Integrations5 min read
ByCostGraph Team
CostGraph launches support for Cencori
Bring Cencori model usage into CostGraph to understand AI spend alongside the compute, databases, and other services behind your applications.
ChatGPTUnderstanding what an AI product costs means looking at the infrastructure behind it. Model calls, the compute running the application, its databases, and network traffic all contribute to the cost of serving a customer.
Cencori is an AI infrastructure company helping developers build, deploy, and run AI applications, agents, and models. Its goal is to make that infrastructure accessible to more people. Through its developer platform, teams can access models, build agents, and monitor their applications in production.
Working together made sense for us at CostGraph. We provide a context layer for compute, connecting usage and spend to the workloads, teams, and customers behind it. Model usage is part of that picture, alongside the cloud and Kubernetes infrastructure supporting an application.
Today, we're announcing support for Cencori. You can bring your Cencori usage into CostGraph and understand model spend alongside the rest of your infrastructure.
Turning model costs into application costs
Cencori records which models a project calls, how many tokens they use, and what it charges. Our integration brings those records into CostGraph so you can compare projects, identify the models driving spend, and include AI usage in your view of an application's costs.
The records include the model and its maker, such as Anthropic or OpenAI, and the type of work, such as text generation, embeddings, or audio. Request counts and separate input, cache read, cache write, and output token counts help explain how that spend accumulates.
Attribution is part of the integration too. You can group spend by Cencori project or configure a pull connection to attribute it to end users or embedded-agent tenants. This lets you understand which customers generate model costs and include those costs in your own billing calculations.
Cencori supplies the usage and billing detail. CostGraph places it alongside your other connected providers and makes it available for cost allocation and anomaly detection. The integration supports both importing records from Cencori and connecting directly from Cencori's console.
Cencori also publishes its per-token model rates to CostGraph's pricing catalogue. Those rates let CostGraph price AI gateway telemetry for Cencori using Cencori's own prices. Imported billing records already contain what Cencori charged.
A closer look at Cencori's own usage
Cencori uses the integration for its own organization. One of its projects is Arcie, its agent infrastructure. Between September 1 and November 4, 2026, Arcie made 4,577 model calls across 13 model makers. Cencori billed $61.07 for that usage.
Two models accounted for about 93% of the total.
- Claude Opus 5.5 accounted for $40.73, roughly two thirds of the bill. Its records included 5.5 million input tokens and 11.5 million cache-read tokens.
- Maximo Atlas 1.3 accounted for $16.10, with 79.6 million input tokens and no recorded cache reads.
That breakdown gives the team a more useful starting point than the total alone. Most of the recorded input and cache-read volume for Opus came from cache. Atlas had a much larger input volume with no recorded cache reuse.
For Atlas, this gives the team a reason to review the request mix and check whether any prompts could benefit from caching. Comparing the same breakdown after a change shows how cache usage and cost changed.
Connect your Cencori organization
You can start in either CostGraph or Cencori. Choose one connection method per Cencori organization, because importing the same organization through both methods counts its spend twice.
Connect from CostGraph
In Cencori, create a read-only usage export key under Organization settings > Advanced > Usage export. Copy your organization ID, then open Integrations in CostGraph and select Cencori. Name the connection, choose a CostGraph tenant, and enter the key and organization ID.
The connection form also lets you select whether to group usage by project, end user, or embedded-agent tenant.
Connect from Cencori
In a Cencori project, open Settings > Integrations and select CostGraph under FinOps. Select Connect, sign in to CostGraph, and choose the tenant that should receive the usage. The connection applies to every project in your Cencori organization.
This is how Arcie's records reach CostGraph. The connection page in Cencori shows the destination, delivery status, and controls for sending usage.
Cencori support is available on all CostGraph plans. Follow the Cencori integration guide to connect your organization, or Cencori's CostGraph guide to start from its console.
For us, this continues the work behind our cloud, database, and CI integrations. Understanding an application's costs means including the services it depends on. Cencori brings the model usage into that view.
Get started with CostGraph or book a demo to see it with your infrastructure.
Frequently asked questions
Cencori supplies daily usage and billing records in the FOCUS 1.2 format. They include the project, model, model maker, type of request, token counts, and what Cencori charged. CostGraph uses the same cost model for these records as for other providers.
For a pull connection, use a Cencori usage export key. An organization owner or admin can create one. It reads usage across the organization and cannot call models. Revoking the key stops subsequent imports.
To connect from Cencori, sign in to CostGraph and authorize the connection. You do not need to copy a Cencori key into CostGraph. A Cencori owner or admin manages the connection, and your CostGraph role must be owner, admin, or billing admin.
Usage arrives daily after the UTC day closes. Today's usage appears the following day. The Cencori connection page shows when usage was last sent.
When you use your own model provider key, that provider bills you directly. The Cencori record therefore reports zero billed cost while retaining the provider's charge separately. Connect the provider's billing integration to include that charge in CostGraph without counting it again as Cencori spend.
A pull connection supports attribution by project, end user, or embedded-agent tenant. End-user attribution uses the identifier your application sends with each request. Connections started from the Cencori console use project attribution.
Some records contain requests whose input and cache tokens cannot be separated. Cencori marks those records, and the separate token counts are incomplete. The billed cost remains available. All 4,577 requests in the Arcie example had token breakdowns.
Yes. Run the CostGraph exporter in a container, a Kubernetes CronJob, or on a machine you manage. It reads Cencori and sends the usage records to CostGraph, keeping the Cencori key in your environment. The exporter guide covers setup.