Analysis updated 2026-05-18
Control Home Assistant smart home devices using voice commands processed entirely on your own hardware.
Chain multiple tools together in a single spoken request.
Install community-built voice workflows from a security-reviewed Tool Registry.
Search the web by voice using DuckDuckGo without sending data to a third-party assistant.
| coreworxlab/caal | zhilin1112/yellowkey-bitlocker | tritano/ultraviewer | |
|---|---|---|---|
| Stars | 395 | 393 | 388 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | hard | — | easy |
| Complexity | 4/5 | — | 1/5 |
| Audience | developer | general | general |
Figures from each repo's GitHub metadata at analysis time.
Needs an NVIDIA GPU (or CPU/Apple Silicon fallback) plus a working n8n instance for automations.
CAAL is a self-hosted voice assistant that runs entirely on your own hardware, meaning your voice commands, credentials, and data never leave your local network. It is built on a platform called LiveKit Agents and uses a purpose-built AI model fine-tuned specifically for voice-activated tool calling. You speak to it, it decides which automation to trigger, and it executes that automation through n8n, a visual workflow tool that handles the actual credentials and API calls. What makes CAAL unusual is its security design: the AI model itself never sees your API keys or passwords. Instead, it just tells n8n which workflow to run and what parameters to use, and n8n handles authentication separately. This means even if someone tricks the AI into doing something unexpected, it cannot access or leak your credentials. You can control smart home devices via Home Assistant, search the web via DuckDuckGo, chain multiple tools together in a single spoken request, and extend CAAL's capabilities by installing community-built workflows from a shared Tool Registry where every submission goes through security review before publishing. It runs on NVIDIA GPU hardware via Docker, with options for CPU-only mode or Apple Silicon Macs. You can use local models via Ollama or connect to cloud providers like Groq or OpenRouter. A wake word ("Hey Cal"), an Android mobile app, and support for English, French, and Italian are all included out of the box.
A self-hosted voice assistant that keeps credentials safe by having n8n, not the AI model, handle authentication for triggered automations.
Mainly TypeScript. The stack also includes TypeScript, LiveKit, n8n.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.