groundplan

Describe what you're building in a sentence — get a production-ready repository, with the architecture chosen for you and the reasoning written down.

npx groundplan new "An AI customer support platform with realtime chat and a knowledge base"

Groundplan reads that sentence, ranks architectures against the qualities it implies, picks the stack and the layers, assembles the repo from composable templates, brands it with your name, and writes the instruction files every AI coding tool expects. Run pnpm install && pnpm build and it builds the first time.

Install

npm install -g groundplan     # or just use: npx groundplan …

Use it

groundplan new "<description>"            # infer → plan → approve → generate
groundplan new "<description>" --dry-run  # print the plan, write nothing
groundplan new "<description>" --yes      # take every default, no questions
groundplan blueprint "<description>"      # the planning docs only — no repo
groundplan recommend "<description>"      # rank architectures, and why
groundplan doctor                         # installed AI CLIs + catalog check

Prefer to decide yourself? Every inference is overridable, and generation is then fully deterministic:

groundplan new acme-support --base nestjs-api --architecture vertical-slice \
  --layers core,db-postgres,auth-jwt

One prerequisite

Turning a sentence into a plan needs an AI coding CLI you already have — Claude Code, Codex, Cursor, Gemini CLI, Aider, Amp, Cline, Continue, Antigravity, or GitHub Copilot. Groundplan drives whichever it finds and never asks for an API key of its own. Pin the choices with --base / --architecture / --layers and no agent is needed at all.

groundplan doctor     # what this machine has, and the install command for what it lacks

Notes

• The templates and catalog live in @opengroundplan/core; installing this CLI brings them with it.
• groundplan doctor also reports which home directory it resolved — run it first if generation ever behaves unexpectedly.

Full documentation, architecture notes, limitations, and the contributor guide are in the repository.

MIT licensed.