Pheron Agent Documentation
Pheron Agent is a fully autonomous AI agent for macOS, built entirely on Apple Silicon. It runs local LLM inference via MLX, orchestrates multi-step tasks through the UNO (Unified Native Orchestration) architecture, and operates without sending your data to the cloud.
Current version: v1.0.6 · Requires: macOS 26.0+ (Tahoe or later), Apple Silicon (M1 or later)
Getting Started
- macOS Permissions & Setup — Grant required permissions and download your first local model
- Models & Hardware Tiers — Find the right model for your chip
- Updating Pheron Agent — Keep your application up to date
Architecture
- Architecture Overview — UNO pipeline, Swift 6 actors, Titan Engine
- Local API (Titan Hub) — REST endpoints for local inference and agent orchestration
Core Features
- Full Tool Inventory — Comprehensive inventory of all 62 native & MCP tools with UBIDs
- External Tool Integrations Reference — Detailed setup, auth, and action reference for MCP, REST, and Blender bridges
- Lark Suite Setup — Step-by-step guide to connecting your Lark Suite app to Pheron Agent
- LemonSqueezy & Kit Setup — Connect your LemonSqueezy store and Kit (ConvertKit) email list to Pheron Agent
- SkillVault — Self-improving procedural memory
- Native Tool Calling — How the agent selects and executes tools
- AudioIntelligence DSP — Music DNA and audio analysis
- System Stability & Self-Healing — AutoRecoveryEngine and context management
Advanced & Reference
- Benchmark & Test Results — Universal AgentTestMethodology v7 evaluation metrics & OWASP security matrix
- Security & Privacy — On-device isolation, Privacy Guard, XPC sandboxing
- Product Roadmap — Upcoming features
- v3 Migration Guide — Upgrading from mlx-swift-lm v2
For questions or support, see the Help & Support page.