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

# What Uses Credits

> When Dume actually spends credits, and how to keep usage down.

Setup doesn't cost anything on its own. Connecting an integration, enabling a skill, or creating a routine just configures Dume, it doesn't run the model. Credits get spent when a session actually runs and Dume generates a response.

## When usage starts

* **Sending a message** runs the model on your request and costs credits.
* **A skill running inside a task** costs the same way; the skill's instructions get loaded in and the model follows them.
* **A routine firing on schedule** costs credits the same as if you'd sent that message yourself, since it's still a session running.

There's no flat per-message fee. Cost scales with how much the model actually reads and writes: a short answer on a light model can be a single credit, a long research task on a premium model can run into the dozens. See the [Credit Cost Guide](/account-settings/credit-cost-guide) for real numbers by task and model.

## Keeping usage down

* Use a cheaper or free model (like Ternary Bonsai 27B) for quick questions, and save premium models for work that actually needs the extra reasoning; see [Choosing a model](/features/multiple_llm).
* Be specific in your prompt. A vague request often costs more in back-and-forth clarification than a precise one would have upfront.
* Scope routines to what you actually need summarized or done, not everything a tool could return.
* For a repeated task, a well-scoped custom skill can make each run shorter and more predictable than starting from scratch every time; see [Building a custom skill](/skills/custom-skills).