Rate limits & usage

What counts as a step#

A step is one reasoning turn by one agent — typically a single model call plus any tool invocations it triggers. A run with a three-agent crew consumes three steps. Each step records its own input and output tokens, which roll up onto the run and into your monthly usage.

Included steps by plan#

Beyond your included steps, usage is billed at your plan's overage rate. Model usage is billed separately, at the provider's list price with no markup (see Models & providers). See Pricing for current per-step rates and seat limits.

Hard caps#

Set a hard cap in your dashboard to stop runs once you hit a monthly step budget — useful for preventing surprise overage on automated pipelines. When the cap is reached, new runs are rejected until the next billing period or until you raise the cap.

Tracking usage#

Each terminal run emits a usage event recording its steps, tokens and model cost. Your dashboard aggregates these into a rolling 30-day view, broken down by workflow and by model. To monitor usage programmatically, sum the total_steps, tokens_in, tokens_out and cost_usd fields returned on each run.

// Roll up steps across recent runs for a workflow
const { data: runs } = await fetch(
  `https://loopllama.ai/api/v1/workflows/${workflowId}/runs`,
  { headers: { Authorization: `Bearer ${process.env.LOOPLLAMA_API_KEY}` } },
).then((r) => r.json());

const steps = runs.reduce((sum, r) => sum + r.total_steps, 0);
const tokens = runs.reduce((sum, r) => sum + r.tokens_in + r.tokens_out, 0);
const modelCostUsd = runs.reduce((sum, r) => sum + r.cost_usd, 0); // billed at cost

Best practices#

• Keep crews as small as the task allows — every extra agent is an extra step on every run.
• Cache or deduplicate inputs upstream so you don't trigger identical runs.
• Use per-agent model overrides to route cheap steps to a smaller model — from any provider — and reserve frontier models for the steps that need them. Model usage is billed at list price, so the savings are yours.