AI CLI Agent · MIT

From idea to deployed MVP,
in one terminal session.

An AI CLI agent that runs your full build pipeline: discovery, planning, coding, acceptance, deploy, and launch copy — without leaving the terminal.

Install NowView on GitHub
Provider-agnostic Sandboxed in Docker MIT licensed
pilot-agent — run
~/projects/taskflow $ pilot-agent runLoading state from .pilot-agent/STATE.md …DISCOVERY What problem does this MVP solve?✓ Scope captured · 4 clarifying questions answeredPLANNING Architecture drafted → 7 milestonesCODING api/ · web/ · db/ scaffolded✓ acceptance 12/12 checks passingDEPLOY live → https://taskflow.app ✓
bring your own model —AnthropicOpenAIOpenRouter
Why Pilot Agent

A sandboxed operator, not a chat window

It plans, writes, verifies and deploys — then explains what it learned. Here's what that replaces.

CapabilityPilot AgentManual workflow
Full pipeline in one command—
State persisted across sessions—
Switch AI provider mid-session—
Docker-sandboxed execution—
Verification loop until acceptance passes—
Self-improving via inspectable lessons—
Get started

Get started in 30 seconds

One command and you're in. Pick the install path that matches your setup.

Linux / macOS / WSL2 — recommended
curl -fsSL https://pilotagent.vercel.app/install.sh | bash
Skip the setup wizard
curl -fsSL https://pilotagent.vercel.app/install.sh | bash -s -- --skip-setup
Manual install with uv
uv tool install git+https://github.com/Hqzdev/pilot-agent
Then — your first build
four commands
pilot-agent setup     # configure AI provider and API key
cd your-project
pilot-agent init      # create .pilot-agent/STATE.md
pilot-agent run       # start the build pipeline
The pipeline

How it works

Six phases, one continuous session. You stay in control at every checkpoint.

  1. pilot-agent setup Configure your provider

    Choose your AI provider — Anthropic, OpenAI or OpenRouter — and add your API key. One-time setup.

  2. pilot-agent init Initialize project state

    Creates .pilot-agent/STATE.md so the agent has a durable memory of your project from line one.

  3. pilot-agent run Discovery phase

    The agent enters discovery and asks clarifying questions until the scope of your MVP is unambiguous.

  4. Plan · code · test — automatically

    It drafts the plan, writes the code and runs the tests on its own. You review and approve at each phase boundary.

  5. Deploy & launch copy

    The same run continues through deployment, then generates ready-to-use launch and marketing copy.

  6. Done — your MVP is live

    A deployed product, a clean state file, and a record of every decision the agent made along the way.

Ship your next MVP from the terminal

One command to install. One session to deploy.

$curl -fsSL https://pilotagent.vercel.app/install.sh | bash
Install NowView on GitHub