n8n OpenAI cost by customer, not by workflow name.

n8n can tell you a workflow ran. It can name the workflow. Finance cannot invoice a workflow name. The OpenAI node spends against one key. The customer who triggered the run is a field in your product. CostSlice is how that field becomes a P&L row.

There is no official n8n plugin. Do not wait for one. Use the OpenAI node you already have. Change the address. Send our key next to yours. Tag the customer.

Set base_url to https://api.costslice.com/v1. Keep OPENAI_API_KEY in the node credentials. Add header x-cs-key from your CostSlice workspace. Add x-cs-tenant from the item — an opaque id, never an email. Add x-cs-feature for the workflow or the step, such as ticket-summary. Add x-cs-env if you run staging and prod.

The 10-minute version for any backend is the same headers. Quickstart has Python, Node, and Vercel AI SDK. n8n is the same contract in a node.

One workflow can serve every customer. That is the usual n8n shape. One OpenAI key behind it. OpenAI cost per tenant is the tag. Without the tag, n8n still works. CostSlice records untagged. The completion still returns. Fix the expression that fills the header.

Feature names matter in n8n because one canvas hides three SKUs. A classify step, a draft step, and a retry look like one workflow. They do not cost the same. OpenAI spend per feature is how you split them. Reprice the expensive step. Leave the cheap one.

We do not store the prompt the node sent. We do not store the completion that came back. n8n may keep its own execution log. That is n8n. CostSlice keeps counts and tags. Security review the meter separately from the workflow history.

We do not cap a customer because the workflow is losing money on them. The P&L is a report. You change a price or you stop sending that customer through the expensive step. Starter will stop new proxy calls when the CostSlice plan hits 10,000 requests. That is our plan cap, not an n8n error on one tenant.

Claude nodes are not this product. CostSlice only sits in front of OpenAI. If the canvas calls another model, that spend is not on this P&L. Other models come later.

Keep both secrets in n8n credentials, not in a pinned workflow that you export to git. x-cs-key is a CostSlice credential, not a license. Rotate it if a workflow leak includes it. Close account revokes CostSlice keys. We do not email the secret again.

If the node has no place for extra headers, put them on the request options the OpenAI node already exposes. The contract is HTTP headers, not an n8n credential type we invented. We do not ship a community node. We do not claim a listing in their marketplace.

Starter $0 / Pro $79 / Scale $199, overage $6/100k. No credit card on Starter. Pricing on the homepage wins if a number here is stale.