AgentManager
An open-source dashboard for running many Claude coding agents at once with a big red stop button, cost tracking, and a record of who did what.
🔗 Visit AgentManager
Description
Running one AI coding agent is easy to watch. Running twenty of them at once — each working on a different task, each burning API credits, each capable of doing something you didn't ask for — is a different problem entirely. AgentManager is built for that second scenario: a control panel that shows every running agent, lets a human step in and pull the plug on any of them instantly, and keeps a running tally of what each one is costing in dollars.
Under the hood it's an open-source (MIT) orchestration layer for Claude agents, built with TypeScript, Next.js, Docker, and GCP, supporting up to 100 concurrent agents with parent-child lifecycle management, a multi-level emergency stop, real-time USD cost tracking per agent, persistence across restarts via cloud storage, and inter-agent messaging. It also manages git worktrees automatically, including cleanup and a cron scheduler for recurring jobs.
💬 Our review
The short version: if you're already running several Claude Code sessions in parallel and losing track of what each one is doing or costing, AgentManager gives you the missing supervision layer for free — but it's a young project (23 GitHub stars) so expect to do some of your own hardening.
There isn't a direct commercial competitor doing exactly this yet — tools like CrewAI, AutoGen, and LangGraph focus on agent orchestration logic (how agents talk to each other and divide work), while AgentManager focuses on operational safety and cost visibility for humans supervising fleets of coding agents specifically. That's a narrower, more practical niche. The tradeoffs are real: it's tightly coupled to GCP and Cloud Run today, so it's not very portable to other infrastructure, and documentation is thin for advanced setups — reasonable if you're already on Google Cloud, less so otherwise.
📊 Global score
🤖 AI-enriched data
Licence MIT, aucun frais de plateforme. Coûts d'utilisation selon la consommation API Anthropic/OpenRouter.
Pros
Orchestration multi-agents native pour Claude
Kill switch d'urgence multi-niveaux
Suivi des coûts par agent en temps réel
Persistance inter-redémarrages
Gestion intégrée des git worktrees
Cons
Jeune projet (23 stars)
Fortement couplé à GCP/Cloud Run
Documentation réduite pour cas complexes