Can an AI agent sign up for its own cloud computer?

TL;DR

Yes. On Orgo, an AI agent can create its own account with one unauthenticated call to POST /api/v1/agents. No email, no browser, and no person present. The call returns an account-wide API key, a workspace, a claim link, and $2 of starting credit, about 36 running hours of a Medium Linux computer (2 vCPU, 8 GB RAM, 40 GB disk) at $0.055/hr (pricing). The agent launches that computer, uses it over the normal Orgo API, asks its person for more credit through a Checkout link, and hands over a claim link so a person can take the account over. An unclaimed account can hold one computer, can't use AI until it has bought $5 of credit, and expires 72 hours after signup unless it has added $5 in all or a person claims it.

What is agent self-signup?

Agent self-signup is an account flow where the AI agent, not a human, is the first owner of a cloud account. The agent calls a signup endpoint, receives its own API key and a small prepaid balance, and starts working right away. A person joins later by claiming the account.

On Orgo the pieces are:

• Signup: POST https://www.orgo.ai/api/v1/agents, no Authorization header, optional name (1 to 80 characters) and contact_email.
• Key: an sk_live_ account-wide key, shown once.
• Computer: a full Linux desktop with a browser, driven over the HTTP API (bash, screenshot, click, type) and VNC.
• Claim: a one-time link plus a 9-character code (XXX-XXX-XXX) a person uses to take over the account. The code skips look-alike characters (0, O, 1, I, L), and case and dashes don't matter.

Orgo provides the computer, not the agent. Bring Claude Computer Use, OpenAI's CUA, Hermes Agent, OpenClaw, or your own loop. For the basics of agents driving a desktop, see What is computer use?

Why would an agent need its own account?

Most agent setups stall at the same step: a human has to sign up, add a card, create a key, and paste it somewhere. That is fine for a team rolling out one integration. It is friction when the agent is the one that discovers it needs a computer mid-task.

Self-signup removes that wait. The agent gets a working machine first, and the human decisions (paying more, owning the account) happen later through links the agent sends. Typical cases:

• A coding or research agent that hits a task needing a real browser or desktop, not just a shell.
• An agent framework that wants a zero-config "try it" path for new users.
• Evaluations and demos where a throwaway computer for a few hours is enough.

Why a full computer instead of a browser tab: some tasks need desktop apps, file dialogs, or a persistent userland. See Cloud desktops vs headless browsers.

How do the ways to give an agent Orgo access compare?

Rule of thumb: if a person is around and will own the work, use MCP or device approval so the agent works inside their plan. If no one is around yet, self-signup gets the agent a computer now and a person can claim it later.

How does an agent sign up and launch a computer?

Five steps (Agent signup guide).

1. Sign up. Send an Idempotency-Key so a dropped connection never creates a second account.

IDEMPOTENCY_KEY=$(uuidgen)
 
curl -s -X POST https://www.orgo.ai/api/v1/agents \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $IDEMPOTENCY_KEY" \
  -d '{"name": "my-agent"}' > signup.json
 
export ORGO_API_KEY=$(jq -r .api_key signup.json)
export WORKSPACE_ID=$(jq -r .account.workspace_id signup.json)

The 201 response includes api_key, account (with workspace_id, status: "unclaimed", and expires_at), claim (with url, code, and its own expires_at), starting_credit_usd: 2, limits, and next hints for each following call. Save api_key before anything else; it is shown once.

2. Launch a computer by the hour.

LAUNCH_KEY=$(uuidgen)
curl -s -X POST https://www.orgo.ai/api/computers \
  -H "Authorization: Bearer $ORGO_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $LAUNCH_KEY" \
  -d "{\"workspace_id\": \"$WORKSPACE_ID\", \"name\": \"my-computer\"}" > computer.json
 
export COMPUTER_ID=$(jq -r .id computer.json)

An agent account's create is hourly by default and runs a Medium Linux computer at $0.055/hr while running and $0.0027/hr while stopped, billed by the second from credit (as of November 2026, pricing). Retrying with the same Idempotency-Key and body returns the original launch instead of a second computer.

3. Use it. Every computer action in the API works: bash, screenshot, click, type, and the rest.

curl -X POST https://www.orgo.ai/api/computers/$COMPUTER_ID/bash \
  -H "Authorization: Bearer $ORGO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"command": "uname -a"}'

4. Add credit. The starting credit runs computers only. To keep a computer running, use AI, or unlock more computers, the agent requests a Checkout link and sends it to its person:

curl -s -X POST https://www.orgo.ai/api/credits/checkout \
  -H "Authorization: Bearer $ORGO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"amount_cents": 1000}' | jq -r .url

amount_cents is any whole-dollar amount from $5 to $10,000 (500 to 1000000). Once the account has bought $5 of credit in total, with none of it refunded or disputed, it can hold up to 5 computers and 5 workspaces, AI unlocks, and the account stops expiring. The signup response's next.add_credits always names the right call for the environment you're in.

5. Hand over the claim link. Give the person claim.url and show them claim.code so they can confirm it is the right account. They sign in or sign up and accept. A person with only the code can enter it at https://www.orgo.ai/claim.

What can an unclaimed agent account do, and what can't it?

Refusals come back with a machine-readable code, which matters because the caller is an agent, not a person reading an error page:

What happens when the agent's account is claimed or expires?

When a person claims it:

• They become an admin of every workspace the agent owns and manage them from their own sign-in.
• The unclaimed limits and the expiry stop applying.
• The signup key keeps working, and the agent's credit keeps paying for its computers. Ownership and billing stay with the agent's account.
• Orgo's own computer keys on the agent's computers (the ORGO_API_KEY a template may put there) are revoked and those computers' VNC passwords change, because the claimer can now act as root on them.
• Claiming adds no credit. One person can claim up to 3 agent accounts; a fourth returns 403 claim_limit.

A claim link works for 7 days and only once. The person can also decline, which kills the link. Refreshing it (POST /api/v1/agents/claim) invalidates the previous link and code, and GET /api/v1/agents/claim reports whether the account has been claimed yet. Both need the account-wide signup key; a workspace-scoped key gets 403 workspace_scoped_key (see Scoped API keys for AI agents).

Errors a claim can hit:

When nobody claims it, the expiry job (it runs every 10 minutes) picks up every unclaimed account whose 72 hours are up and that hasn't added $5 in all:

1. The account is marked expired and every API key on it is deleted. Running computers are then stopped through the normal stop path, which settles hourly billing.
2. During a 2-day grace period, a person can still claim it with an unexpired claim link, get the stopped computers back, and issue the agent a new key.
3. After the grace period, the claim link is cleared and the account's computers are deleted, unless it has added $5 in all by then. The account record and any credit it bought stay on the ledger.

An account that has added $5 in all doesn't expire; a refund or an open dispute counts against that. Below it, plan on a claim. Because the key dies at expiry, an agent that may need its person later should send the claim link early and refresh it before the 72 hours are up.

Is agent self-signup safe to expose?

The guardrails, all enforced in the signup and billing code:

• Idempotency. Idempotency-Key (8 to 255 printable characters) replays the first 201 with Idempotent-Replayed: true. Stored answers last 24 hours and only replay to the same address. Reusing a key with a different body returns 422 idempotency_key_reused.
• Rate limits. Per address (an IPv6 address counts as its /64): 3 an hour, 10 a day. Per network (IPv4 /24 or IPv6 /48): 20 a day. All signups together: 500 a day. Over a limit returns 429 rate_limited with Retry-After. Refused requests don't count.
• Prepaid only. Unclaimed accounts can't run a plan computer or a trial computer, so spend past the $2 is capped by credit someone actually paid.
• No AI on starting credit. Model calls need $5 of purchased credit, so anonymous signups can't run up model spend.
• A small shared pool. Unpaid agent accounts together get 25 running places, so signups can't fill the places other accounts share.
• Feature gates. Invites, client accounts, invoicing, and buying a Mac or iPhone wait for a human claim.
• Per-environment switch. Where agent signup is off, the endpoint returns 404 agent_signup_disabled and points to device sign-in (POST /api/cli/auth/start).

When should you not use agent self-signup?

• You already have an Orgo account and a plan. Use MCP or device approval so the agent works inside your workspace, plan, and sizes (For agents).
• You need Windows, larger sizes, or many computers on day one. Unclaimed accounts run Medium Linux only. Windows is included on Scale. On Starter or Growth it needs a Windows computer add-on, $47/mo in all ($15 for the license plus a Medium's memory, vCPU and disk), or the create returns 403 WINDOWS_REQUIRES_SCALE (pricing). For fleets, see Scale to thousands.
• You bill clients for their computers. Client accounts and invoicing are unavailable until claimed; use a normal account with the Invoicing guide.
• The work must survive past 72 hours with no human around. Add $5 in all or plan for a claim, or the key stops working at expiry.

FAQ

Does an AI agent need an email address to sign up for Orgo? No. POST /api/v1/agents needs no Authorization header and no body. contact_email is optional.

How much credit does an agent account start with? $2 of starting credit, about 36 running hours of a Medium Linux computer. It runs computers only, not AI.

How many computers can an unclaimed agent account have? One, running or stopped. A second create returns 403 HOURLY_LIMIT. After $5 of purchased credit, up to 5.

How long does an unclaimed agent account last? 72 hours after signup, unless it has added $5 in all. Then its keys are deleted and its computers stop. A person can still claim it for 2 more days before the computers are deleted.

Does the agent lose access when a person claims the account? No. The signup key keeps working and its credit keeps paying for its computers. The person becomes an admin of every workspace the agent owns.

What if the signup request times out? Retry with the same Idempotency-Key and the same body within 24 hours. You get the original 201 back, not a second account.

Can a self-signup account use Orgo's hosted AI? Yes, once it has bought $5 of credit in total with no refund or dispute. The smallest Checkout top-up is $5, so any Checkout purchase unlocks it. Then the OpenAI-compatible POST /api/v1/chat/completions endpoint can drive the computer (For agents). Before that, AI calls are refused with 402.

What does a Medium cost per hour? $0.055/hr while running and $0.0027/hr while stopped, billed by the second (as of November 2026, pricing).

Related

• What is computer use?
• Cloud desktops vs headless browsers
• Scoped API keys for AI agents
• Anthropic Computer Use setup on Orgo
• Orgo docs: Let an agent sign up on its own
• Orgo docs: For agents

Sources

• Orgo docs: Let an agent sign up on its own (read 2026-11-05)
• Orgo docs: For agents (read 2026-11-05)
• Orgo pricing (fetched 2026-11-05)
• Orgo agent guide: auth.md (fetched 2026-11-05)