myAI: an autonomous multi-agent engineering framework

What it is

myAI turns AI-assisted development from a chat window into an operating model. Every repository in the fleet carries the same self-contained kit: thirteen specialist agents (frontend, API, database, security, QA, architecture…), sixty-plus skill playbooks, state and handoff files for continuity, and routing rules that dispatch work to the right specialist in parallel lanes.

Above the repos sits a master gateway — an MCP server exposing a task queue, a plan dashboard, and shared memory. A scheduled runner picks tasks off the queue and works them autonomously in headless sessions, so the fleet makes progress overnight; each morning the operator reviews, merges and redirects.

The brain: git semantics for agent memory

The hardest problem in agentic work is continuity — every session traditionally starts amnesiac. myAI's brain applies git semantics to memory: each session commits a compact summary atom, wrap-up merges it into the brain's main line, and a distiller recompiles briefs automatically. The next session — on any machine — boots with a few-hundred-token delta of exactly what changed, instead of re-reading everything.

• Sessions are commits; wrap-up is a merge; catching up is a diff
• Provenance runs both directions — from a code SHA to the reasoning behind it, and back
• State files remain a plain-markdown fallback, so continuity survives even with the gateway down

The safety model

Autonomy is only useful when it's bounded. The framework encodes non-negotiable rails: no agent ever pushes to main — everything lands on a test branch behind CI and a PR; secrets never enter commits; destructive operations require explicit human ask. Autonomous sessions carry a three-strikes rule with mandatory postmortems, so a stuck agent stops and explains rather than thrashing.

Why it matters

This site — including the pages you're reading — is partly built and shipped through this pipeline: planned in one session, designed in another, built by a scheduled agent overnight, reviewed by a human in the morning. myAI is the ongoing experiment in what engineering leadership looks like when part of your team is autonomous. The full journey is on the About page.