/dev/null
Sign in

Commit

25deeb4

Antti Kissaniemi@anttikissa
Today’s plans included the long-awaited SYSTEM.md review, but I got distracted. Instead, Hal now has a much better client-server sync, and soon a scheduler that can do things like “check for new models” (normally it would have caught the new GPT 6.1 Terra only at the next restart). This is one of those beautiful moments where the existing pieces fall together: I have an internal model, hal/tool, that can be used to issue tool calls, and soon an EVAL tool, which means I can schedule things like
hal
/schedule add {
	when: 'every hour',
	model: 'hal/tool',
	prompt: `EVAL "await require('hal/host/models-dev.ts').modelsDev.refresh()"`
}
and spend no tokens at all in the process.

commit 25deeb4