myAI — AI Management Framework

myAI turns a folder of git repos into an AI-managed fleet: every agent session boots with full context, autonomous work runs off-hours on a task queue, and one dashboard shows everything.

Self-hosted, single operator, your own Anthropic key. No SaaS account, no lock-in — the whole stack (gateway + dashboard + MongoDB) runs on localhost via Docker.

Where to start

PageWhat you'll find
QuickstartDownload → running dashboard in 5 minutes
ConceptsThe continuity model, memory tiers, the runner, budgets, and the fleet
BrainGit-versioned agent memory in depth — atoms, branches, the distiller, the CLI
GatewayREST + MCP overview — the two doors into the same daemon, auth model, connecting an agent
KeywordsThe agent mode / ship it / wrap up / brain … vocabulary every managed repo ships with
Self-host guideRun gateway + dashboard + Mongo on your own infrastructure — Docker or Helm
ComparisonsHonest myAI vs Mem0 / Zep / Ruflo / Composio / Langfuse, with a summary matrix — mostly complementary tools, not rivals
CLI referenceEvery myai command — auto-generated from the CLI source
MCP tool referenceEvery gateway MCP tool — auto-generated from the gateway source
API referenceInteractive Redoc explorer for the gateway REST/MCP surface — generated from the OpenAPI spec

Try it hands-on

Two copy-paste walkthroughs prove the core loops in minutes. They ship in the npm package and live at the repo root:

  • Try myAI in ~10 minutes — install the CLI from a local build, scaffold a throwaway project, bring the full stack up on free ports, click around, tear it down. A real cold-start test that never touches a live stack.
  • Try the Brain in ~5 minutes — prove git-versioned agent memory by hand: init → work → stash → pop from a second agent → merge → diff, against a throwaway brain. No gateway, no Docker, no API key.

What it does, in one paragraph

You install one CLI (npm i -g ai-management), run myai init and myai up, and get a local gateway (an MCP server at localhost:3100) plus a dashboard (localhost:3210). Point any MCP-capable agent at the gateway (myai connect-agent) and it boots with your operator context, project state, and semantic memory — no cold starts. Queue tasks (myai schedule) and a headless CLI runner builds them off-hours, surfacing finished work for your review. Everything stays on your machine.