Product Overview

Captain Who is a local-first desktop agent work assistant. It does more than generate a response: within the permissions you grant, it can read projects, research sources, call tools, modify files, run commands, and delegate parts of complex tasks to multiple subagents.

How it differs from a standard chat tool

A standard chat is usually one question followed by one answer. Captain Who works toward a goal by repeating a loop: understand the task, choose the next step, call a tool, inspect the result, then decide whether to continue, request approval, or stop. This is the basic way an agent works.

For example, if you ask, “Find out why this project fails to start,” the agent can inspect the directory, search configuration, read logs, run tests, and summarize its findings in sequence instead of guessing from a short excerpt you pasted.

Work it currently suits

Read and organize local projects; search files and code.
Create, modify, and review files; inspect unstaged, staged, and last-turn Git changes.
Run managed, non-interactive commands for builds, tests, and data processing.
Search the web with Tavily and read web pages.
Browse manually or, with permission, let an agent control the managed in-app browser.
Read attachments and work with Word, Excel, PowerPoint, PDF, and image files.
Install or select skills that provide task-specific instructions and capabilities.
Connect local stdio MCP servers and use the tools they expose.
Let multiple subagents research, implement, or review separate subtasks in parallel.
Create scheduled automations that run on a schedule while the app is running.

Availability also depends on the current operating system, model service, project, permissions, runtime components, and third-party configuration. Rely on the capabilities shown as available in the interface and on the corresponding capability documentation.

What “local-first” means

Conversations, project records, settings, approval state, and task history are stored primarily on your device. Adding a project to Captain Who also does not automatically upload its files to a Captain Who cloud service.

However, the following data leaves your device:

Prompts, necessary context, and tool results sent to the model are transmitted to the model API you configure.
When web search is enabled, queries or target URLs are sent to Tavily.
When image generation or MCP is enabled, relevant inputs are sent to the service or local process you configure.
Browsing the web connects to the relevant websites, as a standard browser does.

Before working with confidential material, review each provider's data policy and your organization's rules.

Capabilities you should not currently expect

There is currently no enabled public download channel, public update feed, or cloud synchronization service.
The app is not a fully offline model runner. Ordinary tasks require an available model API.
Permission policy is not an operating-system-level container, and Full permissions do not mean that every dangerous command will be allowed.
Scheduled automations are not awakened by the operating system after the app exits and do not guarantee second-level timing.
Multi-agent work uses tree-structured collaboration; it is not a visual workflow builder or an arbitrary process orchestrator.
“Files” in the right sidebar is a read-only previewer. It does not render Office files directly.
There is currently no unified, permanent, cross-device artifact library and no one-click user control for erasing all local data.

Next steps

First-time user: Read Installation and First Launch.
App already launches: Go directly to Complete Your First Task.
Want to understand how it works first: Read From Chatbot to Agent.