Agents can do a lot of work per request; knowing where credits go keeps costs predictable.
Gumloop measures usage in credits. Unlike a fixed price per message, you pay for what the agent actually does, so a quick question costs far less than a long research task. (Prices and plan amounts change, so check your account for current numbers.)
What uses credits:
- Chat & Reasoning: the model reading, thinking and writing. Bigger models and longer conversations cost more, and so does every connected tool, because tool descriptions are sent to the model too.
- Tool calls: each successful app action. Failed calls are free. Some tools add their own fees.
- Compute: time the agent spends actively working, such as running code or browsing. Waiting for you is free.
- Brain searches and indexing.
- An orchestration fee: a percentage on top for running the agent loop.
Where to see it: every task shows a running credit total; click it for a breakdown by type. Usage & Limits has detailed logs you can export, and you can set alerts when usage crosses a threshold.
How to spend less:
- Use the simplest model that does the job well.
- Start a new task for a new topic.
- Attach only the connectors and abilities the agent needs.
- Prefer built-in triggers over AI-created polling triggers where one exists.
Bring your own key (BYOK): on paid plans you can connect your own AI provider account, so model costs are billed by that provider instead; tool calls and compute still use credits. In shared agents, usage is usually charged to the person running it.
Remember
- Credits pay for model work, tool calls, compute, Brain and orchestration
- Each task shows its credit total and a breakdown
- Smaller models, fewer tools and fresh tasks reduce spend