If you've used Claude Tag, you know the potential of it for orchestration, a few months ago I built a version of it as a Claude Code plugin, that brings the concept as a local version of it, simply as a always-on task driven Claude Code agent. Our clis, our local memory, our filesystem, our tools, for you or your team.
It's how I manage and orchestrate several projects across different companies, and how I've been running always-on agents since.
Set one up under 5m, it's just Claude Code so you can run with any subscription including 20$, it's fully customizable and it has got some magic focused on token efficiency using claude code native features:
1. Chat through Claude Code channels (Discord, Telegram, iMessage etc). Handle permission prompts in chat, recalls past conversations, keeps per-channel rules ("remember this for this channel"), has per-user allowlists for teams, and you manage the session from chat: !model, !effort, !clear, !compact, !stop, etc.
2. Routines that only wake Claude when needed. Each routine can run on its own model and effort, and a precheck script decides whether the model runs at all. Example: triage new GitHub issues hourly, with a gh CLI precheck that only wakes the model when there are new untriaged issues.
3. Heartbeat that only wakes Claude when there are pending stuff waiting, quiet ticks mean 0 tokens.
4. Spawn and orchestrate local Claude Code sessions with the model and effort you want and let it steer using cross-messaging. In chat each task has it's own thread and you can steer sessions per thread. You can also change !model and !effort on a thread session.
5. Take over the agent or any session on the CLI or the Claude app. (all of them start with remote-control)
6. A watchdog handles auto-restart, and context hygiene compacts long sessions and clears them when idle. All configurable.
Feel free to explore the repo for full list of features ;)
Hey guys,
If you've used Claude Tag, you know the potential of it for orchestration, a few months ago I built a version of it as a Claude Code plugin, that brings the concept as a local version of it, simply as a always-on task driven Claude Code agent. Our clis, our local memory, our filesystem, our tools, for you or your team.
It's how I manage and orchestrate several projects across different companies, and how I've been running always-on agents since. Set one up under 5m, it's just Claude Code so you can run with any subscription including 20$, it's fully customizable and it has got some magic focused on token efficiency using claude code native features:
1. Chat through Claude Code channels (Discord, Telegram, iMessage etc). Handle permission prompts in chat, recalls past conversations, keeps per-channel rules ("remember this for this channel"), has per-user allowlists for teams, and you manage the session from chat: !model, !effort, !clear, !compact, !stop, etc.
2. Routines that only wake Claude when needed. Each routine can run on its own model and effort, and a precheck script decides whether the model runs at all. Example: triage new GitHub issues hourly, with a gh CLI precheck that only wakes the model when there are new untriaged issues.
3. Heartbeat that only wakes Claude when there are pending stuff waiting, quiet ticks mean 0 tokens.
4. Spawn and orchestrate local Claude Code sessions with the model and effort you want and let it steer using cross-messaging. In chat each task has it's own thread and you can steer sessions per thread. You can also change !model and !effort on a thread session.
5. Take over the agent or any session on the CLI or the Claude app. (all of them start with remote-control)
6. A watchdog handles auto-restart, and context hygiene compacts long sessions and clears them when idle. All configurable.
Feel free to explore the repo for full list of features ;)
Contributions are welcome!