Osaurus
Osaurus is an always-on AI edge runtime for macOS that runs local models via MLX on Apple Silicon or routes to cloud providers (Anthropic, OpenAI, xAI, Ollama), exposing both OpenAI-compatible and Anthropic-compatible APIs alongside a built-in MCP server for agent tool access.
Score Breakdown
⚙ Agent Friendliness
🔒 Security
Documentation/Docusaurus MCP tool. Read access to documentation. Low security risk for public docs. Private docs need auth layer on top.
⚡ Reliability
Best When
A macOS developer wants a local-first AI runtime that works with both local models and cloud APIs through a single OpenAI/Anthropic-compatible interface, with MCP tools always available.
Avoid When
Running on Linux/Windows, or when centralized cloud AI infrastructure is preferred over local edge compute.
Use Cases
- • Running local LLM inference on Apple Silicon without cloud costs using MLX-optimized models
- • Providing a unified API endpoint that proxies to local or cloud models interchangeably
- • Building custom AI agents with persistent memory, voice input, and file-monitoring triggers
- • Using MCP tools from a local always-on server across multiple AI applications
Not For
- • Non-macOS or non-Apple Silicon machines
- • Teams needing centralized multi-user AI infrastructure
- • Production server deployments (designed as a personal workstation runtime)
Interface
Authentication
Access Keys (osk-v1 tokens) for API auth stored in macOS Keychain. EIP-191 signature-based agent identity. Remote providers use their own keys stored securely in Keychain.
Pricing
Appears to be open-source. Cloud provider API costs apply when routing to Anthropic/OpenAI/etc.
Agent Metadata
Known Gotchas
- ⚠ macOS 15.5+ required; Apple Foundation Models need macOS 26+
- ⚠ Apple Silicon (M1+) is mandatory for local inference
- ⚠ Runs on port 1337 by default; may conflict with other local services
- ⚠ Work Mode for autonomous tasks may have unpredictable execution boundaries
Alternatives
Full Evaluation Report
Detailed scoring breakdown, competitive positioning, security analysis, and improvement recommendations for Osaurus.
Scores are editorial opinions as of 2026-03-06.