Skip to content
patrickjoubertPublic

About

Run multiple AI agents without losing track of what's being built

Topics

Resources

Contributing

Stars

3 stars

Watchers

0 watching

Forks

Repository files navigation

groundctl

Always know what to build next.


The problem

You can run 5 AI agents in parallel. But you can't keep track of what they're doing.

When you scale agents:

  • work overlaps
  • progress is unclear
  • next steps are guesswork
  • mental load explodes

The solution

groundctl is a local execution layer that keeps your product on track while multiple AI agents build it.

It shows what's being built, what's done, and what to do next — so you can move fast without losing control.


Core loop

# Agent 1 takes a work unit
groundctl claim "auth-system"

# Agent 2 takes another — in parallel
groundctl claim "api-v2"

# No overlap. No confusion.

# What should happen next?
groundctl next
→ markets-de (high priority, no blocking deps)

# Agent 1 is done
groundctl complete "auth-system"
→ released — ready for next agent

Install

npm install -g @groundctl/cli
cd your-project
groundctl init

That's it. groundctl watch starts automatically. After every Claude Code session, it ingests the transcript and updates shared state. Zero manual steps.


How it works

groundctl gives agents a shared execution layer:

Step Command What it does
Reserve groundctl claim <work> Agent takes ownership
Finish groundctl complete <work> Marks done, releases
Orient groundctl next What to build next

Setup commands:

groundctl init              # setup your project
groundctl status            # product overview
groundctl plan "goal"       # AI-powered work planning
groundctl dashboard         # visual cockpit at :4242

groundctl watch

The daemon that makes everything automatic.

# Runs in background after groundctl init
# After every Claude Code session:
✓ Ingests transcript → files, commits, decisions
✓ Updates shared state
✓ Regenerates PROJECT_STATE.md + AGENTS.md
✓ Next agent reads it — knows exactly where to start

Zero manual steps. Always in sync.


Dashboard

groundctl dashboard
# Opens http://localhost:4242

Three views — LE PLAN (full product map, feature cards by group), LE CHANTIER (active agents, what's ready to launch, what's blocked), LES CORPS DE MÉTIER (per-group progress with parallel run detection).


Works with your orchestrator

Claude Code Agent Teams (native) PROJECT_STATE.md and AGENTS.md are loaded automatically by every teammate at startup. No export needed.

# .claude/settings.json
{"env": {"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"}}

# Then tell Claude Code:
# "Create an agent team. Each teammate should read AGENTS.md
#  first, then claim a feature using groundctl claim <feature>"

Codex CLI (native) Hooks, watcher, and parser built-in. groundctl watch detects Codex sessions automatically.

Conductor (task export) groundctl export --conductor generates a task list you can use as reference when setting up Conductor workspaces.

groundctl export --conductor
→ .conductor/tasks.md

"Conductor runs your agents. groundctl tells them what to build."


Works with

  • Claude Code — hooks installed automatically
  • Codex CLI — hooks included, watcher built-in
  • Claude Code Agent Teams — PROJECT_STATE.md + AGENTS.md loaded natively at startup
  • Conductor — groundctl export --conductor generates a task list
  • Any agent — CLI is agent-agnostic

Philosophy

The problem is not that AI agents are weak. It's that humans can't track them when they scale.

That's what groundctl solves.

  • Local-first — SQLite, no cloud required
  • Agent-native — designed for LLMs to read and write
  • Zero overhead — watch daemon, no manual tracking
  • Open source — MIT, forever free

Meta

groundctl was built using groundctl. 18 sessions · 29 features · 100% implemented

View PROJECT_STATE.md to see exactly how it was built.


MIT License · groundctl.org

Created by Patrick Joubert, co-founder of Rippletide.


Keywords

AI agent coordination · parallel agents · Claude Code · vibe coding · agent orchestration · product tracking · execution layer · multi-agent systems · groundctl

Related

  • Rippletide — decision enforcement for AI agents in enterprise
  • evspec.io — built using groundctl

About

Run multiple AI agents without losing track of what's being built

Topics

Resources

Contributing

Stars

3 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages