Everyone was playing with OpenClaw recently. Me too. But soon I realized that OpenClaw is a black box with more than 400K lines of code. Then I struggled to set up agents to communicate with each other - it was unpredictable and unmanageable. I realized that a custom agentic runtime is what I really need. A system where agents will be able to communicate with each other. Brute-forcing the solution, I first came up with an HTTP Router connected to nanobot agents. It allowed them to use API calls to send messages to one another. Well, it was shut down pretty quickly when I realized that without proper context management, this system just explodes. Agents have been chatting constantly, piling their context on top of each other. They were unable to filter or shrink it, and, of course, they were unable to cooperate to solve any task, even a slightly complex one.
Being in the shower after a long day, I’ve been thinking about what I should really do to somehow solve this issue. I needed something lightweight, powerful, and customizable. Remembering my own experience with managing teams, the idea hits me. Tickets! When I needed something specific from someone working in a team, I just created a ticket! It might be an issue on GitHub or a sub-conversation in Slack. There, I always described exactly what I needed and from whom.
Getting back to the laptop, I started drafting what became h1v3 (Hive). A custom runtime written (vibecoded) in Golang. One binary, many Agents - each running as a goroutine. Same with OpenClaw - dedicated workdir, memory, tools, and skills, but smarter. Existing as goroutines, my agents became interconnected via Registry - a central message broker.
Each new conversation, whether with a user or another Agent, becomes a new Ticket. Tickets are nested, and each encapsulates everything needed to solve a specific piece of the job.
That was the solution! Now Agents have only what they really need to act. Everything in a single binary file, in a single Docker container.
Going further, I built h1v3 Monitor - an app that shows what is actually happening under the hood. Agents, tickets, their context, and each tool call - everything becomes transparent and easy to debug.
h1v3 is opensourced and it’s still just taking off. If you ever wanted to run a swarm of agents - join as a contributor or a tester, join your thoughts on the approach, and let's make the future agentic!
This article was written by a human. @cxrtisxl
Top comments (0)