Just some neat wins from today. First up - esdin

[neat magnet]

As much as I love this, it seems very foreign to my usage of openClaw. Are you using a local model for this? Paying crazy bills in token usage? My openclaw barely can sustain replying to more than 1 person at the same time

[west iris]

This is interesting, whats the hook to Sonos?

[alpine skiff]

Mimimax 2.5 for a lot of it, stays well under the 100 calls / 5 hours for their entry level plan. The models themselves use either Minimax 2.5 for coding/general tasks or gpt-4o-mini for creative and gpt-5 for high reasoning creative, at < $4 per day currently.

@west iris It uses soco for everything. First it hosts a small local server with the media to play. It then discovers all speakers, gets group coordinators, and gets their queue position / transport state details, then pauses all of them. Calls play_uri providing the media server, waits the duration of the media plus a buffer, and then resumes all states.