Always know what to build next.
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
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.
# 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 agentnpm install -g @groundctl/cli
cd your-project
groundctl initThat's it. groundctl watch starts automatically.
After every Claude Code session, it ingests the
transcript and updates shared state. Zero manual steps.
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 :4242The 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 startZero manual steps. Always in sync.
groundctl dashboard
# Opens http://localhost:4242Three 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).
groundctl tells your agents what to build. Your orchestrator runs them.
groundctl export --conductor
→ .conductor/tasks.md — ready to import
groundctl export --agent-teams
→ .claude/tasks/groundctl-export.jsonCompatible with Conductor, Claude Code Agent Teams, and any tool that reads task lists.
"Conductor runs your agents. groundctl tells them what to build."
- Claude Code — hooks installed automatically
- Codex CLI — hooks included
- Conductor —
groundctl export --conductor - Claude Code Agent Teams —
groundctl export --agent-teams - Any agent — CLI is agent-agnostic
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
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.
AI agent coordination · parallel agents · Claude Code · vibe coding · agent orchestration · product tracking · execution layer · multi-agent systems · groundctl
- Rippletide — decision enforcement for AI agents in enterprise
- evspec.io — built using groundctl