Skip to content
Commandr
Changelog

What’s new in Commandr

Every release, newest first. You’ll also find these notes in the app under Settings → About → What’s new.

All releases on GitHub

v1.1.0

Latest

Find anything without leaving the keyboard, and stay in control of what each agent spends.

Command palette

Press ⌘K anywhere to run an agent, jump to a project or open a setting. Start typing and the best match is already selected, so most commands are two keys and a Return.

habit-tracker
runesc
Agents
  • Run CheckReview the last build↵
  • Run TestWrite and run tests
  • Ask Debug about an errorPaste a stack trace
  • Sync docs with codeUpdate the README
↑↓ to move↵ to run⌘K anywhere
Typing “run” lists every agent you can start right now.

Spending limits

Give each agent and each project a budget. When an agent reaches its limit, Commandr pauses the run and keeps its work, so you can raise the limit and pick up exactly where it stopped.

habit-tracker
SpendingSet a limit per agent and per project. Commandr pauses the run when it’s reached.
$4.40of $10.00 this project
Frontend$1.62 / $2.00
Backend$2.00 / $2.00
Design$0.48 / $1.00
Check$0.21 / $1.00
Test$0.09 / $0.50
Backend paused at its $2.00 limit, with the API half done.Raise to $3Resume
Backend paused at its limit, with Resume one click away.

Also in this release

Improved

  • Check now reviews only the files that changed since its last run, so it finishes in about half the time.
  • Long agent replies scroll smoothly, even with hundreds of files in the list.

Fixed

  • Pausing during Design no longer loses the colours it had picked.
  • The sidebar keeps its width after a restart.
In the app
Download v1.1.0

v1.0.1

Fixed

  • Projects in folders with spaces in their names open again.
  • Undo run now also restores files an agent deleted.
  • The update card no longer shows twice after a restart.

v1.0.0

Beta

The first public beta of Commandr: thirteen agents that plan, build and check your app, right inside your project folder.

A crew that works in order

Describe your idea once. Scope plans it, Audit looks for gaps, Wireframe and Design shape it, Architect picks the stack, then the build agents write the code. Each one hands its files to the next, and you can pause between any two.

todo-list
PipelineEach agent hands its files to the next. You can pause between any two.
6 of 9 stages
  • Scope EagleTurns your idea into a clear build scope.0:41$0.06Done
  • Audit RaccoonFinds gaps, risks and missing detail.1:12$0.02Done
  • Wireframe BeaverMaps every screen and sketches each one’s layout.2:05$0.09Done
  • Design FlamingoCreates the design system and UI tokens.2:48$0.11Done
  • Architect AntChooses the stack and sets up the project.1:30$0.07Done
  • Frontend ButterflyBuilds components, pages, styling and state.4:16$0.38Building
  • Check CatReviews code and screenshots against the design.——Queued
  • Test RabbitWrites tests and runs them.——Queued
Six of nine stages done, with Frontend building the UI.

Undo any run

Commandr saves a checkpoint before every run. If you don’t like what an agent did, Undo run puts every file back exactly as it was, including files it created or deleted.

habit-tracker
CheckpointsEvery run is saved before it starts. Undo puts your files back exactly as they were.
Added the habits APIBackend · 2 min ago
+ src/api/habits.ts+ src/api/streaks.ts~ src/server.ts
Undo run
Built the Today screenFrontend · 9 min agoCheckpoint saved
Restore
Created the design systemDesign · 18 min agoCheckpoint saved
Restore
The last run added two files and edited one. Undo reverses all three.

Pick a model for each agent

Use a strong model where it matters and a cheap or local one where it doesn’t. Local models run on your Mac through Ollama, so they’re free and your code never leaves your machine.

Settings
ModelsChoose the model each agent uses. Local models run on your Mac and cost nothing.
Scope EagleClaude Sonnet · Anthropic
Audit RaccoonDeepSeek V3 · DeepSeek
Architect AntClaude Sonnet · Anthropic
Frontend ButterflyClaude Sonnet · Anthropic
Test RabbitQwen Coder · Local · Ollama
Document PenguinGPT mini · OpenAI
Claude SonnetAnthropicRecommendedDeepSeek V3DeepSeekGPT miniOpenAIQwen CoderLocal · OllamaFree
Switching Test to a local model.

Also in this release

New

  • A first-run checklist that gets your first project going in a few minutes.
  • One-click updates, with these notes inside the app.
  • .env files are always protected: agents can’t change them and checkpoints never store them.

Commandr is in beta. Things may change between versions, and we’d love to hear what breaks.