Version 1.2.39 New: MCP for all your agents →

All your AI agents.
One workspace.
You in charge.

Turn Codex, Claude Code and any ACP-compatible agent into an organized team. You delegate the work, approve every important step, review the changes and run your project without leaving a single window.

Free · No new account · No middleman servers · Everything stays on your computer

Get started in minutes

From download to your first task, in three steps

  1. Install the app

    The installer on Windows, or a few commands on Debian, Ubuntu or Arch Linux, and it is ready to use.

    See how to install
  2. Connect your agent

    Use the accounts you already have, or try it at no cost with the demo agent.

    • Claude Code
    • Codex
    • Any ACP agent
    • Demo
  3. Create your first task

    Create a space, link your project and delegate. You approve every important step.

    Alt + A new space

The problem

AI already writes code.
Working with it is chaos.

This is what an ordinary day looks like when every agent lives in its own terminal.

  1. You open one terminal for Claude Code and another for Codex.

  2. You no longer remember which agent changed which file.

  3. The agent asks for permission and you approve it without reading because you are in a hurry.

  4. The quota runs out in the middle of a task.

  5. You close the window and the conversation is gone, along with the reasons behind every decision.

Sound familiar?

  • Scattered agents. Each one in its own terminal, with its own format and no shared memory.

  • Conversations that vanish. Close the terminal and the reason behind every change is lost.

  • Blind trust. Watch every line, or grant full access and cross your fingers.

  • Ten windows for one task. Chat, editor, Git and terminal all in different places.

  • One task at a time. While the agent works, you wait.

  • Surprise quotas. You don't know how much usage you have left until it runs out.

Now imagine all of this fitting in a single window.

↓

The solution

Agentic Workspace brings order.
You take back control.

Your agents, your permissions, your code and your history, together in a single window.

Unify

Codex, Claude Code and any ACP agent in the same interface, with the same workflow.

Control

The agent works on its own, but every command, write or sensitive access waits for your approval.

Integrate

Conversation, files, code viewer, Git and terminal in a single window.

Keep

Every instruction, answer and change is saved, backed up and ready to recover.

From chaos to control

  • Before: One terminal per agent

    Now: All your agents in one place; switch agents in the middle of a task

  • Before: Conversations lost when you close

    Now: A permanent history per project and task, exportable to Markdown or JSON

  • Before: Approving everything unread, or granting full access

    Now: Clear permission cards: allow once, always allow or reject

  • Before: Jumping between chat, editor, Git and terminal

    Now: Everything built in: review the diff, commit and push without leaving

  • Before: Waiting for the agent to finish

    Now: Several tasks in parallel, and a notification when they need you

  • Before: Running out of quota mid-task

    Now: A meter with the usage left and a countdown to the reset

  • Before: Losing work to an unexpected crash

    Now: Automatic recovery, retries and verified backups

Features

What makes the difference

01 / 04

Delegate with autonomy, decide with control

AI does the work. The important decisions are still yours.

  • When the agent wants to run a command, access the network or write a file, it shows you a card with Allow once, Always allow or Reject.
  • Modes that fit you: Plan so it proposes without touching anything, Accept edits to move faster, and Full access for trusted projects.
  • By default, agents only read inside your project.

The problem that goes away: choosing between watching every line or granting full access.

02 / 04

Built-in Git, the way a professional IDE does it

Review, accept or revert what the AI did in seconds.

  • Per-file diff of every change the agent made, right next to the list of changes, before you commit it.
  • Staging with checkboxes, Commit, Commit and push, Fetch and Pull with merge or rebase.
  • Branch picker, discarding changes with confirmation, and a .gitignore assistant.

The problem that goes away: jumping to another tool to find out what changed.

03 / 04

Run your project without leaving

Try it as soon as the agent finishes, just like in your IDE.

  • It detects what can run on its own: the package.json scripts and the Python entry points, with their virtual environment.
  • Pick the configuration and press ▶; ■ stops it. The output shows up in the built-in terminal, right under the project files.
  • If dependencies are missing, it tells you and you install them from the same menu.

The problem that goes away: opening another terminal to check whether what the agent did actually works.

04 / 04

Instant context, results in plain sight

Give the agent what it needs without leaving the conversation.

  • Attach files by dragging them, pasting them, from the explorer or with the paperclip.
  • Type / to mention another project or a branch right in the instruction.
  • Built-in viewers for code (with the VS Code engine), Markdown, PDF and images.

The problem that goes away: preparing the context by hand and opening other apps to review the results.

The agent asks for permission to create a file in the project, with the options to allow it once or reject it
Git tab with the modified files on one side, the README.md diff on the other and the Fetch, Pull and Push buttons
Run configurations menu with the detected dev, build and test scripts, next to the run button, and the project running in the terminal under its files
References menu opened by typing a slash in the message, with the project from another space ready to be mentioned

New in version 1.2.38

Connecting MCP servers has never been this easy

GitHub, a database, a browser or a server of your own: until now, you had to configure every server in every agent. Here you add it once to your space and all your agents get access to it.

Until now

By hand, agent by agent

  • One file per client, each with its own format: TOML in Codex, JSON with different keys in Claude and VS Code.
  • The same server, copied by hand into each one.
  • Tokens written into those files.
  • Restarting the agent so it loads the change.

With Agentic Workspace

One place for all your agents

MCP servers in Settings: docs-equipo remote over HTTP, and github and tienda-db local, each with its switch and the spaces where it is on
  • One server, all your agents: add it to the space and Codex, Claude Code and other ACP agents get it, without configuring it in each one.
  • A form for local servers (npx, uvx, docker) or remote ones over HTTP, or Import JSON with what you already have in Claude Desktop, Claude Code, Cursor or VS Code.
  • Test connection shows you its tools before you use it.
  • Turn it on in each space without restarting anything: it applies from the next message.
  • Tokens encrypted in the system's secure storage, never in the database or the backups.

Add it once. Every agent uses it.

No more configuring each server in each agent: add it to your space and any agent you pick for its tasks has access to it. Switch from Codex to Claude Code and its tools are still there.

Codex and Claude Code receive them; other ACP agents get the local ones always and the remote ones if they support them. Every tool they use shows up in the conversation and, with Claude Code, asks for permission with the usual card.

Community

A member of the .agent community

Agentic Workspace is part of the .agent community, the open community of the people building the agentic web: open standards, open work streams and a public map of its members.

Member card: Agentic Workspace × .agent community, at agentcommunity.org/m/agentic-workspace

Member since 2026

Our space on the community map presents Agentic Workspace as a ready-made supervision layer for any agent that speaks ACP: permission prompts, change review and session history, with several agents running at once.

Visit our space in the community ↗
The Agentic Web, the periodic table of agents, with Agentic Workspace highlighted
The Agentic Web, the periodic table of agents. See the interactive poster ↗

Download

Install Agentic Workspace

Version 1.2.39 Free for personal and commercial use. Pick your system: everything you need is included.

Windows

Windows 10 and 11 · .exe installer · x64 · Version 1.2.39

  1. Download the installer Download .exe (195 MB)
  2. Open it and install

    It shows the license, lets you choose the folder and adds a shortcut to the Start menu.

Seeing “Windows protected your PC”? Signing the installer is in progress; in the meantime, click More info and then Run anyway.

To update, open the new version's installer over the existing installation.

Coming soon

macOS

macOS 12 or later · Apple Silicon and Intel · .dmg

The Mac version is on its way. You don't need to do anything: when it is ready, you will find it here.

  • The same features as on Windows and Linux.
  • For Apple Silicon and Intel, from macOS 12.
  • The download will appear right here, on this card.

In preparation

Debian and Ubuntu

Debian 12+ · Ubuntu 22.04+ · x64

  1. Download the package and install it

    Run them in this order, one at a time and in the same terminal:

    curl -fLO https://versions.agenticworkspace.app/apt/pool/main/a/agentic-workspace/agentic-workspace_1.2.39_amd64.deb
    sudo apt install ./agentic-workspace_1.2.39_amd64.deb

The package adds the repository: updates arrive with sudo apt update && sudo apt upgrade.

Arch Linux

And derivatives · PKGBUILD · x64

  1. Download the package recipe Download recipe (PKGBUILD)
  2. Build and install it from the download folder

    Run them in this order, one at a time and in the same terminal:

    rm -rf /tmp/awb && mkdir /tmp/awb
    tar -xzf agentic-workspace-bin-1.2.39.tar.gz -C /tmp/awb --strip-components=1
    cd /tmp/awb
    makepkg -si

makepkg downloads the app, checks its SHA-256 and installs it with pacman.

Verify the downloads

SHA-256 of each file:

9faa7d6c6366497451d5d12e7e5fb6b75e7bf04781fc526139070fddc163def4  agentic-workspace_1.2.39_amd64.deb
f448c70357ff9bfe4de4805d01f09f9e232344500043d9309c1eac6e88b5f6a3  agentic-workspace-1.2.39-linux-x64.tar.gz
cf43b572f9fb46cbe4ad532fa1e68ef654e1564c30d6b7db2c3b5923a3ec772d  agentic-workspace-bin-1.2.39.tar.gz
bfa18397816c9ef715b56b794af5bccfa75d0a1785c166027c2ed2bdea8cea00  agentic-workspace-1.2.39-win-x64.exe

On Windows, compute the checksum in PowerShell from the download folder and compare it:

Get-FileHash .\agentic-workspace-1.2.39-win-x64.exe -Algorithm SHA256

The apt repository is signed with the key agentic-workspace.gpg, whose fingerprint is:

F633993803982F0A9147922E1A3EAC8B1EAAC5D8

Frequently asked questions

Good to know before you install

Is it really free? How is the project sustained?

Yes. You can use Agentic Workspace for free on any number of computers, for personal or commercial purposes, including inside companies. All current features are free. In the future we may add optional paid features; the version you already installed always keeps the free license you got it with.

Do I need a new account or another AI subscription?

No. Agentic Workspace doesn't ask you to sign up: you use the accounts you already have, your Claude Code or Codex sign-in, or your API key. You pay your AI provider the same as you would using it from the terminal.

Can I connect MCP servers?

Yes, since version 1.2.38. Register them in Settings → MCP servers, local with a command or remote over HTTP, or import them from the JSON of Claude Desktop, Claude Code, Cursor or VS Code. Then turn them on in your space: every agent in that space gets access to them, without configuring them in each one. More details in MCP servers.

Where does my data go?

Your history, backups and keys stay on your computer. There are no middleman servers: each agent talks directly to its own AI provider, just as it would from the terminal.

Which systems does it run on?

Windows 10 and 11, Debian 12 or later, Ubuntu 22.04 or later and Arch Linux, on x64 computers. The macOS version is coming soon.

Which languages is it available in?

The app is available in English and Spanish. By default it follows your system language, and you can change it at any time in Settings.

Questions or need help? Write to us at hello@agenticworkspace.app