AI context management
Install once and a whole AI work platform comes up on your machine. Any agent — Claude, Cursor, Gemini, Codex, opencode — joins one swarm, boots already knowing your projects and decisions, and never forgets what's done, what's built, or what's next. Local-first, MCP-native, and yours.
Walkthrough
Install once, and a whole AI work platform runs on your machine. Watch it come together in under a minute.
One npm command brings up the whole local stack in Docker: the myAI gateway, a Mongo memory database for recall, the brain, an autonomous runner and local CI.
Claude, Cursor, Gemini, Codex or any MCP agent plugs into the same gateway and the same memory.
You talk to one session; it assigns work, watches the others, takes over stalls, and ships.
Each machine warms up every morning — code pulled, brain synced, stack checked.
What's done, what's built, what's next, and why — kept by the brain and the backlog.
One place to see all of it and steer.
The problem
Every frontier coding agent is a brilliant amnesiac. Out of the box it has no memory of last week's decisions, no handoff between machines, no idea what it did on your other laptop yesterday. Everyone re-invents this by hand as a pile of markdown and rituals — and pays a re-teach tax in tokens and time at the start of every single session.
myAI's thesis: this layer — context, memory, sessions, handoff, retrieval, scheduling, and a control dashboard — should be portable, user-owned, and model-agnostic. Any agent plugs in and is bootstrapped with your context at session start.
The Brain
The hardest problem in agentic work is continuity — every session traditionally starts amnesiac. The brain fixes that by treating memory the way you already treat code.
How it works
You set direction from the CLI, your phone, or Telegram. A local gateway coordinates specialist agents and an autonomous runner. The brain remembers across all of it.
What you get
A git-versioned brain you own carries what was done, what was built, and what's next across sessions, agents and machines. A returning agent boots with a short catch-up of what changed — not a re-read of everything.
Bring any MCP agent and it joins the swarm. One swarm master hands out the work, watches progress, picks up anything that stalls, and ships — the others build in parallel without stepping on each other.
You talk to one session — from the terminal, your phone, or Telegram — and it drives the rest. Approve, redirect, or check in without sitting at the desk.
Every machine boots itself at 08:00: pulls the latest code, syncs the brain, checks its stack, and is ready before you sit down.
Gateway, memory database, brain, runner and local CI all run in Docker on your machine. Your data stays yours, and checks run locally instead of burning hosted CI minutes.
What's happening now, what needs you, what's done and queued, what it's costing, and what the brain knows — in one view, not a maze of pages.
Progress
myAI is production-grade and used every day across a fleet of real repositories. A few recent milestones:
Private vulnerability reporting wired end-to-end; hardened public surface and a documented disclosure policy.
Health/readiness probes, graceful drain, OpenTelemetry tracing, per-tenant webhook receiver, and a Helm chart for self-hosting.
Git-versioned memory becomes the source of truth every session boots from, with RAG semantic recall and a composite brain health score.
A scheduled runner works the queue overnight across a fleet of repos, driven from one morning console — coffee, review, redirect.
Deeper reading
The architecture of myAI — a git-versioned memory layer, a specialist agent fleet, an autonomous off-hours runner, and an MCP gateway that ties them together. How it works, why it's shaped this way, and where it could go.
Two months of changes to myAI since the last architecture post — a universal daily boot, a brain that surfaces prior art and grades its own health, a fleet-wide MCP bridge, self-hosted CI runners, and the doorway that lets me drive the whole thing from my phone.
AgentFlow turns an idea typed into a text box into a repository you can actually run — the architecture that gets from prompt to working project.
FAQ
One portable brain that you own, served to every agent. Bring your own credentials, run it on your own machine, take your context anywhere.