@simpleagentman allows you to do this with Claude Code, Codex, and OpenCode

That's why I started building @simpleagentman in Jan. Was sick of agents bumping into each other, eating each others' resources, etc. and I couldn't even access them from my phone. The goal is to have a clean single interface to run hundreds of agents in their own isolated envs.

Improved the UI to understand relationships between running agents and subagents.

How do you build your context and your agent's before you build?

Started imagining one in December. Decided to start building it in Jan. Seems like we all had the same idea 😂 Mine is @simpleagentman Open source. Cloudflare control plane. Runs agents in devcontainers in vms on Hetzner (or GCP or Scaleway)

Gonna start posting some stuff about how I work with agents. Maybe it's useful to a couple people out there 🤷‍♂️

Personally with @simpleagentman I'm sort of doing that, but I've been playing with developing tooling for agents to spin up other agents in sandboxes, monitor them, control them. It's still not a proper framework, but I think I see the beginnings of one from these primitives.

Been a while since I ran out of tokens. Restarting all with different agent. I wonder if I should build some tooling in @simpleagentman to detect this kind of thing and automatically reroute to an agent that has usage left.