Cloud workspaces for Claude Code

One workspace. Your team and your agents.

People and AI agents build software together in persistent cloud sessions. Share the conversation, the computer and the repositories, and keep work moving when your laptop closes.

Start a session See how it works

Claude Code first · In the browser, on desktop and phone · Hosted by Kojo

IllustrativeA made-up session drawn with the app's own labels. Not a real customer session.

Fix flaky checkoutNeeds you acme/web

Here now: Ana Ruiz, Ben Kim (typing in terminal main) and the agent

Ana Ruiz Owner Ben Kim Reviewer ci-bot Automation

Chat
Activity
Changes
Terminal (Ben Kim is here)
Browser
Members

Chat

Ana Ruiz Checkout fails on Safari about one run in five. Find out why.

Claude Code Reproduced it in the browser: the cart hook races the price request. Fixed and pushed. One step needs a yes first.

Your approval is needed

Run a command

pnpm db:reset

Resets the test database before the checkout tests run again.

Shown to Ben Kim, a reviewer, while Ana works on something else.

ci-bot Message queued: Add a regression test for the race.

Terminal · main

Ben Kim is typing

$ pnpm test checkout --repeat 20
✓ checkout.spec.ts (20)
  Tests  20 passed
$

Activity

Ran tests1 failed, 13 passed
Editedsrc/cart/useCart.ts
Ran tests20 passed
Pushedagent/7c1e/web-main
Approval neededRun a command
Message queuedby ci-bot

Run: Waiting for approvalComputer awake · sleeps after 15 idle minutes, memory and processes kept

Why Kojo

Build together.

Your teammates and your agent share one conversation, one computer and its terminals. You see who is here and where, and an approval can come from someone other than the person who asked.

Keep the work alive.

Each session keeps its own computer. It sleeps when idle and wakes with memory and processes intact; the agent pushes its branch as it works, and the next run picks up the saved conversation.

Let agents operate Kojo. Preview

Agents join a session as identified members through scoped API keys, and their work shows alongside the team's.

The same session. Different kinds of teammates.

Invite a teammate to pair and work in the same conversation, the same terminals and the same previews. Everyone sees what happened, who asked for it, and what is waiting on whom.

Here nowThe faces of everyone in the session, the agent included while it works, and a cursor in each person's colour on the tab or terminal they are on.
Invite someone to pairAdds a teammate to the session as an editor and emails them the link. “Copy link” shares the address; membership still decides who can open it.
One keyboard at a timeShared terminals keep running across reloads. Take a terminal's keyboard, ask for it or hand it over; the agent is told to leave a terminal someone drives alone.
Approvals from anyone allowedWhen the agent asks before a step, someone other than the person who started the task can approve or deny it. If nobody answers in time, the step is denied, never quietly allowed.

Ana Ruiz Owner Ben Kim Reviewer ci-bot Automation

Roles decide who reads, approves, sends and manages a session: viewer, reviewer, editor, owner, and automation for API keys and apps. Preview Session roles →

Close the tab. Don't lose the work.

Runs have limits; your work shouldn't disappear with them. A session keeps its computer, its branch and its conversation, so the next run picks up the task. The computer sleeps after 15 idle minutes and wakes when you need it.

WorkingThe agent edits, tests and pushes its branch every few minutes. Teammates can join from any browser.
AsleepAfter 15 idle minutes the computer sleeps with its memory and running processes kept. It uses no compute while asleep.
ResumedA message, a preview or a terminal wakes it where it stopped. If the computer is gone, the next run starts a new one from the pushed branch and the saved conversation.

Sleep and wake keep memory and processes. A new computer restores code from Git and the conversation from its transcript, not running processes. See the exact guarantees →

Not just a place to run agents. A workspace agents can use. Preview

Kojo's API covers what the app does: sessions, messages, approvals, activity, commands, ports and preview links, behind scoped API keys. An agent joins a session as an automation member, finds the operations it may call in a catalog, and its work shows in the session under its own name.

Agent / API Illustrative

$ curl -X POST "$KOJO_API/coding/sessions" \
    -H "Authorization: Bearer $KOJO_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "repositories": [{"full_name": "acme/web", "mode": "pr"}],
      "first_turn": "Make the checkout test pass on Safari."
    }'

The same session in Kojo

Member: ci-bot Automation
Activity: Message sent · Run started
Session: Fix flaky checkout
May: read and send messages, not approve or open a shell

Preview: built and tested, not yet proven end to end in production; it may change. Creating API keys from the app is not built yet. API details →

From a session to a pull request.

You pick the repositories a session works in; a model never picks them. The agent works on its own branch and pushes as it goes. When the work is ready, open a draft pull request from the session, described by what was asked and what the agent answered, with its screenshots. A merge happens only when a person approves that merge.

One task across several repositories, published as linked draft pull requests opened in order and retried if one fails. Preview

From repository to working session.

Sign in.Create or join your team.
Connect GitHub.Choose the repositories Kojo is allowed to access.
Start a session.Add credit, pick the repositories and give Claude Code a task.
Collaborate and review.Watch the work, answer approvals, and review the pull request.

Pay for execution, not empty time.

Kojo meters what your sessions use and shows who used it. Awake computer time is charged by the second; a sleeping computer uses no compute. Bring your Claude subscription or API key and that model usage is billed by Claude, not by Kojo. The workspace subscription is $19/month per seat plus metered usage.

Free accounts are temporary. A free account and its workspace are deleted every 3 days; add a card and a seat to keep yours. Read the exact rules before you sign up.

Understand pricing →

Built for a specific workflow. Expanding deliberately.

Claude Code is the supported harness. Open models run through it as fallback routes; Codex and Cursor are not supported.
Kojo runs in the browser, on desktop and phone. There is no native app, no local worktree mode and no push notifications: a session page is live while it is open.
One active run per session. Parallel work uses more sessions.
Hosted service. Not self-hosted, not open source.
Kojo sleeps when nobody uses it. The first visit after a quiet spell shows a waiting page while it wakes, and a tab left idle rests until you continue.
Agents through the API, multi-repository pull requests and monitors are Preview. Preview See what works today.

FAQ

What is a workspace vs. a session vs. a run?: A workspace is your team's account: its people, repositories, credit and settings. A session is one shared conversation with its repositories and its computer. A run is one bounded stretch of agent work inside a session, up to two hours; the next run continues the session.
Does Kojo keep working when I close the browser?: Yes, while the agent is working. After 15 idle minutes the computer sleeps, and the next message wakes it.
Can a teammate approve the agent's request?: Yes: someone else in the session can approve or deny it from their own browser. If nobody answers in time, the step is denied, never quietly allowed.
Can I use my own Claude subscription or API key?: Yes. Model usage on your own subscription or key is billed by Claude, not by Kojo. If it hits a usage limit, the run can move to the next model route you set, in the same conversation.
What happens to my work if the computer is gone?: The branch is pushed every few minutes and the conversation is saved. The next run starts a new computer from the pushed branch and resumes the conversation; running processes do not come back.
What does a sleeping computer cost?: It uses no compute. Its asleep time is metered at the asleep rate, not the awake rate; awake time is charged by the second. What is metered.
What does Kojo store or log?: We keep operational records needed to run, secure and bill your workspaces. We do not provide employee productivity scores. Details.
Can the agent merge to main?: Not on its own. It opens and updates pull requests; a merge happens only after a person approves that merge, every time.
Can I take my work elsewhere?: Your code lives in your own GitHub repositories, on the branches the agent pushes. Exporting a session's conversation is not built yet.

Give your team, and your agents, a place to work.

Start a session See how Kojo works →