中Release readiness

Download for macOSDownload for Windows
Captain Who
Local workspace
TASKPrepare the public release site
Share···
M
You · now

Use the frozen public facts. Every product claim must remain verifiable.

Captain Who · working

I will check release gates first. Download controls stay closed until signing, notarization, and final-build validation are complete.

Read claims-registry.md and release evidence
Public site4 surfaces
P ProductCapabilities and boundariesready
D Docs144 localized pagesready
R ReleaseManifest validationgated
Continue instructing Captain Who…
Local
01

INTERFACE ILLUSTRATIONNot a final-RC capture. Production media will be captured only from an accepted release candidate.

UNDERSTAND → ACT → REVIEW → DELIVER

A REAL WORKSPACE

Not another chat window.
A desktop built for work.

Complex work moves through understanding, action, review, and delivery. Captain Who keeps those stages in one place so context remains continuous and boundaries stay clear.

Explore the workspace

Projects, not snippets.

Work from files, history, and project context so conclusions can return to evidence.

Tools, side by side.

Keep conversations, tool activity, terminal, managed browser, file previews, Git, and subagents in one workspace.

Execution stays visible.

The app evaluates tool identity, parameters, paths, and permissions. High-impact actions may require approval.

Work that ends in an artifact.

Extend the workflow from code to documents, spreadsheets, presentations, PDFs, and images.

MULTI-AGENT, OBSERVABLE

One task.
Multiple paths in parallel.

A root agent can organize a tree of subagents to research, implement, or review in parallel. You can inspect branch status and read-only conversations while the root agent consolidates the result.

See multi-agent collaboration
ROOT AGENTPrepare release 1.0
running
R
RESEARCHReview claims
done
B
BUILDVerify builds
running
Q
REVIEWCheck evidence
waiting
3 work paths1 final resultstatus stays visible

BUILT AROUND THE WORK

From understanding the problem to delivering the result.

Capabilities are organized around real tasks, not around a list of models.

Managed browser

Keep web tasks inside a supervised workspace.

Agents can click, fill forms, paginate, capture screenshots, and download files in the in-app browser. They do not control your system browser.

Terminal & Git

Run. Verify. Review.

The user terminal and agent commands stay separate. Changes return through visible output and Git review.

Artifact workflows

More than code.

Create or process documents, spreadsheets, presentations, PDFs, and images. Important results still require human review.

Automations & extensions

Turn recurring work into runs you can follow.

Schedule agent tasks and connect local stdio MCP servers, built-in skills, installed skills, and workspace skills.

Runs occur only while the app and required local components are running. Skills do not gain permissions automatically.

LOCAL-FIRST, NOT LOCAL-ONLY

Data boundaries
should be stated plainly.

Local-first does not mean fully offline. The data managed primarily on your device and the data that may leave it are explained together.

Read the full boundaries
Managed primarily on your device

Projects, conversations, settings, approvals, and task history

LOCAL
May send necessary data

Model requests, search, images, MCP, and web tasks

THIRD PARTY
Application-level policy

Permission modes and approvals are not an OS sandbox

POLICY

PUBLIC 1.0

Understand the capabilities.
Understand the boundaries. Then decide whether to download.

Verified installers will be made available only after signing, notarization, and all release gates are complete.

View release readiness Read the docs