New in Mavvrik: Agent Studio
Build agents that start with your full cost picture
You don’t need another general-purpose agent. What you need is your complete cost and usage context, across AI, cloud and on-prem, built into every agent. That’s what Agent Studio gives you, with agents that come with it and ones you build yourself.
Clone any agent and change the instructions, skills and tools. Partner agents keep their skills locked.
Custom agents
Build an agent by describing the job
Describe the job the way you’d explain it to someone on your team, like a weekly margin check by product line or a monthly anomaly summary for a client. The Mavvrik Assistant turns it into an agent, and you approve what it writes. It already knows your costs, usage, owners and budgets, so the first run works from your real numbers.
| Instructions | Every Monday, check gross margin for each product line against its cloud and AI cost. Flag any line under 60% and name the owner. |
| Skills | Margin check · Owner lookup |
| Tools | Mavvrik cost data · Jira |
| Channel | #finance in Slack |
| Schedule | Mondays, 8:00 AM |
Two product lines are under 60% this week. Both moved because AI inference cost grew faster than revenue.
| Product line | Revenue | Cost | Margin |
|---|---|---|---|
| Search APIOwner: Search | $410K | $188K | 54.1% |
| AnalyticsOwner: Data platform | $265K | $112K | 57.7% |
| StorageOwner: Infrastructure | $520K | $146K | 71.9% |
From insight to action
Out-of-the-box FinOps agents for routine cost work
Agent Studio agents do more than answer questions. They take on the repetitive tasks that eat up your week, and they check with you before changing anything.
| Task | What the agent does |
|---|---|
| Resource tagging | Finds missing and inconsistent tags, proposes clean mappings and creates virtual tags in Mavvrik once you approve them |
| CapEx to OpEx | Turns hardware, power, cooling, rack and labor costs into hourly running rates for VMware hosts, data centers and on-prem Kubernetes clusters, so on-prem and private AI sit next to cloud and AI spend |
| Private AI token rates | Turns GPU and model hosting costs into per-million-token rates, so you can compare a self-hosted model with a public API |
| AI spend review | Finds teams defaulting to expensive models, prompts that would be cheaper with caching, and LLM spend that could count toward cloud commitments you already have |
| Client alert write-ups | Turns an anomaly alert into a short summary for the client: what changed, why, and who owns it |
| Commitments | Watches reservations and savings plans before they expire, so someone decides in time |
| Quarterly reviews | Builds the 90-day cost review for executives from your own numbers |
Because Mavvrik already brings in your on-prem and private AI costs, these agents work across your data centers and GPU clusters, not only cloud.
What every agent knows
Your costs, already connected.
- Cost
- Usage
- Owners
- Budgets
- Anomalies
- Returns
Context
Your data is already there
Cost, usage, budgets, recommendations and anomalies from every cloud, data center and AI service you connect to Mavvrik.
Evidence
Answers you can check
Every answer shows the steps the agent took and the numbers behind them.
Control
Runs within your rules
Agents work inside your Mavvrik permissions. They read and recommend on their own. Changes, like creating virtual tags, wait for your approval and are logged.
For partners
Build an agent once and run it for every client
Turn the work your team repeats across clients into agents: the monthly savings check, the QBR, the anomaly write-up, the tagging review. Pick which agents go to which clients, then run them on a schedule or share them so those clients run them on their own data.
- Start with what’s in the boxOpen an engagement with the FinOps Maturity Assessment, Quarterly Review, Cloud Cost Compare or CapEx to OpEx agent, then add your own policies, like your tagging standards.
- Match agents to clientsChoose which clients each agent runs for, and set schedules and channels for each one from your partner tenant. Each run uses only that client’s data.
- Share to all clients or the ones you choosePublish to every client tenant or select ones. Clients you add later can get them automatically.
- Your playbook, delivered as your productClients get your results, and the skills behind them stay yours, even when a client copies the agent.
- Keeps working after the engagementAgents you build during an engagement can stay in the client’s tenant and keep running.
| Client | Top finding | Savings / mo |
|---|---|---|
| Retail groupAWS, Azure, Copilot | Unused Azure reservations | $51.4K |
| Regional bankAzure, on-prem GPUs | Idle GPUs in dc-east | $38.2K |
| ManufacturerAI factory, NVIDIA B300s | Low GPU utilization | $27.6K |
| Health networkGCP, VMware | Oversized VMware VMs | $12.9K |
| 34 more clientsEach gets its own report | $481.9K |
FAQ
Questions you might have.
What is Agent Studio?
Mavvrik Agent Studio provides out-of-the-box FinOps agents and lets teams and partners build their own on Mavvrik’s cost and usage data across AI, cloud and on-prem. Agents and skills run inside Mavvrik, and it’s built for AI and hybrid cost governance work, not as a general-purpose agent platform.
What FinOps work can AI agents take on?
Mavvrik Agent Studio includes out-of-the-box FinOps agents for tagging, CapEx to OpEx rates, private AI token rates, forecasting, quarterly reviews and FinOps maturity assessments, plus a studio to build your own. The agents run on Mavvrik’s cost and usage data across AI, cloud and on-prem, and changes wait for your approval.
What’s the difference between a skill and an agent?
A skill teaches an agent how your team works, like your tagging policy or the format of your quarterly review. An agent puts instructions, skills, tools, a channel and a schedule together to do a FinOps job. In Mavvrik Agent Studio, skills and agents both work on your Mavvrik cost and usage data.
Which agents come with it?
Mavvrik Agent Studio comes with ten out-of-the-box agents: FinOps Analyst, Mavvrik Assistant, Quarterly Review, FinOps Maturity Assessment, Tag Advisor, Cloud Cost Compare, Private AI Token Rates, CapEx to OpEx, Forecasting and Savings Policy. You can run them as they are, or clone any agent and change its instructions, skills and tools to fit how your team works.
Does Agent Studio work with on-prem and private AI costs?
Yes. The CapEx to OpEx and Private AI Token Rates agents work on the data center and GPU costs Mavvrik already brings in. Because Mavvrik Agent Studio uses that same data, any agent you build can include on-prem and private AI costs alongside cloud and public AI spend.
Do I need to write code to build an agent?
No. In Mavvrik Agent Studio, you describe the job in plain English, the way you’d explain it to someone on your team. The Mavvrik Assistant turns it into an agent, and you approve what it writes. Then pick the data and tools it can use, and set when and where it runs.
Can agents make changes in my environment?
Yes, within your Mavvrik permissions. Agents in Mavvrik Agent Studio read and recommend on their own, and they can take FinOps actions, like creating virtual tags to clean up inconsistent tagging. Anything that changes data waits for your approval, and every action is logged.
Can partners build agents for their clients?
Yes. Partners use Mavvrik Agent Studio to build skills and agents for their own practice and share them to all client tenants or selected ones. Clients see the agent but not your skills, and each run uses only that client’s data, so one agent can serve every client without mixing their numbers.