Back to guides

Choose where your AI runs.

KAKABAKA is a client for the model service you configure. It does not include a hosted model subscription.

Add the endpoint you intend to use

Enter the service URL, exact model identifier, and authentication details provided by your model service. API compatibility and supported tools vary. Start with a small task to confirm the configuration.

Match capabilities to the task

Screenshot-based phone tasks need a model that accepts images. File and tool workflows need the tool capabilities required by the task. A model being reachable does not guarantee that every feature will work.

Using Ollama

Configure the reachable Ollama endpoint and an available compatible model. Check whether the selected model runs on your own computer or uses a cloud service. An address on localhost is not, by itself, proof that processing stays local.

Review what is shared

The main model can receive your message, relevant history, attachments, and the phone screen or interface text needed for a task. An enabled decision model receives the task goal and relevant interface text and candidates. Review both destinations.

Manage authorization

The Microsoft Store edition requests consent before sharing task data. Changing the model or destination requires fresh authorization. Revoke it in Settings → Data controls to stop active AI tasks and block later requests until you authorize again.

Understand cost and retention

Your selected provider sets usage fees, limits, retention, and deletion options. A successful connection is not a cost limit. Check your provider account and policy before long-running or scheduled work.