Osaurus
Category: AI Infrastructure
Open-source, native macOS AI harness for running local and cloud AI models on Apple Silicon with persistent memory, tools, and sandboxed execution. Osaurus was founded in 2025. The company is led by Terence Pae. Based in Los Angeles, California, USA. Latest round: Undisclosed.
- Founded
- 2025
- Headquarters
- Los Angeles, California, USA
Value proposition
Osaurus is a local-first AI runtime for macOS that lets users run AI models (local via MLX or cloud via OpenAI/Anthropic/etc.) with persistent memory, autonomous agents, and sandboxed code execution — all while keeping data on-device. It's an open-source (MIT), native Swift alternative to cloud-dependent AI tools.
Products and solutions
Osaurus AI Harness (macOS app): local LLM server with OpenAI-compatible API, agent system with persistent memory, sandboxed code execution (Linux VM via Apple Containerization), 20+ native plugins (Mail, Calendar, Vision, Browser, Git, Filesystem, etc.), MCP server/client, voice capabilities, cryptographic identity system, secure WebSocket tunnels for agent access
Unique value
Only native macOS AI harness built purely in Swift on Apple Silicon — no Electron, no runtime dependencies. Combines local-first privacy (data never leaves device) with seamless cloud model switching. Open source (MIT), sandboxed execution, persistent memory that compounds over time, and cryptographic identity — all in one free app.
Target customer
Mac power users, developers, privacy-conscious professionals (legal, healthcare), and AI tinkerers who want local AI with cloud flexibility
Industries served
Consumer AI, Developer Tools, Legal (privacy-sensitive document AI), Healthcare (on-premise AI for patient data), Enterprise productivity
Technology advantage
Native Swift on Apple Silicon using MLX framework for optimal performance; hardware-isolated sandbox via Apple's Containerization framework; three-layer memory system (identity, pinned facts, episodic) with salience scoring; secp256k1 cryptographic identity chain; model-agnostic architecture allowing seamless switching between local and cloud providers; 20+ native macOS plugins
How they differentiate
Unlike Ollama (CLI-focused, cross-platform) and LM Studio (GUI-focused, cross-platform), Osaurus is Apple-only, native Swift, and offers a full agent harness with persistent memory, sandboxed code execution, cryptographic identity, and 20+ native macOS plugins — positioning as a complete "AI OS" for Mac rather than just a model runner. More consumer-friendly than developer-only tools like OpenClaw.
Main competitors
Ollama; LM Studio; Msty; OpenClaw
Key partnerships
Alliance Accelerator (NYC-based crypto/AI accelerator); Hugging Face (model distribution); Apple ecosystem (MLX framework, Containerization, Apple Foundation Models)
Major milestones
Dec 2025: First public release (open-source macOS LLM server); May 2026: TechCrunch coverage, 112K+ downloads; Participating in Alliance accelerator (NYC); Evolved from predecessor product Dinoki (AI desktop companion)
Growth metrics
123,900+ downloads; 5,500+ GitHub stars; 295 forks; 2,509 commits
Market positioning
Early-stage entrant in the local AI runtime space, differentiated by Apple-exclusive focus, native Swift performance, and comprehensive agent/memory/tools stack. Competing against more established but cross-platform tools (Ollama, LM Studio) by offering deeper macOS integration and a more complete "personal AI" experience.
Geographic focus
US (primarily), global via open-source distribution
About Terence Pae
Ex-Software Engineer at Tesla (3 years, built backend infrastructure for Model 3 launch); Ex-Software Engineer at Netflix; Co-Founder & CTO of NiftyKit (NFT platform); Ex-iOS Engineer at Trulia/Zillow; BA Computer Science, UC Santa Barbara
Official website: https://osaurus.ai