RouterClaw is a Go-based autonomous AI agent designed to run on OpenWrt routers with about 32MB RAM, featuring Telegram and Google Workspace integration, WoL, web browsing, and direct shell access.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
RouterClaw is an autonomous AI agent written in Go, designed to run on resource-constrained embedded systems like OpenWrt routers. It acts as an always-on, intelligent brain for a home network, with Telegram bot interface, Google Workspace integration, web browsing, OpenWrt shell access, and Wake-on-LAN.
How it works
The project includes a Go binary built for MIPS (GOOS=linux GOARCH=mips GOMIPS=softfloat) to run in RAM on 32MB systems. It provides modules for an orchestrator, Telegram bot, Google integration, web interactions, and LLM integration. The repository describes a minimal OpenWrt setup and a cross-compilation/deployment flow to transfer the binary to the router and run it in the background.
Getting started
Configuration is via a config.json storing API keys (DeepSeek, Telegram, Google) and worker node MAC addresses. Build and deployment steps shown:
- Compile for MIPS: "env GO111MODULE=on GOOS=linux GOARCH=mips GOMIPS=softfloat go build -trimpath -ldflags '-s -w' -o routerclaw ."
- Transfer to router: "scp routerclaw root@192.168.1.2:/tmp/"
- Run in background: "cd /tmp && ./routerclaw /root/config.json &"
The README also describes a minimal OpenWrt image build process due to limited flash (4MB) and RAM constraints, including a command sequence to build a tiny image and flash via UART.
Recent releases
There are no releases listed (latest 0: none).
Traction
Stars: 26.
Behind the repo
No startup/company linked in the provided facts.
Caveats
License: GPL-3.0. The project is described as experimental and capable of executing arbitrary shell commands on the host, with safeguards to prevent reboot, but caution is advised on production networks. The README notes the need to build a minimal OpenWrt image for very constrained devices, implying manual setup steps and hardware-specific considerations.