How to control costs and read the dashboard

This page is for admins who pay for the models or share them with a team. You set limits on how much the Foundry may spend, see where the money goes, and use evals to find the cheapest model that still does the job.

Before you start#

You are signed in as an admin.
You know how your agents are logged in: a Claude subscription, an API key, a ChatGPT login or a local model.

Steps#

1. Know what the amounts mean#

The costs shown are estimates at API prices, as reported by Claude Code. Whether you pay them depends on the login:

Claude Code with a Claude subscription (Pro/Max): no per-token charges; usage counts toward your plan's limits.
Claude Code with an API key (ANTHROPIC_API_KEY): billed per token, so the amounts are real.
Codex with your ChatGPT login: recorded as $0 with the token count.
Local models: free, recorded as $0.

Budgets use these amounts either way, so they also protect your subscription limits.

2. Set the daily budget#

Open Administration → Settings → Budget & capacity. The daily budget stops new work when today's estimated spend reaches it. Paused runs continue the next day, or when you raise the budget and resume them.

On a fixed-price subscription you can turn off Enforce cost limits (cost_limits: false). Costs are still recorded, but nothing is stopped because of money. The usage limits that Claude and Codex report themselves still pause runs.

3. Cap runs and steps in a flow#

limits:
  max_cost_usd: 20      # the whole run stops at this cost
defaults:
  max_budget_usd: 5     # per agent step

A run that reaches its budget fails with a message that says so. A change to the flow counts only for a new run; a resumed run keeps the flow it started with.

4. Set limits per user#

On Administration → Users, Default limits sets three limits for every account: runs at the same time, runs per day and the daily budget in USD. Open an account and use Limits to give it its own values. An empty field means "no limit". A job over a limit waits in the queue and starts by itself when the limit allows.

5. Read the dashboard#

Open Administration → Dashboard. It shows spend today and over 30 days, the success rate, Needs a human, the Waiting card, cost per day, the By user card, results per flow and per repository, the steps where runs fail most, and eval results. An account at a limit gets an at limit mark. For a quick look, Administration → Overview has a Limits card with today's spending and the accounts at a limit.

6. Compare models with an eval#

An eval suite runs sample tasks through flows and models and scores them:

# evals/my-suite.yaml
name: my-suite
flows: [quick]
cases:
  - name: add-multiply
    repo: fixtures/calc
    task: Add a multiply(a, b) function to calc.js.
    vars: {test_cmd: node --test}
    check: node --test

scf eval evals/my-suite.yaml --models sonnet,codex,ollama:qwen3-coder

The report shows pass rate, average cost, tokens, time and fix loops per variant, and appears on the Dashboard. If a cheaper model passes as often, route more steps to it.

Check that it worked#

The Dashboard shows today's spend below your daily budget.
The By user card shows each account's limits next to its numbers, for example 2 / 5.
Your eval report appears with the eval results on the Dashboard.