Manage the assistant model key centrally on the admin machine #261

Open
opened 2026-06-07 00:21:43 +00:00 by mik-tf · 0 comments
Owner

Goal: manage the assistant model access in one place on the admin machine, the way the shared embedder and voice engines already work.

The shared embedder and voice engines run on the admin machine, and each tester reaches them through a per tester token, so their keys live in one place. We should do the same for the assistant model access: a shared admin side endpoint that testers call through, giving a single key to manage and rotate and central control of spend, instead of configuring model access on each machine. This is an operations and hardening improvement, not urgent, and fits after the current sandbox milestones.

Related: #256

Signed-by: mik-tf mik-tf@noreply.invalid

Goal: manage the assistant model access in one place on the admin machine, the way the shared embedder and voice engines already work. The shared embedder and voice engines run on the admin machine, and each tester reaches them through a per tester token, so their keys live in one place. We should do the same for the assistant model access: a shared admin side endpoint that testers call through, giving a single key to manage and rotate and central control of spend, instead of configuring model access on each machine. This is an operations and hardening improvement, not urgent, and fits after the current sandbox milestones. Related: https://forge.ourworld.tf/lhumina_code/home/issues/256 Signed-by: mik-tf <mik-tf@noreply.invalid>
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
lhumina_code/home#261
No description provided.