Zero inference on our servers
Typillar runs zero inference on its own servers. Every token used to plan or write your project is generated on your side, through your own Cloudflare account. This is a deliberate architectural choice, not a configuration option: there is no model running on Typillar’s infrastructure to fall back to, so there is nothing to fall back to.
What “zero inference” means
Section titled “What “zero inference” means”- Typillar’s control plane decides what work to do and orchestrates it.
- The thinking — every model call that reasons or writes code — runs on your own Cloudflare account.
- No prompts or generated code are processed by a model hosted by Typillar, because Typillar hosts none.
- With no Cloudflare connection there is simply no model. Typillar tells you so and stops; it does not quietly borrow one of ours.
See Your models for why inference runs on your side.
Both halves of the catalog, still on your account
Section titled “Both halves of the catalog, still on your account”Most of the catalog runs on Workers AI under your account, billed as ordinary Workers AI usage. Three frontier models — Claude, GPT and Gemini — are reached through Cloudflare’s AI Gateway, billed to credits you load in your own Cloudflare dashboard.
The distinction is a billing one, not a trust one. Both paths are your Cloudflare connection, your account, and your bill. Neither one routes a prompt through a Typillar inference server, because we don’t operate one.
Why it’s built this way
Section titled “Why it’s built this way”- A clean trust boundary. Your prompts and code don’t pass through our inference. The control plane coordinates; your infrastructure does the reasoning and the running.
- Cost transparency. You pay for inference directly on your own account — no proxy, no markup, no hidden token bill.
- Control. You choose which model your builds run on, and the Cloudflare connection that powers it is yours to revoke whenever you want.
How it fits the bigger picture
Section titled “How it fits the bigger picture”Zero inference is the inference half of Typillar’s ownership split: you own the account, the repo, the data, and every token of inference; Typillar owns the agents and the orchestration that put those tokens to work. The control plane is the office; the thinking and the running happen at your address.