Hey guys,
I've been running a handful agents with Claude Code and claude-code-hermit plugin for the last 4 months under my subscription.
You setup an agent with it's identity under 3 minutes in a fresh folder or existing project.
Add your own agent routines, mcps, watches, etc. It's claude code but stateful, persistent and proactive that self-improves through an operator-gated proposal system.
In terms of token efficiency it has some magic with routine delegation to subagents when want (like haiku), you can setup your own routines & heartbeat precheck scripts so the model doesn't wake for nothing and sessions self-management with auto-compaction & auto clear.
I set these up in companies & for myself, you can literally run a bunch of these in a single laptop, there's a quick /docker-setup wizard for them to run isolated in docker and /channel-setup to wire discord & telegram really quick using the official discord & telegram official channels.
Hey guys, I've been running a handful agents with Claude Code and claude-code-hermit plugin for the last 4 months under my subscription.
You setup an agent with it's identity under 3 minutes in a fresh folder or existing project.
Add your own agent routines, mcps, watches, etc. It's claude code but stateful, persistent and proactive that self-improves through an operator-gated proposal system.
In terms of token efficiency it has some magic with routine delegation to subagents when want (like haiku), you can setup your own routines & heartbeat precheck scripts so the model doesn't wake for nothing and sessions self-management with auto-compaction & auto clear.
I set these up in companies & for myself, you can literally run a bunch of these in a single laptop, there's a quick /docker-setup wizard for them to run isolated in docker and /channel-setup to wire discord & telegram really quick using the official discord & telegram official channels.
I run almost all of them on sonnet & auto mode.
Let me know what you think!