Cherry Studio is a cross-platform desktop AI client that connects to many LLM providers, local models and 300+ preset assistants.
osaurus-ai/osaurus
A native macOS AI harness that runs any local or cloud model with per-agent memory, sandboxed tools, and cryptographic identity, fully offline.
Open-source alternative to:ChatGPT
- Stars
- 8,135
- Forks
- 492
- Open issues
- 229
- Last commit
- 16 hours ago
What is it?
Osaurus is a macOS app that sits between you and AI models, whether they run locally on Apple Silicon or come from cloud providers. It keeps the layer around the model on your machine: agent prompts, persistent memory, tools, and identity. Each custom agent has its own prompts, memory, and working folder, and tools and skills are selected automatically through RAG search based on the task.
Agents can execute code in an isolated Linux VM built on Apple's Containerization framework, or fall back to a Seatbelt sandbox on earlier macOS versions. Osaurus also works as an MCP server and client, exposes OpenAI, Anthropic and Ollama compatible endpoints on localhost, and includes built-in Apple app tools such as Calendar, Reminders and Mail. Local models run through MLX, and Apple Foundation Models are available on macOS 26+.
Cloud requests can be routed through an on-device privacy filter that redacts names, emails, secrets and similar data before sending, and blocks the send if anything leaks. The project is written in Swift, requires macOS 15.5+ and Apple Silicon, and is MIT licensed.
Key features
- 01Runs local MLX models on Apple Silicon or connects to OpenAI, Anthropic, Gemini and other cloud providers.
- 02Per-agent persistent memory that distills conversations at session end.
- 03Agents execute code in an isolated Linux VM with per-agent users and a vsock bridge.
- 04Acts as an MCP server and client, and exposes OpenAI, Anthropic and Ollama compatible APIs.
- 05Privacy filter redacts sensitive data before cloud requests and blocks leaks.
- 06Cryptographic identity with access keys, plus end-to-end encrypted agent-to-agent channels.
Use cases
- Running a private AI assistant on a Mac that works without an internet connection.
- Pointing existing OpenAI or Ollama compatible tools at a local model server.
- Giving a coding agent a sandboxed shell and a working folder to execute tasks.
- Sending prompts to cloud models while redacting personal data first.
Quick start
- 1
Install the app with Homebrew
brew install --cask osaurus - 2
Open the chat UI
osaurus ui - 3
Start the server
osaurus serve
Things to know
- macOS 15.5+ on Apple Silicon only.
- The Linux VM sandbox requires macOS 26+; earlier versions use a less isolated fallback.
- The full build bundles a roughly 4 GB model download.
In collections
Topics
FAQ
- What is osaurus?
- A native macOS AI harness that runs any local or cloud model with per-agent memory, sandboxed tools, and cryptographic identity, fully offline. Osaurus is a macOS app that sits between you and AI models, whether they run locally on Apple Silicon or come from cloud providers. It keeps the layer around the model on your machine: agent prompts, persistent memory, tools, and identity. Each custom agent has its own prompts, memory, and working folder, and tools and skills are selected automatically through RAG search based on the task.
- Is osaurus free to use? What license does it have?
- osaurus is open source under the MIT license. Check the license terms before using it in commercial products.
- Is osaurus still maintained?
- Actively maintained. The last commit to osaurus was on Nov 6, 2026.
- What can I use osaurus for?
- Running a private AI assistant on a Mac that works without an internet connection. Pointing existing OpenAI or Ollama compatible tools at a local model server. Giving a coding agent a sandboxed shell and a working folder to execute tasks. Sending prompts to cloud models while redacting personal data first.
- How do I install osaurus?
- Install the app with Homebrew: brew install --cask osaurus Open the chat UI: osaurus ui Start the server: osaurus serve
- Where can I get osaurus?
- The source code and installation instructions are on GitHub: https://github.com/osaurus-ai/osaurus
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
LLM inference server for Apple Silicon Macs with continuous batching, tiered RAM/SSD KV caching, and menu bar management.
A self-improving AI agent that creates and refines skills from experience, runs on many model providers, and works through CLI and messaging apps.
Self-hosted personal AI agent framework in Python with a browser WebUI, terminal client, chat app channels, tools, memory and MCP support.
Open-source personal AI assistant and Agent Harness that plans tasks, runs tools and skills, and keeps memory across web and IM channels.
A self-hosted personal AI assistant in Rust with WASM-sandboxed tools, encrypted credential storage, and a web UI, inspired by OpenClaw.