Model access
Which AI models your agents may use.
Who can do what
| Task | User | Agent owner | Admin |
|---|---|---|---|
| Choose a model for an agent, from what's enabled | — as Editor | ✅ | ✅ |
| Enable or disable models for the organization | — | — | ✅ |
| Add models to the platform catalogue | — | — | — |
Organization → Models. The catalogue itself is maintained by the platform; you control which of it your organization can reach.
Two gates
A model is available to your agents when both are true:
- The platform has published and enabled it.
- Your organization hasn't disabled it.
The default is allow: a newly published model is available unless you turn it off. Toggle models individually or in bulk.
Your plan doesn't restrict the catalogue
A common assumption, and it's wrong: being on the free tier does not hide paid models. Free-tier organizations can enable and use any published model. What bounds a free organization is its lifetime spend allowance, not a narrower menu — cost is enforced at runtime by budgets, not by tier.
So if you want to keep spend down, the control is a budget or a disabled model, not the plan.
Choosing models for an agent
Each agent has a Model in Agent → Settings → General, chosen from what you've enabled. The default routes intelligently rather than pinning one model.
Pin a specific model when the agent's work depends on it. Otherwise leave the default — it adapts as the catalogue changes, and you won't be left on a deprecated model.
Disabling a model that's in use
Agents pinned to a model you disable can no longer use it. Before disabling something widely used, check Usage & spend — the per-agent breakdown expands to show usage by model, so you can see exactly which agents would be affected.
Bringing your own key
With BYOK, models from that provider run on your account and Delegate adds no fee. Availability still follows the two gates above.