https://reddit.com/link/1ss24su/video/5yhh67c37mwg1/player

I love Claude Code, and I liked what OpenClaw showed about autonomous agents, but I wanted something simpler and more native to Claude Code itself.

So I built claude-code-hermit , it turns Claude Code into an 24/7 assistant that lives in a folder, survives reboots, works through recurring routines, and gets better over time by reflecting on repeated blockers.

The rule from day one was: use Claude Code’s own primitives as much as possible: memory, channels, remote control, plugins, etc. No separate agent framework. No big reinvention layer. If Claude Code improves, hermit should improve with it.

How I’m using it right now:

Paulinho — watches my Home Assistant setup, spots repeated patterns, energy consuption etc and proposes automations.

Amélia — helps reconcile emails / payments / reports

Tars — builds daily briefings from sources I care about: X, etc.

Rex — tracks training patterns and recovery

Maria — handles content planning/reporting for my wife’s business through Discord

Same plugin. Same subscription. Different assistants. Each one runs independently in Docker and can ping me when it needs input.

What it actually does:

keeps Claude Code running as an always-on assistant through a heartbeat & routines system

sets up the official channels to talk to it through discord & telegram

runs recurring checks / routines using claude code crons

uses a proposal pipeline

reflects on repeated friction and proposes self-improvements you can approve

builds up useful knowledge over time instead of starting cold every session

A big influence on the knowledge side was Andrej Karpathy’s LLM wiki idea: raw observations get distilled into compiled knowledge the hermit actually reuses.

Setup is simple:

Install the plugins:

Run the setup wizard: /claude-code-hermit:hatch

Docker wizard setup: /claude-code-hermit:docker-setup

Obsidian wizard setup: /claude-code-hermit:obsidian-setup

Most of the features & bug fixes from it's development came from the other hermits proposals, which is pretty cool.

Repo: github.com/gtapps/claude-code-hermit

Feel free to give it a try.

[留言]

为什么值得关注

能改变理解方式，而不只是重复常识；符合当前抓取需求；它提供了新的理解或解释，而不只是表面观点

来源：reddit，领域：projects，保留分：0.60