Per-task agent
Claude Code, Codex, or Gemini — chosen per task and switchable anytime. Set a specific model override per run.
Features
Everything you need to command a fleet
Command Fleet wraps the agent CLIs you already use in a real operations layer — portfolio organization, safe parallel runs, review-before-merge, and an autonomous build loop. Here's the whole picture.
Organize
A portfolio, not a single repo
Structure your work the way it actually exists: an organization holds workspaces, workspaces hold projects, projects hold tasks. Every project gets a clean Kanban board — To do, In progress, In review, Done.
- Drag-and-drop board with priority-sorted cards
- Move projects between workspaces in one drag
- Archive finished work without deleting it
- A home dashboard with counts, a needs-review queue, and a live activity feed
Dispatch
Hand any task to any agent
Pick the right model for the job, every time.
Plan or build mode
Run plan-only to investigate without edits, or full build. A full-access toggle lets trusted runs skip per-command approval.
Live streaming output
Watch the agent work in real time as output streams into the task history, with a running indicator you can cancel.
Resumable sessions
Send a follow-up — “now also fix X” — and the agent picks up in the same worktree where it left off.
Attachments
Attach screenshots, specs, or logs to a task; the agent is handed the file paths so it can read them for context.
Prompt templates
Reusable snippets you drop into any task by typing @name — your bug-report or release-notes boilerplate, once.
Safe by design
Parallel agents that never collide
Every run happens in its own git worktree on a dedicated branch. Spin up five agents across five projects and none of them touch your working tree until you say so.
- Per-run isolated worktree + branch
- Setup script (e.g.
pnpm install) prepares each fresh worktree - Cleanup script lints/formats after every turn, so the diff is clean
- In-app diff with +/- coloring, then one-click merge — or discard
- Optional verify gate: a Done task that fails your
build & testis routed back to review
Autonomous
The workspace PM that builds for you
A manager agent oversees a whole workspace and delegates real work.
Scaffold
Describe an app; it creates the project from a stack pack and wires the basics.
Plan
Decomposes the build into an ordered task graph with dependencies between tasks.
Execute
A scheduler runs ready tasks in parallel, capped for safety, each in its own worktree.
Review & merge
Failed work is auto-retried, then escalated to you; finished branches auto-merge.
Ship it
Preview and deploy, built in
Run any project's dev server inside the app, browse its files, and ship to production with a one-button pipeline — no leaving Command Fleet.
- Preview web, desktop, or custom command targets with a live URL
- Built-in file browser for static assets and docs
- Six deploy stack packs: Firebase, Cloudflare, Vercel, Netlify, Supabase, Fly
- Ordered deploy pipeline, gated until every required credential is set
- Credentials kept off the AI-writable manifest and out of prompts
One contract, two machines
Preview and Deploy are the same engine pointed at different commands — declared in a small manifest the AI can write or you can hand-edit. What you preview is what you ship.
Stay in the loop, hands-off
- Desktop notifications when an unattended run finishes
- Event webhooks to Slack, Discord, or your own endpoint
- A local JSONL audit log of every run
- Auto-resume tasks when a provider usage limit resets
Local-first, by default
- Your code and data never leave your machine
- API keys stored in your OS credential vault, never in plaintext
- A per-project secrets vault that is never sent to any agent
- House-rule preambles prepended to every prompt for consistency
See it run on your portfolio.
Start your 14-day free trial — no credit card required.