Tesseract-Prime
Your personal AI assistant — powered by local LLMs, running entirely on your Mac. Private, agentic, and extensible.
Features
- Multi-model LLM orchestration via Ollama — switch between Gemma, Mistral, Qwen, Llama and more
- 4-tier tool capability classification — automatic model-aware tool routing
- MCP (Model Context Protocol) integration — connect to external tools via stdio servers
- Agentic workflows with ReAct prompting, function-call interception, and proactive tool inference
- Non-sandboxed helper app for system-level actions — CGEvent input simulation, command execution
- Auto-discovery of MCP servers from Claude and other configurations
- Completely on-device — no API keys, no cloud, no data leaves your Mac
- Native macOS SwiftUI application with Mach XPC architecture
Architecture
Built for power users who want AI without compromise.
Tool Orchestrator
Intelligent 4-tier classification routes each model to its optimal tool-calling strategy — from native JSON function calls to ReAct prompting to raw text interception.
MCP Ecosystem
Meta-tool pattern expands external MCP servers at runtime. Auto-discovery finds servers from your existing Claude configuration. Per-server timeouts and caching.
TesseractHelper
A non-sandboxed companion process handles privileged operations — MCP stdio transport, keyboard simulation, shell commands — communicating via Mach XPC.
Open Source
Tesseract-Prime is distributed as a signed, notarized .pkg installer.
View on GitHub