Osaurus

Overview

Osaurus is a native macOS application that functions as a personal AI harness, sitting between users and any language model—local or cloud—to provide persistent memory, autonomous execution, and sandboxed tool usage. Built entirely in Swift for Apple Silicon, it operates fully offline using on-device MLX inference while offering seamless connections to major cloud AI providers. The platform features a three-layer salience-scored memory architecture and acts as a full Model Context Protocol (MCP) server and aggregator.

Founder story

Osaurus was created by Terence, a veteran software engineer with 20 years of experience shipping products at companies like Netflix, Tesla, and Zillow. Frustrated that most AI tools force users to send sensitive files and personal context to external servers behind recurring monthly subscriptions, the founder built an open-source, local-first macOS AI harness designed to give users complete ownership over their data.

What it does

• - Runs open-source LLMs locally on Apple Silicon via an optimized Swift MLX runtime without cloud dependencies
• - Provides a three-layer salience-scored memory system to inject relevant context without bloating windows
• - Executes code safely inside isolated Apple Containerization linux development environments
• - Exposes installed tools and native plugins via a built-in Model Context Protocol (MCP) server.

Who it's for

• AI Engineers
• Mac Power Users
• Software Developers
• Privacy-Conscious Professionals

Why it works

• Operates fully offline on Apple Silicon hardware, ensuring user data and private documents never leave the local machine
• Utilizes high-performance native Swift code instead of heavy cross-platform Electron frameworks for optimal speed
• Provides open-source flexibility under a permissive MIT license with no usage caps, subscriptions, or telemetry
• Combines local model execution with remote cloud fallback options through an extensible plugin and MCP framework.

Growth strategies

• Open- source distribution and community engagement via GitHub repository stars and forks
• Developer adoption driven by native Model Context Protocol (MCP) server compatibility and CLI tools
• Product- led growth through direct Homebrew cask installation and website downloads
• Word- of-mouth recommendations among privacy-focused AI practitioners and macOS power users

Alternatives

• LM Studio
• Ollama
• AnythingLLM
• Jan.ai

Comparison overview

• Standard cloud AI assistants route all personal context, memory, and queries through third-party servers under monthly subscription models.
• Osaurus runs locally on macOS as an open-source harness, keeping memory, tools, and execution entirely on-device with optional cloud integration.