Agents everywhere. Nobody steering.
They overlap, wait on each other and pile up in your notifications.
Hello, world.
The open-source cockpit for running a whole flock of AI agents. They work on your Kubernetes, and only honk when a human should decide.
helm install flokpit [YOUR-CHART]
Why a cockpit
They overlap, wait on each other and pile up in your notifications.
Each agent knows its slot, and you see the whole flock at a glance.
What makes flokpit different
Frameworks help you build an agent. Coding agents finish one ticket at a time. flokpit is where a human runs a whole flock of them, safely, every day.
You decide which actions honk. Everything else flies on its own, within your rules and budget. Your attention goes where a human is needed.
Chief agents lead sub-agents in formations you can see. One view shows who does what, what it costs and where it is stuck.
Commands and @mentions in chat. Answer a honk on your phone, by email or with your voice. No dashboard to babysit.
Open source and runs on your Kubernetes. Bring any AI account or local models. Every decision lands in an audit log you own.
| Agent frameworkse.g. LangGraph, CrewAI, AutoGen | Workflow builderse.g. n8n, Temporal | Hosted coding agentse.g. Devin, Copilot coding agent | flokpit | |
|---|---|---|---|---|
| Main job | Build an agent in code | Automate fixed flows | Finish a ticket, open a PR | Run many agents as a team |
| How you steer it | Code and config | Visual flow editor | Issue, Slack or web prompt | Chat, phone, email, voice |
| When a human is asked | Wherever you code an interrupt | Wait steps in the flow | At PR review | By rule, per action and flock |
| Many agents at once | You wire it yourself | Parallel branches | Often one task per agent | Chiefs, sub-agents, formations |
| Budget and audit | Via tracing tools | Run history | Per seat or usage plan | Budget guard and audit log per flock |
| Where it runs, which model | Your code, any model | Self-host or cloud | Mostly vendor cloud, limited model choice | Your Kubernetes, any model |
flokpit is meant to sit on top, not replace: agents built with a framework can fly in a flock. [Confirm before launch]
How it flies
flokpit runs on your own Kubernetes. Your agents, your data, your rules. Bring an API account for the models, like Anthropic, OpenAI or a local model. Which ones?
Start from a template or one agent at a time. Chief agents bring their own sub-agents, and each files a flight plan before take-off.
Give direction with commands and @mentions. When an agent needs your call, it honks, in the cockpit, on your phone or by email.
Set approval rules and a budget per flock. Every decision lands in the audit log, and most actions can be undone.
Demo with simulated replies. Open the full app →
Your cockpit
See the whole formation, approve what matters and land agents with one command.
Open the appflokpit pocket
The companion app for iPhone and Android. When an agent honks, you decide in seconds, wherever you are.
Coming soon Store links are placeholders [link]. Until then, try it in the demo under ••• → Pocket cockpit.
wants to merge PR #91 “Rate limit for /search”. Tests pass.
Flock facts
We study how flocks, schools and colonies coordinate, then build that into flokpit.
Each starling tracks only its 7 or so nearest neighbours. No leader needed.
Simple rules. Massive coordination.
Ballerini et al., PNAS 2008
Lower heart rate for pelicans flying in formation compared with flying solo.
Together, every agent works less.
Weimerskirch et al., Nature 2001
A slime mould grew a network similar to Tokyo’s rail system.
Simple units. Smart routing.
Tero et al., Science 2010Follow the flock
Open source