Assistant

An AI-powered personal assistant that processes emails, pull requests, issues, and other inputs using LLMs — safely.

Assistant uses AI to sort through the noise of digital communication, classifying messages and taking action based on rules you define. It transforms sender-benefit communication into receiver-benefit communication, so what reaches you is actually worth your time.

Core Principles

Reversibility — Every autonomous action must be reversible. Draft instead of send. Archive instead of delete. If an action cannot be undone, it requires human approval.

Audibility — Every action the AI takes must be auditable. The filesystem is the database: task queues are YAML files you can inspect, notes are markdown, and daily logs are human-readable.

Accountability — AI has ability but no accountability. The LLM classifies; deterministic code decides. The dispatch layer is the safety boundary, not the LLM.

Learn More

• Why Assistant Exists — the problem this project solves
• Design — design principles and decisions
• Architecture — how the system is built
• API — endpoint reference
• Development — getting started as a contributor
• Testing — test guide and philosophy