AI Usage Tracking¶
Monitor and control AI and LLM usage across all cloud providers.
Partial Availability
Federated AI access and account-level budget governance are available today. Per-model metering and AI-specific spend caps are coming Q2 2026.
Overview¶
As organizations adopt AI services across AWS (Bedrock, SageMaker), Azure (OpenAI Service, ML), and GCP (Vertex AI, Gemini), tracking usage and costs becomes critical. AI Usage Tracking extends RosettaOps to provide visibility and governance over AI and LLM consumption across all connected clouds.
According to the State of FinOps 2026, 98% of organizations now manage AI costs (up from 31% in 2024). FinOps for AI is the number one forward priority for cloud-operating organizations.
Available Today¶
The following capabilities are operational:
| Capability | Description |
|---|---|
| Federated Bedrock access | Researchers access AWS Bedrock through sandboxed cloud accounts managed by RosettaHub |
| Account-level budget governance | Real-time budget enforcement on the cloud account containing AI services |
| Sandbox isolation | SCP enforcement ensures researchers cannot escape their sandbox or access unintended services |
The architecture for governed AI access is in place -- researchers access foundation models through RosettaHub-managed accounts with full budget and permission controls.
High-Priority Roadmap¶
Usage Visibility¶
| Feature | Description |
|---|---|
| Per-model tracking | Usage metrics broken down by AI model (GPT-4, Claude, Gemini, Llama, etc.) |
| Per-user attribution | Track which users and projects consume AI resources |
| Token and cost metrics | Monitor input/output tokens, API calls, and associated costs |
| Cross-provider view | Unified view across AWS Bedrock, Azure OpenAI, GCP Vertex AI, and self-hosted models |
Governance Controls¶
| Feature | Description |
|---|---|
| Spend caps | Set maximum AI spending per user, project, or organization |
| Model restrictions | Control which AI models are available to specific users or teams |
| Rate limiting | Configure request rate limits to prevent runaway automation |
| Approval workflows | Require manager approval for access to expensive models or high-volume usage |
Reporting¶
- Usage dashboards -- real-time and historical AI consumption metrics
- Cost allocation -- attribute AI costs to projects, departments, or cost centers
- Trend analysis -- identify usage patterns and forecast future AI spend
- Anomaly alerts -- flag unusual consumption spikes
RosettaOps AI Agents (Roadmap)¶
RosettaHub is building MCP-powered ResOps Agents -- AI agents embedded in the RosettaHub desktop application that provide natural-language interfaces to cloud operations:
| Agent Domain | What It Automates |
|---|---|
| CloudOps | Infrastructure provisioning, scaling, and lifecycle management |
| FinOps | Budget optimization, cost anomaly investigation, savings recommendations |
| DevOps | Formation CI/CD, image pipeline automation |
| GreenOps | Carbon-aware scheduling, idle resource detection |
These agents use the Model Context Protocol (MCP) to interact with RosettaHub's APIs, providing researchers and administrators with conversational access to platform capabilities.
Why This Matters¶
AI services can generate significant and unpredictable costs. A single misconfigured automation can consume thousands of dollars in API calls within hours. AI Usage Tracking applies the same real-time cost enforcement that RosettaOps provides for compute and storage to the emerging category of AI services.
RosettaOps Tier¶
| Tier | Capability |
|---|---|
| Observe | View AI usage metrics, track costs per model and user, receive alerts |
| Govern | Set spend caps, restrict models, configure rate limits |
| Automate | Auto-block usage when caps are reached, enforce approval workflows |
Related Topics¶
- Cloud Operations Overview -- The governance layer of the Supercloud
- Cost Management -- Real-time cost tracking and budget enforcement
- The RosettaOps Model -- Observe, Govern, Automate progression
- For Data Science -- AI and ML workloads on the Supercloud