AI Usage & Model Settings

The AI Usage page records every AI event in your organisation: assistant and agent conversations, agent runs, brief analysis, image and video generation, and search embeddings. Organisation administrators and owners can open it from the sidebar under Admin → AI Usage.

Use the range control (7d, 30d, or 90d) to set the reporting period. All costs are shown in US dollars.

Overview

The Overview tab shows headline totals for the period: input and output tokens, total cost, agent credits, images and video seconds generated, and the number of events. Charts break down daily token usage (input vs output) and daily cost.

If any events used a model with no configured price, a warning shows how many are unpriced so you can fix the price list.

Models and pricing

The Models tab lists each model used in the period with its provider, the features it was used for, call counts, token totals, and cost.

The Model Pricing table on the same tab holds the rates used to cost usage. You can edit a model’s input and output rates (per million tokens), the per-unit price for images, video seconds, and agent calls, and the credits charged per agent call. After adding or fixing a price, choose Recompute Costs to backfill any events that were recorded without a price.

Note

Events with unknown models are never dropped. They are stored without a cost and flagged as unpriced until you add a price and recompute.

Model Configuration

The Model Configuration tab controls which model each internal AI feature uses, grouped by area (chat assistants, Studio developer agents, briefs, content analysis, search embeddings, and image and video generation). Each row shows the default model and lets you set an override; leave the override blank to use the default. Overrides apply platform-wide and take effect within about 30 seconds.

The agent rows (AI Agents, Brief Analysis, and Resource Planning) set the default model for agent runs. An individual agent can override its own model in the agent editor, and that setting takes precedence.

The model you choose also selects the provider automatically. For example, choosing a Claude model routes that feature through your Anthropic key, and choosing a Gemini model routes it through your Google key.

The Image & Video Generation group sets the fal model used for each generation capability: image generation, image editing, video generation, and image to video. The override field suggests the supported models, and both video capabilities default to Gemini Omni Flash.

Caution

Changing the search embeddings model changes how content is indexed, and existing search indexes must be rebuilt before results are reliable again.

Provider API keys

The Provider API Keys panel on the same tab lets administrators supply their own credentials per provider: OpenAI (with an optional base URL override), Anthropic (Claude), Google (Gemini), and fal (image and video generation).

• Keys are stored encrypted and are never shown again after saving; only a masked hint (such as ••••1234) is displayed.
• Each provider shows whether it is using a custom key, an environment default, or has no key.
• Saving or removing a key takes effect without a restart. Adding a fal key enables image and video generation immediately.

Users, Features, and Agents & Credits

• Users shows who is using AI: LLM calls, tokens, media generated, agent calls, credits, and cost per person.
• Features shows what the AI is used for across the organisation, such as general chat, brief building, embeddings, Studio agents, and media generation.
• Agents & Credits shows runs per agent. Each agent run is charged as credits on top of the tokens its underlying model calls use.

Audit Log

The Audit Log tab is a filterable browser of every recorded event. Filter by type and status, and open an event to see its tokens, cost, latency, and recorded output.

For the settings that control projects and clients rather than AI, see Advanced Settings.