Skip to main content
Cencori is AI infrastructure that meters every model call your projects make and bills it at its own price. One connection captures that spend across every project, model, and upstream model maker in your Cencori organization. CostGraph reads Cencori as FOCUS 1.2 records at a grain of day x project x model x endpoint x sub-account. Each record is priced at what Cencori billed you.

​
What lands in CostGraph

  • Model spend per day, for each Cencori project and model.
  • Optionally, spend per end user or per tenant, so you can charge model cost back to your own customers.
  • Request counts and token counts, split into input, cache read, cache write, and output tokens.
  • The model maker behind each call, such as Anthropic or OpenAI, and the kind of work it did, such as text generation, embeddings, or image generation.

​
Choose how the data reaches us

CostGraph pulls it

You supply a read-only usage export key and we call Cencori once a day. Nothing to run.

Cencori pushes it

You connect CostGraph from the Cencori console and Cencori sends us the same records.
Use one or the other for a Cencori organization, not both. Connecting it both ways counts the same spend twice.

​
Before you begin

  • In Cencori, open Organization settings > Advanced > Usage export and create a key. Organization owners and admins can create one. Cencori shows the key once, so copy it before you close the dialog. It starts with cuk_.
  • Copy your Cencori organization id from Organization settings > General > System identifier. CostGraph uses it as the billing account.
A usage export key is read-only and covers every project in the organization. A project secret key also works, but it sees only its own project and can spend money if it leaks. Use a usage export key.

​
Let CostGraph pull

  1. Open Settings > Integrations and pick Cencori.
  2. Name the connection and pick the tenant these charges belong to.
  3. Paste the usage export key and the Cencori organization id.
  4. Optionally, choose what spend is attributed to:
    • project attributes spend to each Cencori project. This is the default.
    • end_user attributes spend to the end-user id your application sends on each request.
    • tenant attributes spend to the tenants of your embedded agents.
    • none leaves the sub-account empty.
  5. Select Connect.
CostGraph verifies the credentials with a short fetch before saving, so a wrong value or a missing permission fails while you are still on the form. CostGraph reads each day after it has closed in UTC, so today’s spend appears the following day.
Credentials are encrypted at rest and used only to read billing data. Revoke the key in Cencori at any time and the connection stops. Nothing else is affected.

​
Connect from Cencori

You can also start the connection from Cencori instead. In the Cencori console, open Settings > Integrations > CostGraph and follow the prompts. Cencori then pushes the same records to CostGraph, so you have no key to paste here. See Cencori’s CostGraph guide for the steps on their side.

​
Run the exporter

To keep the usage export key inside your own infrastructure, run our exporter. It reads Cencori on a schedule and pushes the records for you. Run it in a container, a Kubernetes CronJob, or a binary on a machine you already have.
Only the finished FOCUS records reach CostGraph. Run the exporter covers the three ways to run it and the variables it reads. To build the request yourself, see Push FOCUS data.

​
Calls made with your own provider key

When a call runs on your own provider key, the model maker bills you for it, not Cencori. Those calls land in CostGraph at zero cost, so your Cencori spend shows only what Cencori billed. The amount the model maker charged is kept separately on the record in x_UpstreamProviderCost. If you also connect that model maker, for example Anthropic or OpenAI, its bill carries the real charge and nothing is counted twice.

​
Cache reads

Each record carries x_CacheReadTokens beside x_InputTokens. Comparing the two shows how much of each prompt Cencori served from cache, which is one of the largest levers on token spend. A non-zero x_UnsplitRequests means Cencori could not split some requests into input and cache tokens, so the token columns undercount for that record.

​
What CostGraph does with it

Records land as raw billing rows, are normalised into line items, and roll up into daily cost. From there Cencori spend appears in Cost Overview and in anomaly detection alongside every other provider. Group by sub-account to see spend per project, end user, or tenant. Cencori usage export describes the export on Cencori’s side.