# Vela AI Agent

An AI agent framework for the openvela embedded operating system, running on watches, glasses, speakers and similar devices, with multiple LLM backends, a ReAct tool-calling loop, an MCP client, and configuration tuned for devices with as little as 256KB of RAM.

Vela AI Agent is an AI agent framework built into the openvela embedded operating system, intended to run on any smart device running openvela - watches, glasses, speakers, headphones - with devices interconnecting over the OpenClaw Node protocol. It supports switching between several LLM backends, including Kimi, Qwen, DeepSeek, GLM, MiMo, OpenAI, Claude and OpenRouter-aggregated models, and a router mode that can hold up to four backends at once and pick between them automatically by task complexity or cost profile.

A ReAct-style agent loop with more than 35 built-in tools supports parallel tool execution, and the agent can be reached through a CLI, Feishu or WeChat bots, WebSocket, MQTT, or voice with speech recognition and synthesis, with each channel independently toggleable through Kconfig. A device-side MCP client connects to remote MCP servers over Streamable HTTP, and a Markdown-based skills system lets new skills be created through conversation. Low-memory optimizations include heap detection, memory pooling and streaming output.

It is built and run through the openvela build system and QEMU emulator, and requires enabling the AI Agent package in menuconfig.

Website: https://github.com/open-vela/packages_ai_agent