AI cost and usage observability

AI cost and usage observability gives you a unified view of AI activity and cost across your Snowflake environment. It is broken down by agent, user, model, and team. It replaces manual dashboards and spreadsheet exports with three connected views:

An AI usage dashboard.
A list of every AI agent and surface running in your account.
A detail page for drilling into any one of them.

What it covers

AI cost and usage observability tracks Snowflake’s native AI surfaces, including:

Cortex Agents (custom agents registered in your account)
Cortex Analyst and Cortex Search
Cortex Code (Snowsight, CLI, and Desktop)
AI SQL functions (for example, AI_COMPLETE, AI_SUMMARIZE, and AI_EMBED)
Snowflake Intelligence / CoWork

All of this is tracked using metadata pulled from Snowflake, not your underlying data nor raw query results.

AI cost and usage observability does not cover AI activity that happens outside Snowflake. If a third-party AI tool or application calls into Snowflake (for example, an external LLM app querying your account), that activity is not attributed nor shown here.

Who can see this

AI cost and usage observability is available to all roles in your Slingshot tenant.

ℹ️

This feature only shows usage and cost data for Snowflake accounts that have been onboarded to Slingshot. Accounts that aren’t onboarded won’t appear — you won’t see partial or zeroed-out data for them.

AI Usage dashboard

Get a holistic view of AI spend across your onboarded Snowflake accounts.

How to access the AI usage dashboard

From the left navigation, select Dashboards, and then AI Usage.

What the dashboard shows

Summary metrics: Blended Total Cost, AI Inference Cost, Data Compute Cost, Total Tokens, and Average Cost per Request.
Daily Cost Trend: A chart of cost over time. Select a day to filter the rest of the page to that date.
Who Uses What: A breakdown table with tabs for Users, Entity (agent, surface, or function), Model, Role, and Product, each showing Cost, Credits, Requests, and Tokens.

Cost is shown as a Blended Total, combining two charge types:

AI Inference Cost: LLM fees based on input/output tokens.
Data Compute Cost: Standard warehouse credits used running SQL for AI-tagged queries.

These two costs are independent. An agent can incur inference cost without compute cost (if it calls a model without running SQL), or the reverse (if it runs queries without calling a model).

Filter and adjust the date range

Select Add Filter to filter by Role, User, Account, Warehouse, Entity, Model, or Product.
Use the date range picker in the top right to set a custom range, or choose a preset:
- Last 7 days
- Last 30 days
- Last 90 days
- This month
- Last month

AI Agents list

See every registered Cortex Agent, AI surface, and AI function running in your account.

How to access the AI agents list

From the left navigation, select Platforms > Snowflake, and then select the AI Agents tab.

The list shows, for each agent/surface/function over the trailing 30 days:

Name: As well as its database and schema.
Type: Agent, Surface, or Function.
Cost Trend: A sparkline of recent cost.
Cost, Credits, and Requests.
Last seen

ℹ️

An External row may appear in the list. This represents Cortex Agent API usage that isn’t tied to a specific registered agent.

Use the filter flyout menu to narrow the list by type (Agents, Surfaces, or Functions). This view is fixed to the trailing 30 days and doesn’t support a custom date range.

Select any row to open that agent’s detail page.

Agent detail page

Drill into a single agent to see its cost history, who’s using it, and what it’s doing.

How to access the agent detail page

Select an agent from the AI Agents list, or from the Who Uses What table on the AI Usage dashboard.

The detail page shows:

Stat cards: Total Cost, Total Credits, Total Tokens, Requests, and Average Latency
Token Breakdown: Usage split by model.
Cost by User: which users or systems triggered the agent, and their associated cost.
Activity and lineage: The tasks that triggered the agent and the tables it read from or wrote to over the trailing 30 days. This reflects anything that occurred at least once during that time window.
Tasks: A table of the agent’s triggered tasks, with links to underlying query details.

Use the date range picker (custom range, or presets for Last 7/30/90 days) to adjust the cost and usage charts. The activity and lineage view is always scoped to the trailing 30 days, independent of the date range picker.