Let agents choose their computer.
One compute pool for your agents: sandboxes, desktops, GPUs and your own machines, across every vendor.
Paste it into Claude Code, Codex or Cursor, or give your agent the skill.
Incoming jobs
- my-macyour files · free
- my-serveryour Linux box · free
- e2bbrowser · desktop
- vercelscripts · code
- aws · gcpyour cloud account
- daytonaWindows · Mac
- modalGPU
- cloudflarelight jobs
Watching your load…
They are giving you free compute. Use it.
Each provider gives a new account free compute. Connect your accounts once, and your agents get all of it in one pool.
 E2B | $100 | to start |
 Daytona | $200 | to start |
| Modal | $30 | every month |
| Vercel | 5 CPU hours | every month |
| Total | $300 | to start, then $30 and 5 CPU hours every month |
Free on Cerver too: two provider accounts, plus your own machines. That is a real pool: E2B, Modal and your Mac.
# 1. your machine joins the pool (this also installs the CLI)
curl -fsSL https://cerver.ai/install.sh | bash
# 2. add your free E2B account. Modal connects the same way.
curl -X POST https://gateway.cerver.ai/v2/account/providers \
-H "Authorization: Bearer $CERVER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"provider":"e2b","credentials":{"e2b_api_key":"e2b_..."}}'
# 3. one pool, a different computer for each task
cerver computes
cerver run --on provider_e2b "fix the failing test"
cerver run --on provider_modal "run the simulation"
cerver run --on my-mac "update the board deck"
my-mac is the name your machine shows in cerver computes. Full steps in the quickstart.
- 1Log in and export the accountsDesktop
- 2Score 2M support ticketsGPU
- 3Run the analysisSandbox
- 4Update the board deckYour laptop
Different work needs a different computer.
Hours × each provider's published hourly rate, Nov 2026. You pay the provider directly.
One vendor is one point of failure.
Every provider has outages, its own prices and work it does best. Use several through one API and you get all three on your side.
An outage doesn't stop your agents
When a provider goes down, the work moves to another one and keeps going. One vendor down no longer means your agents are down.
Pay for what each step needs
Each step runs on the cheapest computer that can do it. Your own machine is free, and the free tiers you already have get used first.
The best computer for each job
A GPU for the model, a screen for the browser, your Mac for your files. Not the one you happened to sign up for first.
One compute pool for your agents. Every vendor, every kind of computer.
Your pool is every computer your agents can use: the provider accounts you connected and your own machines. Today the agent picks a computer for each task, or moves it mid-task. Soon a router does it for you: keep a few of each kind warm, and when they are full, wait, start one on demand, or use the next best kind. Plain rules, or an agent that routes.
Incoming
0 waitingRouter
Your pool
# one policy for your whole pool pool: desktop: { on: e2b, warm: 3, max: 6 } gpu: { on: gpu-box, warm: 1, max: 2 } fast: { on: vercel, warm: 4, max: 8 } private: { on: my-mac-minis, fixed: 4 } route: agent # or: rules (by tagged need) when_full: cheapest of [wait, burst, next best kind]
- Rules or an agent. Rules match a tagged need. The router agent reads the task itself and explains every choice.
- Your own machines take the steady load. Cloud computers only cover the spikes.
- See your whole pool. Busy, idle, waiting and cost, so you size each kind to your traffic.
The right computer for each step
Desktop, GPU, sandbox or your own machine. Start each step on the one that fits, and move a session between the computers on your account without losing a step. GPUs through Modal are coming soon.
Watch and take over
See each computer's screen live. When the agent is stuck on a login or a 2FA code, click in, type it, and hand it back.
One view, every computer
Each step priced on the computer it used, rolled up per agent and per customer. A $20 cap stops at $20.