Do I need my own Anthropic or OpenAI account?

No. That is the point of credits — the agent runs on our provider account. You can still bring your own agent instead, free on every plan.

What does a task cost?

It depends on the agent and the length of the work. A Standard-tier task starts around four cents; a frontier model such as Claude Opus 5 costs more. The wallet shows the real figure for every task you run.

No. There is no expiry, no subscription, and no monthly minimum.

What happens when I run out?

Runs stop. Nothing rolls over into a bill, because credits are prepaid — so an unexpected invoice is not possible.

Can I still use my own agent?

Yes, and you can switch on every deploy. Bringing your own agent is free on every plan and is unchanged.

Which model does Claude run?

Opus 5 by default. You can pick Fable 5.1, Sonnet 5 or Haiku 4.5 per session from Profile › Agents; each is priced separately.

Free

Usage credits (pay as you go)

Run an agent without your own provider account — prepaid credits are billed per task, and never expire.

01 · CHOOSE

Your agent or ours

Asked on every deploy

02 · TOP UPPrepaid

Buy a credit pack

$10, $25 or $50

03 · RUNPer task

The agent works

Billed as it runs

04 · SEETracked

Every run, every cent

In your wallet

There are two ways to run an agent, and you choose on every deploy. Bring your own agent — Claude Code, Codex, Cursor, Copilot and the rest, on the accounts you already pay for. That path is free on every plan and has not changed.

Or run a CodeAgent agent: we hold the provider account, you hold prepaid usage credits, and each task is deducted as the agent works. There is nothing to sign up for with Anthropic or OpenAI, no second subscription, and no monthly minimum.

No provider account

The agent runs on our account. You never create, pay for, or manage a key with the model provider.

Billed per task, as it runs

A task on the Standard tier starts around four cents; a frontier model costs more. The wallet shows what each task actually cost.

Credits never expire

A pack you buy today is still there months from now. There is no subscription and no monthly minimum attached to it.

It stops at zero

When the balance reaches zero, runs stop rather than rolling into a bill. A surprise invoice is impossible because it is prepaid.

Which agents you can run

Four agents are available on credits, in two tiers. Standard covers everyday work at the lowest cost per task; Frontier is for the hardest reasoning.

Claude — Frontier

Runs Opus 5 by default. Fable 5.1, Sonnet 5 and Haiku 4.5 are selectable per session from Profile › Agents.

Codex — Frontier

OpenAI's coding agent with reasoning on, and full tool use.

DeepSeek V4 — Standard

Fast and the lowest cost per task. Also available as V4 Flash for shorter work.

Qwen3 Coder — Standard

Long-context work: large refactors and codebase-wide changes.

1
Start a deploy
The agent chooser asks how you want to run this codespace: your own agents, or CodeAgent agents.
2
Pick a CodeAgent agent
Choose the tier and the agent. Nothing to link, nothing to authenticate.
3
Top up if you need to
Checkout opens in your browser. Your balance updates in the app when you come back.
4
Watch the cost as it runs
The wallet groups spend by task, so you see what a piece of work cost rather than a wall of API calls.

NOTE

Credits and your plan are separate things. Your plan sets the app limits — how many tasks a day, how many devices, Team Spaces. Credits pay for compute on our provider accounts. Pro subscribers get $1 in credits to try a CodeAgent agent.

FAQ

Plans & pricing Supported AI agents Cloud (GitHub Codespaces)

Ready to try it?

Start free See pricing