Billing & Credits

Billing and Credits

AgentContainer uses a simple prepaid credit model. There are no subscriptions, no tiers, and no monthly commitments. You load credits, run tasks, and pay only for the container time you use. AI token costs are billed directly by your chosen provider (Anthropic, OpenRouter, or OpenAI) — you bring your own API key.

Prepaid Credits

Credits are stored as a USD balance on your account. To add credits, open the Billing page in the dashboard and load any amount (minimum $5) via Stripe checkout. Your balance is available immediately after payment.

There is no expiration on credits. Your balance carries forward indefinitely until you use it.

What Tasks Cost

AgentContainer charges for one thing: container time. The cost is calculated after the task completes, not estimated upfront.

• Container time — a fixed per-minute rate of $0.05/min for the isolated environment your agent runs in.
• AI tokens (BYOK) — billed directly by your provider. You configure your own API key (Anthropic, OpenRouter, or OpenAI) in Settings > API Keys. AgentContainer does not proxy, markup, or meter token usage.

Example cost breakdown

A 12-minute task:

Token costs depend on the model and provider you choose. Check your provider's pricing page for current rates.

The Reservation Model

When you dispatch a task, AgentContainer reserves $0.50 from your balance. This reservation ensures the platform can cover the minimum cost of spinning up a container and running the agent. If your balance is below $0.50, the task cannot start.

After the task completes, the reservation is replaced by the actual cost. If the task cost $0.60, your balance is debited $0.60 (not $0.50). If the actual cost exceeds the reservation, the full actual cost is still deducted — a negative balance is allowed for the completing run only. This means a task that is already running will always finish, even if it costs more than expected.

Mid-Run Balance Depletion

If your balance runs out while a task is still running, the task is paused and its status is set back to pending. You do not lose progress. When you top up your credits, any pending tasks automatically restart and pick up where they left off.

This design means you never lose work due to an empty balance. The worst case is a temporary pause until you add more credits.

Viewing Usage

The Billing page in the dashboard shows your current balance and a full transaction history, including both top-ups and task charges. Each completed task displays the container minutes used.

The same data is available programmatically on the task object via the API. See the Tasks and Output docs for the full task schema.

{
  "taskId": "tsk_abc123",
  "status": "finished",
  "costUsd": 0.60,
  "containerMinutes": 12
}

Cost Optimization Tips

A few ways to keep your costs predictable and low:

• Set maxRuntimeSeconds — put a ceiling on task duration so a runaway agent does not consume your entire balance. If the task exceeds this limit, it is automatically stopped.
• Choose the right model — smaller models like claude-haiku cost significantly less per token than larger ones. Use a capable-enough model for the job, not the most powerful one by default.
• Write specific task descriptions — vague prompts lead to more agent exploration, more tool calls, and more tokens. A clear, scoped description keeps costs down.
• Disable internet access when not needed — setting internetAccess: false does not save money directly, but it avoids unnecessary network calls that can slow the agent down and increase container time.

{
  "description": "Parse the uploaded CSV and return summary statistics as JSON.",
  "model": "claude-haiku",
  "maxRuntimeSeconds": 300,
  "internetAccess": false
}

For more on task configuration options, see the Quickstart guide or the full Tasks and Output reference.