Price on the item, not in a list
Model, tool and human-decision prices live on the catalog items that incur them, so there's no separate price book to keep in sync and every figure traces back to a rate.
Agent Management · Cost Control
Know what every agent, run and approval actually costs.
Gamma prices the resources your agents consume and turns every run into a cost record you can attribute. Model tokens, MCP tool calls and the human decisions your agents wait on are all priced at the AI Gateway, then rolled up per agent, so you can read model spend, tool spend and human decision cost, set against the value each agent delivered.
4.6
(35 reviews)
4.6
(56 reviews)
Trusted to run APIs at the world’s top enterprises
RATP 
Walmart 
Blue Yonder 
Schneider 
Roche 
BMW 
Xylem 
Auchan 
Monday 
EY 
The Challenge
Agent spend grows fast. Understanding it doesn't.
As enterprises put agents into production, spend scatters across model providers, tool platforms and cloud consoles. Most tools price the LLM call and nothing else, leaving finance and platform teams with token counts where they need a cost model.
Only the model call is priced
The model call is easy to price. But the tools an agent uses and the people it asks to approve things cost money too, and that part goes uncounted, so you only see a fraction of what a run really costs.
Cost lives in one place, price lives in another
When prices sit in a separate list, a cost figure can't be traced back to the rate that produced it, and a catalog refresh can quietly change the number underneath it.
Cost with no sense of value
A spend figure on its own reads as a bill. Without the value of a run delivered beside it, there's no way to tell an expensive agent that earns its keep from one that doesn't.
The Solution
Price at the source, meter at the gateway, read per agent.
There's no separate price list to maintain. Each catalog item carries its own cost attributes: a model carries an input and output price per million tokens, an MCP tool carries a reference cost per call, and a human decision carries the rate set on its rule. The AI Gateway prices traffic against those attributes as it flows, and the figures roll up per proxy on the dashboards and per agent on the agent's Cost page, set against the value the agent's tools declared.
The Outcome
Cost you can trace, read against value.
Because price is an attribute of the catalog item and cost is metered at the gateway, every figure can be pointed back at the rate that produced it, and read next to what the run delivered.
The whole run, priced
Model tokens, MCP tool calls and human decisions are all priced at the AI Gateway, so an agent's cost reflects the whole run rather than the model call alone.
Read per agent
An agent's cost page adds up its model spend, tool spend and human decisions for a period, says what changed since the period before, and shows where the money went.
Cost set against value
Each MCP tool's owner declares what a successful call is worth, and the platform credits that value to the agent's successful calls, next to what they cost.
How It Works
From attributed spend to chargeback.
The story runs in five moves: attribute what agents spend, weigh it against the value they return, treat human approvals as a line item, set budgets that can act, and hand each team its share of the bill.
Step 1
Agent FinOps: price the catalog
Attach cost attributes to the catalog items agents consume. There's no separate price list; each item carries its own.
A model carries an Input price and an Output price per 1M tokens, with currency and an effective date
An MCP tool carries a Reference cost in USD per call; a tool with none reads Not priced, which isn't a price of zero
A human decision carries a Cost per decision (USD), set in the HITL rule defaults or on a single rule
Step 2
Metered at the AI Gateway
Every request is priced as it passes through the gateway. The model's input and output prices are applied to the token counts of the request; MCP tool calls are priced against the price book the proxy carries.
Input cost and output cost recorded as metrics on each model request
Tool cost reported on the MCP: Overview dashboard, with a Tool Price Coverage widget showing which billed calls carried a rate
The Token Rate Limit policy can cap token usage at the gateway
Step 3
Read what an agent cost
An agent's Cost page adds up its three kinds of priced events over a period, on the traffic attributed to the agent's gateway application.
Key figures: Spend, Runs, and Avg cost per run, each with its trend against the period before
What changed, Where the money goes (Models, Tools, Human decision), and Spend over the period
The Agent: Overview dashboard charts the same figures over time: Total Model Spend, Total Tool Spend, Total Human Decision Cost, Cost by Model, Top 5 Tools by Cost
Step 4
Attribute business value
Counting tokens shows what a run cost; value attribution shows what the spend bought. A tool's owner declares what one successful call is worth, and the platform credits it to the agent's successful calls.
A Reference value in USD per successful call, set inline on the tool's page; a tool with none reads No declared value
Value delivered appears on the Cost page's Spend and value card, marked Declared value
Total Tool Value and Value Over Time chart it on the Agent Overview dashboard
Our Impact
The Stories Behind the Success
“AI agents are becoming an important interface for enterprise software, and enabling them securely is a key part of our strategy. With Gravitee Agent Management, we're able to extend our existing API program to support AI agents, transforming trusted APIs into governed MCP tools while preserving the security, observability, and operational controls our customers expect from Tealium. This allows us to innovate faster and adopt new AI capabilities without introducing unnecessary complexity."
Matthew Houser
SVP of Product and Engineering at Tealium
“By implementing Gravitee’s AI Agent Management, we can centralize and govern access to language models and eliminate shadow IT. It also helps us consolidate our existing chatbots, which today operate in silos, and prepare for future agent-to-agent communication across use cases like HR, maintenance, and field operations.
For us, having one governed entry point for AI means better governance, cost tracking, token visibility, and control over sensitive data.”
Matthieu Klotz
Data Architect & Head of Exchange Platforms @ RATP Group
Do you know what your agents cost?
Price models, tools and human decisions in the catalog. Meter every run at the AI Gateway. Read what each agent cost, set against the value it delivered.
