Platform Settings

Platform Settings is where administrators configure the AI runtime for the whole tenant — which model the agents use, a fallback model, and token/cost budgets. You can find it under Configuration > Platform Settings in the sidebar.

Every setting is optional: fields left blank inherit the platform’s environment defaults. A badge next to each field shows whether the effective value comes from your tenant configuration or the platform default, and the resolved effective value is displayed underneath. Changes take effect on the next agent turn — no restart or redeploy needed.

Models

• Default model — The model used for every agent turn. Pick from the curated list (each entry shows the model id, context window, and a short note), or choose Inherit environment default.
• Fallback model — Optional. When set, a turn whose primary model errors is retried once on the fallback model. Pick a different model than the default.
• Thinking — Whether the model gets an internal reasoning phase before answering. Adaptive (recommended) lets the model decide when and how much to reason; Disabled skips reasoning for faster, cheaper answers at some quality cost on complex tasks. Applies to Anthropic models; reasoning tokens count against the per-call output limit below.

Per-Execution Budgets

Hard caps for a single agent turn, cumulative across every LLM call the turn makes:

• Max tokens per execution
• Max cost per execution (USD)

A turn that exceeds them is stopped.

There is also a Max output tokens per LLM call setting — the ceiling for a single model response (its reasoning plus its visible output), not the whole turn. Setting it too low is dangerous: the model can spend the entire ceiling reasoning and produce no visible answer, which stops the turn with an error. The minimum accepted value is 8,000; leave it on the default unless you have a specific reason.

Daily Caps (Tenant-Wide)

Aggregate limits across all users and agents for the calendar day:

• Max tokens per day
• Max cost per day (USD)

When a daily cap is reached, new agent turns are blocked until the next day — the chat shows an error instead of starting the turn. Because usage is counted after each turn completes, a turn already in flight when the cap is hit is allowed to finish; it is the next turn that gets blocked.

Per-User Daily Overrides

Below the tenant-wide caps you can add per-user daily overrides — a tighter daily token and/or cost cap for a specific user. Pick the user from the directory picker, enter at least one of the two limits, and save. Remove an override to return the user to the tenant-wide caps.

Per-user overrides cap individual users; the tenant daily caps always remain the aggregate ceiling.

Saving

Click Save changes to apply. Every change is recorded in the tamper-evident audit trail with your user identity.