Documentation / The desktop app
The desktop app
Plan, build, and review agent work in one macOS app, with your own Claude Code, Cursor, or Codex.
Install and sign in
The desktop app runs on macOS. Request access to receive setup guidance and the current build; there isn’t a public download yet. The current build is unsigned, so only remove macOS quarantine if you trust the build and its source.
Open the app and choose Sign in with GitHub. Sign-in finishes in your browser, and the credential stays in the macOS credential store, outside the app’s web view. You don’t need the CLI to use the app.
Connect a local project
Open Settings, choose Claude Code, Cursor, or Codex, and map your Wayform project to the local product-repo folder. Your chosen agent needs to be installed and signed in separately. The app uses your installed agent; it doesn’t supply a hosted coding model.
Plan in a chat
Start a new chat and describe what you want to build. The agent can inspect the mapped repository, use web research, and read the team’s decisions before it writes a plan.
Planning chats run with read-only repository access. Saving or revising a Wayform plan is allowed; planning doesn’t edit your application files. Chat transcripts stay on your machine.
The Plans view lists every plan grouped by state. Open one to see its goal, approach, step status, what it builds on, and open questions. You can edit it and read earlier versions.
Build a plan in the app
Choose Build on an agreed plan. The build runs in a build chat:
- Each build gets its own worktree on a
wayform/<plan>-<agent>branch, so your working copy stays untouched. - Activity is folded into readable steps, and each checklist item shows its progress.
- Actions that need your call arrive as approval cards in the dock where you’d type, one at a time.
- You can send follow-ups to steer the build while it runs. Ended builds reopen from their saved transcript.
The build uses the agent you picked, with the team’s decisions briefed into each task and the plan checked against the diff before the pull request.
Auto mode
With Auto on, the agent’s own reviewer approves routine actions so you aren’t asked about every edit. Wayform’s own rules still apply first: protected paths and writes outside the build’s worktree are denied, and risky or irreversible commands still come to you as a card. Auto is available for Claude Code and Cursor builds and can be toggled per build.
When CI fails on the pull request
If a build’s linked pull request fails its checks, Wayform can wake that build. The agent resumes the same chat with restricted access, reads the failure, and pushes a repair to the PR branch. Wakeup turns never run in Auto mode. Automatic wakeups are available for Claude Code and Cursor.
See and fix memory
The Memory view lists the decisions and context your team has recorded, with what each one applies to. Correct a fact by superseding it, and review proposed replacements before they take effect. Removed facts stay visible so nothing disappears silently.
Models, effort, and usage
Pick the model and reasoning effort for a chat or build from the composer, and change them mid-session. Type /usage in a chat or build, or open Settings, to see your agent plan’s usage.
Keyboard shortcuts
| Shortcut | Action |
|---|---|
⌘ K |
Open the command palette |
⌘ N |
New chat |
⌘ Enter |
Send a chat message |
Esc |
Stop a chat turn or cancel an edit |
J / K |
Next / previous plan |
E |
Edit the current plan |
⌘ S |
Save a plan edit |
] |
Toggle plan details |
⌘ , |
Open Settings |
Connection settings
The app uses the default hosted gateway unless you set a different URL in Settings. An app opened from Finder or the Dock doesn’t inherit your terminal’s environment, so configure a custom gateway in the app itself.
If sign-in or an agent isn’t working, see troubleshooting.
Source: app/README.md; app/src (Settings, BuildChat, AgentDock, Memory, Usage); shipped plans #24, #29, #31, #33, #53, #58, #59, #60, #61, #65, #69