> For the complete documentation index, see [llms.txt](https://notara-1.gitbook.io/notara-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://notara-1.gitbook.io/notara-docs/byok/spend-limit.md).

# Spend limit

A monthly spend limit stops a runaway agent from running up your model bill. It applies to the model API key you save in Notara.

## What it covers

The limit applies only when the Notara agent runs on your workspace's own model key. These are paid plan features:

* Slack replies
* Scheduled runs
* Webhook runs

MCP clients (Claude Code, Cursor and others) use their own model and are not affected. Free has no hosted AI, so the limit does not apply on Free.

## The default limit

Every new key starts with a monthly limit of **US$100 of estimated model spend**.

An admin can change the limit (US$1 to US$100,000) or remove it:

1. Go to **Connections → Slack → Model & API key**.
2. Find the **Monthly spend limit** card on Models.
3. Enter a new limit and save, or remove the limit.

## How spend is estimated

Notara estimates spend from token usage at list prices. For models without a list price in Notara, Notara uses a conservative (high) estimate, so the limit may trip before your provider bill reaches it.

Your provider bill is always the source of truth. Notara never charges for model usage.

## Alerts

Notara alerts you at 50%, 80% and 100% of the limit:

* A Slack DM to workspace admins
* An email to the owner
* A banner on **Usage** and **Models** while the agent is paused

Each threshold alerts once per month per limit. Changing the limit re-arms the alerts.

## When you reach the limit

At 100%, Slack replies and scheduled and webhook runs pause until one of these happens:

* The 1st of the next month (UTC) arrives
* An admin raises or removes the limit

People who message the agent in Slack while it is paused get a reply saying so.

## Per-run token stop

Each single run also stops at **500,000 tokens** (input + output) and says it stopped. If you need more, ask it to continue. This catches loops fast.

## The month

The limit runs on the UTC calendar month.