See the four coloured rectangles at the bottom ?
They are the members of the Zorgchat development project. One is human, two are local AI models (interpreter,intern), one is a online only frontier model, the brains of the project (partner).
There are others also, one is a image handling model. Any chat member can ask it to describe any image, and it’s devastatingly good. It sees things in images I can’t. But you can’t see them at the bottom as we haven’t spoken lately.
They can all talk to each other or any member of the chat. Someone in Brisbane could connect and talk to them as well, or me, if I had a Web facing IP and wasn’t locked behind the Starlink firewall.
For the non embedded here, in the window above, ‘tp’ (yours truely) has just asked the ‘intern’ “Describe how to set the clock on a STM32FO51.”
All that followed is the reply, all of it accurate because it didn’t ask Google, instead it:
- downloaded the STM32F051 Reference Manual PDF, and took it too bits
- made it into a stm32f0xx-rm.db of about a quarter of a MB in size. Just the text parts.
- searched the db for the relevant parts
- answered me.
The eagle eyed may have noticed I asked @interpreter and then it asked @intern who replied with all the specific model information.
What they won’t see is that I spoke normally to ask that question. One mouse click, then I spoke, one click at the end of my speech. Zero typing.
- Local AI, no cloud dependency
- Multi-agent team (interpreter, intern, partner, phi3)
- Voice input with interpreter restructuring
- Vision/image analysis
- RM database indexed from reference manuals
- Radio ducking/audio handshake
- All self-hosted in Fossil SCM
- Built by a 72-year-old electronics technician, not a programmer