Beta·every release is a pre-releaseAPIs and harness behavior are still changing·every release until v1 is a pre-release

The terminal workspace for coding agents

Agents come and go.Your work stays.

A terminal-native ADE: the same plugins and AGENTS.md in Claude Code, Codex, OpenCode and Antigravity, running side by side, each in its own worktree.

$ curl -fsSL https://uze.sh/i | sh

x86_64 or aarch64, signature and checksum verified.

Get started

Three commands.

  1. Add a marketplace

    $ uze market add hiukky/ai

  2. Install a plugin into every agent

    $ uze git@ai

    • ✓Claude Code
    • ✓Codex
    • ✓OpenCode
    • ✓Antigravity

  3. Run them side by side

    $ uze workspace

    Each agent on a branch of its own, in its own worktree.

What each agent receives, capability by capability.

What uze does for each of your agents

Install once. Run them side by side. Bring the work home.

Plugins

One plugin. Every agent.

Install a plugin once and each agent receives it through its own mechanism: a real plugin where the agent has one, its native files where it does not.

  • Skills, subagents, MCP servers and hooks, in each agent’s own format
  • Recorded in agents.yaml, so a teammate’s uze install gets the same
  • A receipt for every file placed: drift stops uze, it never overwrites
How delivery works
~/demo-app

Workspace

Every agent on a branch of its own.

Start four agents on one repository and none of them touches another’s files, or yours. The sidebar says where each one stands, read from Git, not from what the agent claims.

  • A worktree per agent, reused warm: no fresh node_modules, no cold build
  • Branches named from the work: add rate limiting is feat/add-rate-limiting
  • Close an agent mid-task and its work is kept, never lost
Where an agent works
uze workspace

Delivery

One key brings the work home.

When an agent is ready, Alt+I rebases its branch onto yours, runs your checks, and merges it or pushes it. Nothing reaches your branch until the checks pass.

  • A conflict or a failed check goes back to the agent, in your project’s words
  • Your checks run only after you read and approve them
  • Deliver all: every ready agent in one go
Bringing the work home
uze workspace

Continuity

Out of tokens? Switch harness, keep going.

When one harness hits its limit, the same agent carries on in another: same directory, same branch, same name. The one it left keeps its conversation for when you switch back.

  • A briefing of what it was asked, said and committed
  • Its uncommitted files and last command included
  • Kept out of your checkout, readable by you alone
Continuing in another harness
uze workspace

Questions people ask first

The short answers. The docs have the long ones.

Something else? Read the full FAQ or ask on GitHub.

Is uze another coding agent?
  • Not an agent
  • No model
  • No API key

Your agents keep the logins and subscriptions they already have. uze installs plugins into them and runs them, and you talk to each one exactly as you do today.

Does it send my code anywhere?
Your code goes only where your agents already send it. uze itself fetches the marketplaces you registered, and checks GitHub for a new release. Release builds contain no telemetry.
Why not symlink one skills folder into every agent?
Because they do not read the same formats. Codex has no Markdown agent format, so a subagent becomes TOML there; a skill only you may run is written differently in each agent; hooks run through a different mechanism in each. uze also keeps a receipt for everything it places, which is what lets it remove a plugin cleanly later.
Why not just use Claude Code’s own plugins?
On Claude Code, uze delivers through Claude Code’s own plugin mechanism, so nothing is lost. What you add is the same plugin reaching Codex, OpenCode and Antigravity, and a project that records it, so every teammate gets it too.
Do I need the workspace to use the plugins?
No. They are two tools in one binary and each works without the other. The package manager serves agents you start yourself, in a terminal, an editor or CI.
What does the workspace add to tmux and git worktree?
Each agent gets a tab, a branch and a worktree of its own without you creating any of them. You see its diff a keystroke away, and when the work is ready one key rebases it, runs your checks, and merges it or pushes the branch. An agent can carry on in another harness without losing its place, and work nobody finished is kept and listed in Recent, across every project. Closing the terminal stops nothing.
Can a plugin run code on my machine?
Yes: MCP servers and hooks are commands. Before installing, uze lists every command a plugin can run and asks, and asks again when an update adds one. Plugins are fetched without your Git credentials and never run a repository’s own hooks.
Does it work on Windows?
Yes, natively, on Windows 10 22H2 and 11, x64 or Arm: install with irm https://uze.sh/i | iex. It needs Git for Windows, and WSL works with the Linux command too.
What does it cost?
Nothing. uze is open source under the Apache License 2.0, and it is in beta: commands and file formats may still change before v1.

Set it up once, on this machine.

uze finds the coding agents you already have, installs any you want through the vendor's own installer, and reports what it could not do rather than guessing. One Rust binary, open source under the Apache License 2.0.

Get startedSource on GitHub
$ curl -fsSL https://uze.sh/i | sh

x86_64 or aarch64, signature and checksum verified.

Built by Romullo (@hiukky).

All product names, logos and brands are the property of their respective owners. uze names the coding agents it interoperates with for identification only, and is not affiliated with, endorsed by, or sponsored by any of them.