An assistant connected
to my environment.
I connected a local language model to web search, Windows diagnostics, SSH, email and calendar tools. Open WebUI is the conversational interface; my own PC handles inference.
The interesting part was getting from “I don't have access” to a real tool call, with an actual result that I can verify.
Explore the agent architecture ↗- Request and contextThe model selects a relevant tool.
- Data from my environmentPC diagnostics, search, Gmail or SSH.
- Control over actionsI approve email drafts before sending.
- The GPU becomes available againImage models are unloaded after rendering.