Agent Cost & Usage

Mavvrik gives teams a unified view of how AI agents consume models, tools, and other paid services, and attributes that spend to the business activity that generated it.

Use Agent Cost & Usage to understand what an agent costs, what generated that cost, and who or what the spend belongs to.

What you can do

Track model usage and cost for supported agent applications and platforms.
Break agent spend down by model, provider, session, user, customer, application, use case, region, and tags.
Inspect the execution activity that produced the cost.
Add paid non-LLM consumption such as API calls, OCR pages, document-processing units, storage, or other metered services.
Apply your contracted LLM rates when they differ from Mavvrik's default model pricing.

Common use cases

Understand cost per agent or workflow

See how much each agent, session, or execution consumes and which model or operation drives that spend.

Attribute agent spend to the business

Use business context to answer questions such as:

How much does the support agent cost per customer?
What is the AI spend per user or application?
Which use cases are driving the highest agent cost?

Measure the full cost of an agent workflow

Supported model calls are priced automatically. Add metered usage for paid non-LLM services that contribute to the same workflow cost.

Use your own LLM commercial rates

Configure customer-specific LLM pricing when negotiated rates differ from Mavvrik's default pricing.

How Mavvrik gets agent data

Choose the integration that matches how the agent runs today.

Your environment	Integration	What it does
Python agent application	Python SDK	Captures supported providers, agent frameworks, vector databases, and business context from Python applications.
JavaScript / TypeScript agent application	JavaScript / TypeScript SDK	Captures supported providers, frameworks, vector databases, and business context from JavaScript / TypeScript applications.
eve agent application	Connect eve	Uses eve's native OpenTelemetry instrumentation to send agent telemetry directly to Mavvrik. No Mavvrik SDK is required.
Existing Langfuse project	Connect Langfuse	Reads supported Langfuse observations and attributes model usage and cost to connected agents.
LiteLLM proxy	Connect LiteLLM	Captures model-gateway telemetry and can attribute gateway usage to individual agents.
Self-hosted n8n	Connect n8n	Captures workflow executions, supported AI operations, tools, and business context from n8n.

For SDK applications, use Supported Agentic Stacks to confirm language, provider, framework, vector database, web framework, and serverless support before installation.

How agent cost is built

Mavvrik combines four layers of information:

Automatic AI usage — supported model calls, token or equivalent usage, duration, status, and pricing inputs.
Business context — customer, user, session, application, use case, request, region, and tags.
Metered usage — paid non-LLM consumption such as calls, pages, seconds, or other units.
LLM pricing — Mavvrik default model pricing or customer-configured LLM rates.

These layers allow the same agent activity to be analyzed from both an execution and cost perspective.

Start with your first integration

If this is the first time you are connecting an agent, follow Get Started.

The first-user flow is:

Choose an integration → connect it → run one test request → verify Cost and Sessions → add business context or metered usage where required.

Explore the documentation

Get Started — first-time setup and verification.
Integrations — SDK, eve, Langfuse, LiteLLM, and n8n integration paths.
Cost & Attribution — business context, metered usage, and LLM pricing.
Reference — security, configuration, and troubleshooting.