A model is the reasoning engine behind a run. Okou handles the tools, the context, and the procedure; the model does the thinking. Which one is in play affects quality, speed, and how many credits a run consumes.
The model lives on the chat thread
The model picker sits beside the chat composer and controls the model for that chat. The selection is stored on the thread, not on the agent and not on the workflow.
- A new chat starts from your member default, and falls back to the workspace default when you haven't set one.
- Changing the model mid-chat applies to the next run in the thread; a run already in progress finishes on the model it started with.
- An automation chat picks up its model when it is first created, and later firings reuse it.
- In Slack, the /zero model command binds the model for the next thread you start — see Slack.
Agents and workflows deliberately have no model setting of their own. The thread is the single place to look.
What's available
The full catalogue, with what each model is good at and what it costs relative to the others, is on the Models page. It spans frontier reasoning models, cheaper high-volume models, and dedicated image and video models, and it changes as new models ship.
Two plan rules:
- Free includes three models. They're enough to build an agent, connect tools, and run real work.
- Pro, Team, and Enterprise open the full list.
Workspace admins choose which of those models members can actually pick, and which one is the workspace default, under Settings → Models.
Where the inference runs
A model can be routed more than one way, and the route is set by workspace policy rather than by the picker:
- Built-in — Okou runs the model and it bills against workspace credits. Nothing to connect.
- Your own subscription — each member connects their own Claude or Codex account and their runs go through it.
- A shared provider key or gateway — an admin supplies the credential once for the workspace.
Bringing your own provider is a paid-plan feature. Whichever route a model uses, tool calls, generation, and storage still bill against Okou credits.
The picker chooses a model. It does not override the route policy has assigned to that model.
Choosing one
- Match the model to the task, not to the price list. A cheap model on a long agentic run can cost more than an expensive one, because it takes more steps to get there.
- Use a cheap model for volume. Triage, classification, and bulk pre-filtering rarely need a frontier model.
- Use a frontier model for judgement. Long-horizon work, code, and anything where a wrong answer is expensive.
- Set a member default for the model you want most chats to start on, rather than changing the picker every time.
Troubleshooting
| What you see | What to do |
|---|---|
| Selected model is not available | The thread's model isn't usable under the current workspace policy or provider connection. Pick another model in the composer, or reconnect the provider under Settings → Models. |
| A model you expect is missing | Admins control availability per workspace. On Free, only the three included models appear. |
| Credits draining faster than expected | Check which model the thread is on. Rates differ substantially between tiers. |
What's next
- See Credits & billing for how model choice shows up on the bill.
- See Chat for the thread the model is attached to.