> ## Documentation Index
> Fetch the complete documentation index at: https://docs.attensira.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Credits

> What a credit is, which agent actions spend one, and what happens as a workspace balance falls to zero.

A credit is the unit Attensira uses to meter **agent work** — the actions where a language model does something on your behalf rather than simply returning data you already have. Your workspace holds a credit balance. Agent actions draw it down. Reading numbers you have already collected does not.

## What spends a credit

Three kinds of action spend credits, whether you start them in the app or over [MCP](/mcp/overview):

* **Chats.** A conversation with the Attensira agent. Each turn is agent work.
* **Automation runs.** Every execution of an automation, whether you trigger it by hand or it runs on its schedule.
* **Ask calls.** The `ask` tool, which puts a question to the agent and returns an answer or a running session.

Over MCP, exactly three tools cost credits: `add_prompts`, `run_automation`, and `ask`. The read-only tools — `search`, `get_analytics`, `get_prompt`, `get_page`, `get_session`, `get_account`, `list_automations`, `get_automation` — do not.

<Warning>
  Opening a dashboard, filtering a chart, or exporting numbers you already have does not spend credits. If you are watching your balance move while doing only those things, something else in the workspace is running — check your automation schedules.
</Warning>

## How a credit is consumed

Consumption is per action, not per second and not per seat. Two people in the same workspace draw from one shared balance, so a busy automation schedule and a busy chat habit compete for the same pool.

Cost scales with how much work the action actually does. A short chat turn costs less than an automation that fans out across many prompts and pages. This is why an automation you run by hand can feel cheap while the same automation on an hourly schedule is not.

<Tip>
  When you trigger an automation over MCP, pass an `idempotency_key` to `run_automation`. It is sent upstream as the `Idempotency-Key` header, so a retried call does not produce a second billed run.
</Tip>

## What happens as the balance falls

Nothing degrades quietly. Your balance is shown in the workspace, and the agent surfaces remain fully functional right up to the point where a request cannot be covered.

As the balance gets low, the practical effect is that long-running agent work becomes the first thing at risk: an automation that fans out widely needs more headroom than a single chat turn, so it is the kind of action most likely to be the one that cannot start.

## What a workspace sees at zero

At zero, credit-consuming actions stop. Starting a chat, running an automation, calling `ask`, or adding prompts through the agent will be refused rather than partially performed. You are told the workspace is out of credits; you do not get a half-finished automation run.

Everything that does not consume credits keeps working. You can still open the workspace, read every metric already collected, filter and compare, and use the read-only MCP tools. Scheduled automations do not silently disappear — they remain configured and resume once the balance is restored.

Restore a balance by topping up or changing plan in your workspace settings. See [Plans](/account/plans) for what each plan includes.