Scheduled maintenance in progress — sign-in and sign-up are temporarily disabled. The rest of the site works normally.

Skip to content
All posts
Product

Your model, your bill: why we pass tokens through at cost

Most agent platforms mark up your tokens. We don't — the model isn't our business, the infrastructure is. Move the sliders and see your real bill.

J
Jhosetp Chino
Founder & Engineer · Jun 17, 2026 · 4 min read
Share

There’s a quiet tax in most AI agent platforms: they resell you the model. You send a message, it hits their account, and they bill you the tokens plus a margin. On a small pilot it’s invisible. At production volume it’s a line item that grows with your success.

We do the opposite. The model isn’t where we make money — the infrastructure is. So tokens pass through at cost, and you can bring your own key or your own model entirely.

Move the sliders and watch your real monthly bill — and the markup you aren’t paying:

Token cost, at costInteractive
Model
Conversations / month20,000
Input tokens / conversation1,200
Output tokens / conversation400
Your monthly cost on Thaliq
$192/mo
Thaliq markup$0
What a 20% platform markup would add+$38.40

You pay the provider directly, at cost. We never mark up tokens.

What “at cost” actually means

Two things, concretely:

  • BYO-key — plug in your Anthropic, Bedrock or Vertex credentials. The provider bills you directly. Thaliq never sits in the middle of that transaction.
  • BYO-LLM — go further and run the model in your own account (Claude on Vertex, for example). In self-host, your tokens never even leave your perimeter.

Either way, the number the calculator shows is the number you pay the provider. Our margin is zero on inference — by design.

Why give up the easy money?

Because token markup quietly makes us the wrong kind of company — one that wins when you spend more on the model. That’s misaligned. We’d rather win when the platform is worth paying for: the agentic loop, the tools and MCP, durable memory, HITL, observability, the channels, the self-host story.

Charge for the hard part. Pass the model through. It keeps our incentives pointed at building better infrastructure instead of inflating your token bill.

No lock-in falls out of this for free

If you bring the model, you can also take it with you. Swap Claude for another provider without rebuilding your agent. Export your specs and data. Self-host the whole thing. A platform that doesn’t profit from your tokens has no reason to trap them — and every reason to make the infrastructure good enough that you stay because you want to.

Ready to build your first agent?

The infrastructure is done. Bring your model and ship.

Get started