AI Usage & Model Settings

The AI Usage page records AI activity across your organisation, including Ru conversations and background tasks, brief and resource planning, search, and image and video generation. Organisation owners and administrators can open it from Settings → AI Usage.

Choose 7, 30, or 90 days to change the reporting period. What you see depends on how your organisation pays for AI.

Platform credits

For an organisation using platform credits, the report focuses on what was used and how many credits it consumed. It does not show provider model names, token counts, or the provider’s underlying cost. The same protection applies when an older audit link is opened, so technical provider details are not exposed through another reporting path.

The available tabs are:

Overview: total credits, retail value, calls, images, video seconds, and activity over time. The Credits used total counts only credits charged to the wallet. When document indexing occurred, a note shows its separate uncharged credit value.
Categories: credit use grouped into chat and agents, images, video, and background AI (including search embeddings). Client workspaces see the same credit view as the agency that manages their billing.
Users: credit use by person.
Features: credit use by Runnit feature.
Agents & Credits: credit-charged planning and assistant runs.
Audit Log: individual events with their status, category, credits, and retail value. Owners and admins see every event in the organisation and can filter by person. Other members see their own events only.

Use these views to find what is drawing down the balance, then open AI Credits to check transactions, set a warning, or top up.

Document indexing is recorded but is not charged. This means the report can show indexing activity and its value without reducing the wallet. The charged total is the amount to use when you reconcile the balance.

Unmetered or own-key access

Unmetered organisations and organisations using their own provider keys see the detailed technical report. It includes provider models, input and output tokens, provider cost in US dollars, images, video seconds, calls, and events. Credit-only fields are not shown.

The full report includes:

Overview for headline totals and daily charts;
Models for model, provider, feature, token, call, and cost breakdowns;
Users and Features for attribution;
Agents & Credits for assistant runs with a configured flat credit rate; and
Audit Log for individual event details, status, latency, and recorded output.

If an event used a model without a configured price, it is kept in the report and marked as unpriced. Model prices follow their effective dates. A price change applies to new activity; existing usage keeps its recorded cost.

Model settings

Runnit’s platform operators can also see Model Configuration and pricing controls. Operators can set platform-wide defaults or apply model settings to one organisation only. Organisation administrators cannot change these settings either way. If your organisation needs a different model for a feature, ask your Runnit administrator.

Model Configuration selects the model used for each internal AI feature. The General Chat setting controls Ru in conversations and background Ru tasks. Brief Analysis and Resource Planning have their own settings. Image and video settings control each media capability separately.

Retired models cannot be selected. The model picker lists only supported models, and saving a retired one is refused with a message naming the lowest supported model. If an older model is still named in a deployment’s configuration, the platform uses the lowest supported model instead of failing the call, and the Model Configuration page shows the retired value that was replaced so it can be corrected. Reports keep the model recorded against past activity, so historical costs stay accurate.

Configure AI recovery

Enterprise owners and admins can configure recovery under AI Usage → Recovery. Platform admins can also select an enterprise organisation on that tab. Other plans show read-only defaults.

The default sequence allows five attempts: two with the primary model, then one each with gemini-flash-latest, deepseek/deepseek-v4-flash-latest and gpt-5.6-luna, in that order. Success stops the sequence. The first fallback is still attempted when it matches the primary model.

Set Primary model, or leave it blank to use each task’s selected model. Entering a model applies it to the tenant and its client workspaces.
Set Primary attempts from 1 to 5.
Enter up to three fallback models in order. Each allows 1, 2 or 3 attempts. Use Remove fallback or Add fallback to change the list. The total cannot exceed 10.
Click Save recovery settings. New AI requests use these settings. Models without supported IDs or complete active pricing cannot be saved.

Restore defaults removes the override. An Enterprise downgrade also stops applying it. Organisations using their own provider keys retry the primary only. Recovery retains completed actions. Access, credit and cancellation errors stop the request. Outages can exhaust all attempts, including when models share the same connection. The audit retains each attempt’s HTTP status, or connection error codes when no HTTP response arrived. Chat shows retry and fallback progress and the final error. See Chat Assistant.

Provider keys and embeddings

Changing the search embedding model requires existing search indexes to be rebuilt before results are reliable again.

Platform provider keys are also restricted to platform operators. An organisation in own-key mode manages its separate credentials under Settings → AI API Keys instead. See AI Credits for the setup steps.

For organisation controls unrelated to AI, see Advanced Settings.