and free forever, open source
- Your data stays on your Mac, encrypted at rest
- Supports any model: local, Apple Foundation, or cloud
- Chat overlay accessible anywhere with ⌘; hotkey
- Persistent memory that learns from your conversations
- Linux Sandbox for safe code execution in isolated VM
- 20+ native plugins for Mail, Calendar, Git, Browser, Vision
- OpenAI, Anthropic, Ollama, and Open Responses API compatible
- Full MCP server and client for Cursor/Claude Desktop integration
- Voice input with wake-word and hands-free dictation
- Folder watchers and schedules for automation
- Cryptographic identity with per-agent access keys
- Public links without ports via secure tunnel
- Custom themes importable/exportable as JSON
- Guided 5-step onboarding, no config files needed
- Safe file/git tools scoped to working folder only
Osaurus cons
- macOS only, requires Apple Silicon (M1/M2/M3 or newer)
- Requires macOS 15.5 or later
- Sandbox and Apple Foundation Models need macOS 26+
- No native Windows or Linux support
- No built-in image generation yet
- Indie project, smaller team than corporate alternatives
- Local model performance depends on your Mac's RAM
- Some advanced features require technical setup
- No mobile app for iOS/iPadOS
- Cloud provider API keys still cost money if used
Frequently asked questions about Osaurus
What is Osaurus?
Osaurus is the native macOS app for AI agents that run on your machine. It lets you create custom agents that write code, use your tools, and evolve with you. It supports any model (local or cloud), reads your files, remembers your work via persistent memory, and keeps everything on your Mac unless you explicitly choose a cloud provider.
Is Osaurus free?
Yes, Osaurus is completely free and MIT licensed forever. There are no paid plans, subscriptions, or feature paywalls. The only potential costs are if you choose to use cloud AI providers like OpenAI or Anthropic, which charge for their APIs. Local MLX models run entirely free on your Mac.
What models does Osaurus support?
Osaurus supports local MLX models (MiniMax M2.5, Gemma 4, Qwen3.6, Llama, DeepSeek V4, GPT-OSS), Apple Foundation Models on macOS 26+, and cloud providers including OpenAI, Anthropic, Gemini, xAI/Grok, Venice AI, OpenRouter, Ollama, and LM Studio. You can swap between any model at any time.
Does Osaurus work offline?
Yes, Osaurus works fully offline with local MLX models. Nothing leaves your Mac unless you explicitly choose a cloud provider. You can turn off Wi-Fi and Osaurus will still work with local models, keeping your data completely private and on-device.
What are the system requirements?
Osaurus requires macOS 15.5 or later and Apple Silicon (M1, M2, M3, or newer). It is built in native Swift optimized for Apple Silicon and does not support Intel Macs. The Linux Sandbox and Apple Foundation Models features require macOS 26 (Tahoe) or later.
How do I install Osaurus?
You can install Osaurus via Homebrew with 'brew install osaurus' or download it directly from the website. Installation takes less than a minute. After installation, a guided 5-step onboarding walks you through creating your first agent, picking a model, and setting up identity—no config files needed.
What is the Linux Sandbox feature?
The Linux Sandbox (available on macOS 26+) lets AI agents run real code—shell, Python, Node—in an isolated Linux VM with zero risk to your Mac. Each agent gets its own Linux user and home directory for multi-agent isolation. It's powered by Apple Virtualization framework with native Apple Silicon performance, no Docker needed.
How does Osaurus memory work?
Osaurus memory distills past conversations into compact facts and surfaces them only when relevant—no firehose of irrelevant context. It uses a buffer-and-distill pipeline with a relevance gate that decides whether memory is needed, then picks at most one memory section (identity, pinned facts, episodes, or transcript) to inject into your chat.
Can I use Osaurus with Cursor or Claude Desktop?
Yes, Osaurus is a full MCP server and client, so Cursor, Claude Desktop, and other MCP harnesses get instant access to your installed tools. It speaks OpenAI, Anthropic, Open Responses, and Ollama APIs at the same port, making any SDK you already use work seamlessly.
How do I create custom agents?
You can create custom agents through the AgentsView in the app. Each agent has its own system prompt, theme, history, and generation settings (default model, temperature, max tokens). Agents automatically select tools and skills via RAG search based on the task. You can import/export agents as JSON files to share them.