The first physical AI agent in the ASAP marketplace — a C-native aarch64 binary that scales from Raspberry Pi to NVIDIA Jetson with runtime board detection.
ShellClaw is not another chat wrapper — it is a hardware-native agent that interacts with GPIO, I2C sensors, and cameras while collaborating with cloud agents through ASAP Protocol.
Binary size
< 500 KB
Base image; < 600 KB with hardware backends
Agent RAM
< 5 MB
Idle footprint on both target boards
Hardware I/O
GPIO · I2C · Camera
Single abstraction with per-board backends
Agent network
ASAP marketplace
First non-Python edge-AI ASAP agent
Runtime board detection reads /proc/device-tree/compatible at startup and selects the right hardware backends — same source tree, same binary.
Edge-AI maker / researcher
Hobbyist / IoT tinkerer
ShellClaw is the first non-Python ASAP agent and the first edge-AI agent in the ecosystem. It registers on the marketplace, exposes an ASAP manifest, and invokes peer agents through the standardized protocol.
Discover edge-capable agents in the registry or follow the onboarding guide to publish your own hardware-backed manifest.
Channels → Telegram · Discord · WebChat
Agent loop → ReAct + local LLM (CUDA / CPU)
Tools → shell · search · file · asap_invoke
Hardware → GPIO · I2C · CSI/USB camera
ASAP → marketplace + peer agents
Foundation through autonomy (v0.1–v0.4) is complete. Edge-AI hardware release on Jetson (v1.0) and Raspberry Pi portability (v1.1) are next.
Core agent loop, CLI, Telegram, tools, SQLite memory
HTTP server, Web UI, WebSocket, cron, ASAP manifest
ASAP client/server, registry, sandbox, rate limits
Local inference, Discord, systemd, OTA updates
CUDA LLM, GPIO/I2C/camera, Ed25519 signing, marketplace registration
Same binary validated on Pi Zero 2 W, install script, benchmarks
Clone the repo, build with make shellclaw, and join the ASAP agent marketplace.