Feature Overview

TokenOps is a view within CoreStack’s FinOps module that surfaces token-level usage, cost, and rate analytics for AI services onboarded to the platform. It applies once a Platform Cost Trend (New) or Platform Summary widget bound to AI Service data has a token-based Consumed Unit configured. Without one, the widget shows Cost only. TokenOps is most valuable to FinOps Practitioners and FinOps Admins who need to track AI spend and consumption patterns across models. It does not track Claude API usage by Agent or API key. That level of attribution belongs to Governance > AI Services > Frontier AI, documented separately.

Prerequisites

Before you begin, ensure the following:

• Role: You have the FinOps Practitioner or FinOps Admin role assigned in CoreStack.
• Onboarding: An AWS account with Bedrock enabled is already onboarded to CoreStack. No additional connector or manual onboarding step is required — Bedrock usage is auto-discovered within 24 hours of being active.
• Access: You have permission to add and configure widgets on your dashboards.

Key Capabilities

1. Token-Based Consumed Units — Each provider’s token-bearing Consumed Unit becomes its own summary card:

2. Consumed Unit Configuration — A Consumed Unit is set from the widget’s Data Binding tab: turning on Override Filters reveals a filter Key/Value pair, where the Key is set to Consumed Unit and the Value is set to the token unit for the provider. Once set, the widget’s Usage and Rate metrics resolve to token counts instead of a cloud-usage unit (GB, Hrs, vCPU-Hours).

3. Primary/Secondary Metric Selection — One of Cost, Usage, or Rate is chosen as the Primary metric on the Cost Trend (New) widget — distinct from the standard Cost Trend widget. The Primary metric renders as the main bar graph, while the remaining two render as secondary line items on the same chart. For TokenOps widgets, all three metrics should be added as visible, rather than a subset.

4. Breakdown by AI Model — Data can be grouped by dimension, such as AI Model Name, to isolate a specific model’s consumption, cost, or rate over a selected time range.

5. Rate as an Average — Rate is calculated as an average cost per unit of consumption (for example, cost per 1,000 or 1 million tokens), recalculated based on the selected Time Range and Granularity.

6. Current Provider Scope — This Consumed Unit mechanism is active for AWS, Azure, and GCP accounts generally. For AI Services integrations specifically, it's confirmed for AWS Bedrock today; Anthropic Claude API usage is tracked separately under Frontier AI rather than TokenOps.

Frequently Asked Questions

Q: Why don’t I see Usage or Rate options on my AI Service widget?

A Consumed Unit must be set. Open the widget’s Data Binding tab, turn on Override Filters, and set the filter Key to Consumed Unit. Without one selected, the widget shows Cost only, and Usage/Rate options don’t appear.

Q: What does the Rate metric represent?

Rate is an average cost per unit of consumption (for example, cost per 1,000 or 1 million tokens), recalculated based on the Time Range and Granularity filters you’ve selected.

Q: Can I add more than one Consumed Unit to the same widget?

Based on the configuration flow, each widget’s Override Filters setting supports a single Key/Value pair, so a separate widget is needed per Consumed Unit you want to track side by side.

Troubleshooting

No data appears on the widget

Cause: No Consumed Unit has been selected in the widget’s Override Filters, or the AI service account hasn’t finished onboarding.

Solution:

1. Confirm the AWS account with Bedrock enabled is onboarded and active in CoreStack.
2. Allow up to 24 hours for auto-discovery to complete.
3. Open the widget’s Data Binding tab, confirm Override Filters is enabled, and confirm the filter Key is set to Consumed Unit with a Value selected.

If the issue persists, contact CoreStack support with the widget’s dashboard name, the AWS account ID, and the Consumed Unit value you selected.

Rate shows no variation over time

Cause: This can be expected if pricing hasn’t changed for the selected model over the chosen time range, or the Granularity filter is too coarse to show short-term changes.

Solution:

1. Select a specific model or dimension to isolate its rate trend.
2. Try a finer granularity (for example, daily instead of monthly).