@ada can you pull the top 20 support tickets this week and group them by product area?
Your AI agent's own computer. Always on.
A cloud desktop with Claude Code, Codex or Hermes ready to work. Use your own AI plan, or ours.
Talk to it from wherever you already are.
The agent lives on its computer, so the conversation does too. Start a task on your phone, check the result in the browser, and let your AI tools hand it work over MCP.
Grouped 20 tickets into 4 areas. Checkout errors are up 3× since Tuesday's release.
📎 tickets-w45.csv · 18 KB$ npx agb connect Signing in with GitHub… done Found Claude Code, Cursor, Codex ✓ Added agb to 3 apps $ claude › On my agb computer, open the staging site in Chrome and screenshot every page with a console error. ✓ agb.browser 12 pages checked, 2 with errors ✓ agb.files saved to ~/shots/staging-errors/
More than a container. A desktop that stays put.
Each agent gets a full cloud desktop with a browser, a file system and a GUI. Files, logins and installed tools are still there tomorrow.
Watch it work, take over anytime
Stream the desktop in your browser. Grab the mouse when a site needs your sign-in, then hand it back.
Online around the clock
No idle shutdown. Scheduled jobs run at 3am whether or not anyone is logged in.
Reset to clean in one click
Something broke? Restore the computer to its original image and keep the agent's settings.
- Compute
- 4 vCPU · 8 GiB8 vCPU · 16 GiB on Max
- Storage
- 80 GiB system diskUpgrade to 120 GiB anytime
- Desktop
- Full GUI with Chrome, terminal and file manager
- Agents
- Claude Code, Codex, Hermes, Kimi Code, QoderPick one at setup, switch later
- Models
- Your own Claude or ChatGPT plan, or an API keyModel credits included on Pro and above
- Uptime
- 24/7 · no idle sleep
- Region
- Singapore · ap-southeast-1
No API keys. No surprise bills.
Already pay for Claude or ChatGPT? Sign in with it on any plan. Or let agb handle models: Pro and above come with monthly credits. Pick a tier, and agb routes to the right model for it. When credits run low, you get a heads-up at 80% and 95%. When they run out, the agent switches to a lighter model and keeps working.
Same credits work in chat, scheduled tasks and the OpenAI-compatible API.
Running in about the time it takes to make coffee.
Sign in and pick an agent
Use Google or GitHub. Choose Claude Code, Codex or Hermes. You can switch later.
agent: claude-code · tier: autoYour computer starts
agb sets up a desktop in Singapore with the agent installed and signed in to models.
ada-workstation · onlineHand it work
Chat in the browser, connect your tools with one command, or schedule something to run every morning.
$ npx agb connectGive it a routine, then get on with your day.
Schedule work in plain language. The agent runs it on its own computer, saves the output and pings you where you chat. If it needs a decision, it waits for you.
Move your local agent to the cloud without starting over.
The agb desktop app copies your agent's settings, skills and past sessions from your laptop to its new computer. Pick up a conversation exactly where you left it.
- ~/.claude/settings.json
- 12 skills · 3 MCP servers
- 48 sessions
- Settings applied
- Skills and MCP servers installed
- Sessions ready to resume
One agent per teammate, one place to manage them.
Admins hand out computers, set budgets and see what every agent did. Each person's agent runs on its own isolated computer.
Single sign-on
SAML, Microsoft Entra ID, Google Workspace and Lark. Roles for admins and members.
Shared credit pools
Top up once for the team, set a limit per agent, and get alerts before anyone runs dry.
Execution logs
Every task, every scheduled run, the model tier it used and the credits it spent.
Guardrails
Sensitive-data filtering, prompt-injection checks and a review step for new skills.
Network policy
Allow or block outbound sites per team, and connect to your private network.
Reassign in a click
When someone leaves, move their computer to a new teammate or reset it clean.
Start with the computer. Add models when you need them.
Every plan starts with a free 2-hour trial. No credit card.
An always-on computer for your agent. Bring the AI plan you already pay for.
- Run Claude Code, Codex or Hermes on a desktop that never sleeps
- Sign in with your Claude or ChatGPT plan, or paste an API key
- Watch the screen live and take over anytime
- Check in from the browser or your phone
- Hand it work from Claude Code, Cursor and Codex over MCP
- Model credits not included
Models included. Give your agent routines and let it work while you sleep.
- Use top models with no API key or AI subscription
- Schedule routines that run every morning or every night
- Mention your agent in Telegram, Slack or Lark
- Picks the right model for each task, and keeps going on a lighter one if credits run low
- Move your local agent's settings and sessions to the cloud
For long builds, many browser tabs and agents that work all day.
- Twice the computer for heavy builds and test suites
- Enough credits for agents that run all day
- Add your own key as a backup when credits run out
- Priority support from the agb team
An agent for every teammate, managed in one place.
- Sign in with your company SSO
- Share one credit pool and set a limit per agent
- See every task each agent ran and what it cost
- Guardrails and network rules for the whole team
- One invoice in SGD or USD
Compare all plans
| Starter$19.99 | Pro$69 | Max$149 | Team$79 / member | |
|---|---|---|---|---|
| Your computer | ||||
| Always-on cloud desktop, no idle sleep | ✓ | ✓ | ✓ | ✓ |
| Watch live and take over | ✓ | ✓ | ✓ | ✓ |
| Reset to a clean image | ✓ | ✓ | ✓ | ✓ |
| Size | 4 vCPU · 8 GiB | 4 vCPU · 8 GiB | 8 vCPU · 16 GiB | 4 vCPU · 8 GiB |
| Disk | 80 GiB | 80 GiB | 120 GiB | 80 GiB |
| Agents and models | ||||
| Claude Code, Codex, Hermes | ✓ | ✓ | ✓ | ✓ |
| Use your own AI plan or API key | ✓ | ✓ | ✓ | ✓ |
| Monthly model credits | — | 1,600 | 4,000 | Shared pool |
| Smart model routing and fallback | — | ✓ | ✓ | ✓ |
| Ways in | ||||
| Browser and phone | ✓ | ✓ | ✓ | ✓ |
| MCP for Claude Code, Cursor, Codex | ✓ | ✓ | ✓ | ✓ |
| Telegram, Slack, Lark | — | ✓ | ✓ | ✓ |
| Automation | ||||
| Scheduled routines | — | ✓ | ✓ | ✓ |
| Move your local agent to the cloud | — | ✓ | ✓ | ✓ |
| Team and security | ||||
| Company SSO and admin roles | — | — | — | ✓ |
| Execution logs, guardrails, network rules | — | — | — | ✓ |
| Support | Priority | Priority | ||
Straight answers.
Still deciding? Read the docs or ask our team on Telegram.
Do I need my own AI subscription or API key?
On Starter, yes: your agent signs in with your Claude or ChatGPT plan, or with an API key you paste in. Pro, Max and Team include monthly model credits, so you don't need either. You can still add your own key on any plan.
What happens when I close my laptop?
Nothing changes for the agent. It runs on its own computer in Singapore, so tasks keep going and scheduled jobs still run. Open the app later and the session is where you left it.
How is this different from Claude Code on the web or GitHub Codespaces?
Those give you a temporary environment for one tool. agb gives your agent a persistent desktop with a GUI browser, runs 24/7, works with several agents, and can include model credits. You can reach it from chat apps and your other AI tools too.
Can I watch what the agent is doing?
Yes. Open the desktop view to see the screen live. You can take over the mouse and keyboard at any time, for example to sign in to a site, and then hand control back.
Where is my data stored?
On your computer's disk in the Singapore region. Each computer is isolated from every other customer's. Your files stay until you reset or delete the computer.
What happens when my credits run out?
On Pro and above, the agent switches to the Lite tier and keeps working. You get alerts at 80% and 95%, and you can buy more credits or upgrade at any time.