Cyifq Bot is being connected to a model running on a local PC. The current arrangement uses a private relay between the always-on server and the computer, allowing the public web service to request a response without exposing the computer’s model endpoint directly to the internet.
Owner Daniel Bayliss wanted a larger model than the one previously running on the T630 server. The PC now hosts a Qwen3 4B Instruct based model configured for Cyifq Bot. The older T630 model service has been stopped so the server is not carrying an unused model workload.
The design includes Wake-on-LAN support: the server can send a wake request if the PC is asleep, then wait for the model relay to become available. A successful message while the PC is already awake has been observed; a full wake-from-sleep cycle still needs a separate real-world check.
This is an infrastructure update, not a promise that every request will be instant. A cold model load, a sleeping computer or a network interruption can lengthen the first reply. The team is continuing to monitor the experience.