Models and context

Aru uses AI models made available by your organization. The model list and options depend on the organization and model. Different models may handle the same request differently.

Choose a model

1. Select the current model in the composer.
2. Browse the list or search by name.
3. Choose a model.
4. Confirm the name in the composer before sending.

You can favorite models you use often. The selection is stored per task, so check it when switching tasks.

If the list is empty or shows No model, check the current organization and its available models. See Troubleshooting if the problem continues.

Reasoning effort and speed

Models that support it offer Reasoning effort. Consider a higher setting for complex analysis or multi-constraint comparisons, and start with the default for short tasks. A higher setting does not guarantee a better result.

Speed is also model-dependent. Faster modes may use more quota, so review the description before choosing one.

If Aru adjusts an unsupported combination, it may show Model settings adjusted. Confirm the model and options that were applied.

Context usage

Select the usage indicator in the composer or open the Context panel to see how much of the model’s context window the conversation uses. In a long task, repeat critical constraints and result paths in a follow-up request. Start a new task for a new goal.

Context is the conversation and material the model can consider. Tokens measure the amount of text processed; they are not the same as characters or file size.

Continue a long task

Aru may summarize earlier conversation to continue a long task. Restate conditions that must remain in effect and important result paths. You can also ask it to reread decisions saved in project files.

The confirmed requirements are: write the report in Korean, compare against the previous month, and keep the source files unchanged. Read results/performance-summary.md and continue under those requirements.

When the goal changes or the conversation contains too much unrelated material, start a new task with only the necessary sources and decisions.

Change models during a conversation

Changing the model applies the new model’s limits and behavior. Review any performance or context-compaction notice, then restate the important requirements and where to continue.

See Tasks and conversations for follow-up requests.