Sophia NLU is an on device, low compute, Rust based natural language understanding engine that acts as a conversation agent. It offers the fluidity of a LLM without the compute, requiring only 160MB of RAM, no GPU. Handles unlimited devices, multiple intents, ambiguity, contextual awareness, millisecond latency, doesn't connect to the internet, and never calls home.
Upgrade to v1.2 just released with details at: https://nlu.tech/r/ha_v10200
This upgrade comes with many improvements:
- Sophia will now ask for clarification if it's uncertain which entity you meant. * State persistence. For example, if you ask how many lights are on in the kitchen, you can now say "turn them off for me" in the next message and Sophia will remember which lights you mean. * Help improve Sophia! When Sophia doesn't correctly understand a message, simply tell Sophia the last message was wrong. It will be logged, and once in a while, manually upload all pending through the Log menu of the web interface for review and repair. * And more... Grab a free trial of Sophia, and be setup within minutes at: https://nlu.tech/ha/
After the free trial, you can either pay a small fee or simply refer a single paying customer and get your lifetime license absolutely free. Essentially, a 2 for 1 deal.
The built-in Assist is very rigid as it only allows for single intents, and requires the friendly names of devices. LLMs require a GPU, hallucinate and give probabilistic mismatches. And nobody wants to send their messages over the wire to big tech. Sophia solves this pain point with quality and privacy.
Hand crafted, not vibe coded. This is actually 2.5 years of R&D with many months and upgrades for the HA edition now. I have huge ambitions for this tech, so am 100% here to ensure you get the best software possible and it works exactly as you need. If there's any issues at all, just simply let them be know, and they will be resolved.
My inbox is always open, if you need me, I'm at matt@cicero.sh. I've had a very rough handful of years since going suddenly blind, so if you could do me a favor and share this around it would be greatly appreciated.
Comments URL: [https://news.ycombinator.com/item?id=48499703](https://news.ycombinator.com/item?id=48499703)
Points: 2