Uai is live: your AI dev team, hosted.

For a year the default way to code with AI has been to babysit one chat. It works one turn at a time. The commits land authored by a bot, on your history. There's no second agent to catch the first one's mistakes. And too often it ran somewhere you couldn't watch or review — you got a result, not a diff.

We think the unit of AI coding should be a team you design, not a box you feed prompts to. So that's what we built.

A team, not a chatbot

You design the crew once, and reuse it.

The building block is a persona: a reusable, named teammate you design once and drop onto any task. Each carries a name and @handle, an avatar, a role and brief (its system prompt), an engine — Claude or Codex, with model and effort — attached skills, permissions for what it may do, and memory that carries between tasks. Group personas into Teams with a leader and drop a whole team onto a task as a unit.

The agents on a single task are its crew. They answer in the task chat and review each other — an @architect proposes, a @reviewer pushes back — before anything reaches you. Point a planner at a goal and it drafts the backlog: fully-configured task and subtask drafts that stay inert until you approve them with Start.

Parallel & reviewable

Many tasks at once — and you review every diff.

Each task is fully isolated in its own container and its own git worktree, so N run in parallel without stepping on each other; your plan sets the concurrency cap. You follow each task's chat live and can steer it mid-run — the orchestrator routes your @mentions.

When a task is ready, you review the diff in the product — a drawer that spans every repo in the task, viewable split or unified, comparing task/<slug> against origin/<default>. Not "go check GitHub." Every running task also exposes a live preview URL, so you can see the app running before you merge.

Your machine, but never loose on it

Your hardware, your keys, your name — no blast radius.

There are three ways to run an AI agent. In someone else's cloud, where you hand over your code, your keys, and control. Loose on your own box, un-sandboxed, with full reach over your files and credentials. Or Uai's way: on your machine, with every task sealed in a disposable container and a fresh worktree.

Tasks run on your host with your AI keys (never marked up) and your GitHub + SSH identity — so commits and PRs are attributed to you, not a bot. Yet each task is fenced in: it never bind-mounts ~/.claude or ~/.ssh, Uai uses its own SSH identity, ports bind to 127.0.0.1 only, and the container and worktree are deleted when the task ends.

And the cloud is secret-blind. It stores an environment schema — key names and labels, never values — and holds no AI keys, no SSH keys, and no GitHub tokens. Your secrets live encrypted on your host. The cloud only knows "this host has creds: yes/no."

Multiplayer

Bring the whole team — humans and agents — into one task.

Invite teammates into a task's chat as collaborators. @mention agents and people in the same channel and the orchestrator routes to each. When an agent needs you, it @mentions you — and that mention pushes to your browser or phone, so you can weigh in from anywhere.

What's real, and what isn't yet.

We'd rather you hear it from us. The desktop apps — the Uai client and the Uai Host supervisor, both for macOS — are real, signed, and notarized. The mobile app is a scaffold: it typechecks, it isn't on any store, and we're not going to pretend otherwise. It's coming.

Access is behind a waitlist right now. The product is running and design partners are shipping with it, but we're letting people in in batches rather than opening the doors and watching hosts fall over. Join the waitlist and we'll get you set up — we'd rather make you wait a few days than hand you a broken first task.

Pricing is locked, and it's simple. Every account is an organization. Free is solo and free — one seat, your own machine. Dev, Pro, and Max add seats, and every seat adds a host, more parallel tasks, and a bigger crew; the per-seat rate drops as the plan scales ($19.99 on Dev, $17.99 on Pro, $15.99 on Max). Self-host isn't a tier — it's a deployment mode; your host is unbilled compute. You pay for orchestration and collaboration, never for inference.

Where it's going.

The team abstraction isn't specific to code. Planning work, reviewing it, and shipping it in parallel generalizes well past pull requests. Software is the first thing a crew is obviously good at — not the ceiling. That's the direction; today's release is the foundation. It's built on Claude Code and Codex under the hood.