# Agent Workspace

Agent Workspace is an open-source, local-first orchestration layer for CLI coding agents.

## Product Summary

- **What it is:** A GUI control plane for agent CLIs and server terminals.
- **Who it is for:** Developers running parallel coding tasks, reviews, and runtime checks across worktrees.
- **Core model:** Keep execution and orchestration local; use your own provider/CLI accounts for inference.

## Plan-Based Value

- **Claude Max + Codex plans:** Agent Workspace works with Claude Code on Claude Max and Codex CLI on ChatGPT/Codex plans.
- **No API keys required for many users:** For those plan-backed CLI workflows, separate API keys are often unnecessary.
- **Lower cost path:** That can reduce API spend because usage can stay on the user's existing plan instead of moving to per-token API billing.
- **Still flexible:** Users can still bring API keys and other CLI tools when they want broader provider coverage.

## Core Capabilities

1. **Workspaces + Worktrees**
   - One workspace can contain one or many repositories.
   - Each repository can run multiple worktrees in parallel.
2. **Native Terminal Grid**
   - Agent sessions run in real terminals.
   - Side-by-side server panes can run websites, game runtimes, tests, and CLI commands.
3. **Review and Diff Workflow**
   - Built-in review context with links to GitHub pull requests and diffs.
   - Re-prompt and verification loops from the same interface.
4. **Task and Tier Flow**
   - Tier assignment (T1-T4) helps route focus, backup, batch review, and overnight runs.
   - Task board and project board views support ongoing planning and execution.
5. **Ports and Runtime Visibility**
   - Track active services and assigned local ports.

## Why Developers Use It

- **Reduce human context switching:** Keep multiple agents, worktrees, servers, reviews, and diffs visible in one place instead of jumping between terminals and browser tabs.
- **Protect focus:** Tiered work, review surfaces, and side-by-side runtime panes help users stay locked onto the task that matters while still supervising parallel work.
- **Manage multiple threads safely:** Users can run focused work, background work, reviews, and runtime checks without losing track of what each agent is doing.

## Quick Answers

- **Does Agent Workspace run locally?**
  Yes. Sessions, workspaces, and orchestration run locally.
- **Where do model requests run?**
  Through the provider CLIs and accounts selected by the user.
- **Do users need API keys?**
  Not always. Users running Claude Code on Claude Max or Codex CLI on ChatGPT/Codex plans can often work without separate API keys.
- **Can it run more than one repository at once?**
  Yes, via mixed-repository workspace configuration.
- **Can users jump directly to GitHub PRs and diffs?**
  Yes, the review flow includes direct GitHub link actions.
- **Does anyone actually use this?**
  Yes. We built it over a few days in July 2025 and have used it ever since to ship websites and multiple games, with more than half a million plays across those projects. We have not opened an IDE since.

## Installation

### Windows (recommended)
- Download: <https://github.com/web3dev1337/agent-workspace/releases/latest>

### Source
```bash
git clone https://github.com/web3dev1337/agent-workspace.git
cd agent-workspace && npm install
npm start
```

## Canonical Links

- Homepage: <https://agent-workspace.ai/>
- Repository: <https://github.com/web3dev1337/agent-workspace>
- README: <https://github.com/web3dev1337/agent-workspace#readme>
- Terms: <https://agent-workspace.ai/terms.html>
- Privacy: <https://agent-workspace.ai/privacy.html>
- Sponsors: <https://github.com/sponsors/web3dev1337>