@yigit Thanks — this is exactly the right question to poke at. Integration is a code change, not admin access. You add our SDK and wrap your existing LLM client in one line.

That's the whole integration. A few things that matter for security:

We never take your Anthropic/OpenAI/Bedrock keys. You give us a TokenOps key (ours), never your vendor credentials.
We don't proxy your traffic. Your calls go straight to the vendor in your own runtime — we're not a man-in-the-middle. We read the usage metadata the vendor already returns (model, token counts, latency) after the response lands, attribute it to the customerId you pass, and send that to your tenant.
Prompt/response content is opt-in and stays in your tenant by default. Most teams leave it off — we only need the usage numbers to do the math.

The only place a provider's billing API comes in is monthly invoice reconciliation, and it's optional + read-only: you forward the PDF/CSV, or we pull it via the vendor's billing API where one exists. No write access, ever.

@yigit Thanks — this is exactly the right question to poke at. Integration is a code change, not admin access. You add our SDK and wrap your existing LLM client in one line.

That's the whole integration. A few things that matter for security:

We never take your Anthropic/OpenAI/Bedrock keys. You give us a TokenOps key (ours), never your vendor credentials.
We don't proxy your traffic. Your calls go straight to the vendor in your own runtime — we're not a man-in-the-middle. We read the usage metadata the vendor already returns (model, token counts, latency) after the response lands, attribute it to the customerId you pass, and send that to your tenant.
Prompt/response content is opt-in and stays in your tenant by default. Most teams leave it off — we only need the usage numbers to do the math.

TokenOps by Lovie

AI unit economics platform for AI companies

AI unit economics platform for AI companies