Frequently Asked Questions
Can Captain Who be used offline?
The interface, project history, and many local Tools run locally, but language-model requests normally require a connection to the Provider you configure. Online search, web access, image generation, GitHub Skill installation, and network-connected MCP Servers also require a network connection.
Must I use a specific model service?
No. The current adapters support OpenAI-compatible, Anthropic-compatible, and DeepSeek V4 Chat configurations, but a specific endpoint must meet the protocol requirements. Compatible gateways must be verified by the user.
Does Captain Who charge me automatically?
Costs shown by the application are local estimates calculated from the per-thousand-Token prices entered by the user. They are not the provider's bill. Whether models, search, image generation, or third-party tools incur charges is determined by the relevant service provider.
Is all data stored only on the device?
Projects, conversations, and settings are stored primarily on the device, but when you start a model, search, image, or web task, the data needed to complete that request is sent to the relevant third party. An MCP Server configured by the user may also access the network independently. See Data and Permissions.
Where are Tokens stored?
Signed macOS releases store the model Token, Tavily Key, and image-generation credentials in separate system-keychain namespaces. SQLite retains only opaque references. Unsigned development builds use an independent private development credential store; this is not a storage commitment for release builds.
Why does the Agent request approval?
File writes, commands, patches, external MCP calls, Skill Scripts, or sensitive browser actions may have side effects. An approval is bound to a specific action; it does not grant the model permanent authorization. Review the target, parameters, and impact before deciding.
What is the difference between a Skill and MCP?
A Skill primarily tells the Agent “how to do something”; an MCP Server primarily provides “tools that can be invoked.” A Skill does not automatically gain file, command, or network permissions because its source is trusted.
Does Scheduled Automation continue after the application exits?
No. It is not an operating-system background service. Tasks start only while Captain Who and Core Server are running. Multiple schedules missed while offline are merged into one recovery Run on the next launch rather than replayed one by one.
Does pausing an Automation stop the current Run?
No. Pausing prevents only future schedules. The current interface also has no separate button for stopping an active Automation Run. Deleting a task starts a cancellation and cleanup process, but external side effects that have already occurred cannot be reversed.
Does removing a project delete project files?
It does not modify files in the project directory, but it permanently removes that project's local conversations, messages, and attachment records from Captain Who. Confirm that you no longer need this history before proceeding.
Are automatic updates supported?
The app includes an updater that accepts only the signed stable channel. The public channel is not currently enabled, so Settings shows it as disabled and the app does not access the update network. Once the formal channel is enabled, you can check, download, and install an update after a safe exit under “Settings → About & Updates.” There is currently no automatic rollback service. See the Upgrade Guide for details.
Does browser automation use my system browser?
No. Agent automation controls only the browser environment managed by the application and does not connect to arbitrary external debugging endpoints. It also has an authorization boundary separate from the manual browsing interface.