mai is a persistent home for a fleet of Claude & Codex sessions — tile a whole team into one screen, run any model, let a Hive of agents build and ship, and see every token. Below: watch it work, then browse all 200+ capabilities.
Every API call is captured, so you see the exact dollar cost of every session, project, and agent — live. No mystery bills, no black box.
Shells, dev servers, builds, long-running agents — start them once and they stay alive through reboots, crashes, and dropped connections.
Tile a whole fleet of sessions into one screen — split, swap, save layouts, pop out. Drive your entire operation at a glance.
Run dozens of agents in parallel and review results instead of babysitting them. The throughput of a whole team, driven by one person.
Real captures of the actual mai UI — the tiling workspace, any model, the Hive, live cost, and more. Hover or scroll a clip to play it.
mai is extensible to the point of play — a live pixel office, an LCD pet, and a desktop companion, each driven by your real session activity.
Each of these is impractical — or impossible — in a single plain Claude Code session.
mai <verb>Headless CLI. Every operation, with no running server and no token — fast enough to run inside a Claude hook, a cron job, or another agent's loop.
mcp__mai-sessions__*MCP server, auto-wired into every mai session. An agent can spawn a sub-agent, watch its screen, wait for it to finish, and act on the result.
/api/v1/*HTTP + a pushed event stream. An external app can create and steer its own owner-tagged fleet of Claude sessions over the wire.
Compiled from a full-codebase sweep. Each card lists what the capability lets you — or your agents — do, and the interfaces that expose it.
Install the web / chat tier in one line, then drive the whole thing from a browser, a phone, or another agent.
curl -fsSL https://buy-mai.com/install.sh | sh