Model per job

OneCamp does a lot of AI work in the background: catch-ups, channel summaries,
morning briefings, team reports, meeting recaps and workspace memory. By
default all of it runs on your workspace's default model. Model per job
lets an admin send each kind of work to a different model.

Typical choices:

• A fast or local model for summaries and briefings, which run often and are
  short, so they cost little and arrive quickly.
• A model with a large context window for meeting recaps, because
  transcripts are long.

Set it up

1. Add the models you want to the allowlist in Admin, AI models.
2. In Model per job, choose a model for each kind of work, or leave it on
   Workspace default.
3. Press Save.

The rules

• Only models on the allowlist, enabled and with their provider enabled, can be
  chosen.
• Local-only mode still refuses cloud models: a cloud model chosen here is
  replaced by your local default, and the refusal is recorded.
• If a chosen model can no longer be used, for example because it was removed
  from the allowlist, that work runs on the workspace default instead of
  failing, and the server logs why.
• Choices someone already made are never overridden: a person's own chat uses
  the model they picked, an agent uses its own model, and a channel with a
  pinned model keeps it.