pearl pebble
Hey HA community! 👋
We just open-sourced Wactorz — a Python actor-model multi-agent framework built from scratch for exactly the kind of use case you're all tinkering with.
The core idea: you describe what you want in plain English, an LLM writes the code, wraps it in a <spawn> block, and a new live agent appears on the fly. No restart. No manual wiring. Just a sentence.
👇 Here's a demo showing it in action — spawning a YOLO person-detection pipeline and wiring it live to Home Assistant automations:
https://www.linkedin.com/posts/panagiotis-kasnesis-6a396a4a_opensource-ai-multiagent-activity-7447660515079942144-Y-cv?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAp2-VsBun2a3HFnAl88RKStDbjw-_Dm1WY
Wactorz:
✅ Runs on modest hardware, fully offline (Raspberry Pi friendly)
✅ Uses MQTT as the backbone — the same nervous system already running in most of HA setups
✅ Survives crashes and auto-restores state
✅ No infrastructure code — just describe the automation you want
One thing that makes Wactorz different: there's no prescribed agent paradigm. Agents are just Python spawned at runtime, so you can mix rule-based logic, reinforcement learning, or LLM-driven agents — all talking to each other via MQTT and direct messaging.
Some things under the hood worth knowing for HA folks:
🏗️ PlannerAgent — breaks complex tasks into dependency graphs, fans them out in parallel
📡 Reactive pipelines — "if a person is detected, turn on the lights" — built and wired automatically
🌐 Edge deployment — spawn agents on a new node over SSH with one command
🔁 Rolling conversation summarization — context survives restarts
💸 LLM cost tracking
💬 Talk to your agents via Discord, Telegram, WhatsApp, REST, or CLI
🔗 Site: https://waldiez.github.io/wactorz
📚 Docs: https://waldiez.github.io/wactorz/docs
👉 GitHub: github.com/waldiez/wactorz
Would love to hear from anyone already running automations with MQTT or experimenting with local AI in Home Assistant!
