Your agent is welcome here. Here is the contract.
01Why a travel site built for agents
An AI agent is the most honest customer a travel site can have. It cannot be rushed by a countdown timer, it does not see dark patterns, and it compares totals, not teasers. A site that wins agent bookings has to publish real all-in prices with deterministic rules. That is what Levelfare does for humans anyway, so we are building the machine side in from the start rather than bolting it on.
The industry lesson of 2026 points the same way. OpenAI walked back in-chat travel checkout in March 2026 because travel is too complex for a chat transaction, and the operators who are making agent booking work are doing it with deterministic booking APIs underneath the model. Language models guess; booking systems must not.
02The contract we are building to
| Commitment | What it means for your agent |
|---|---|
| One fare, all-in | The amount in the offer payload is the amount charged. Taxes, fees, and our disclosed 3 to 5% markup are inside it. No surprises appended at order time. |
| Deterministic booking | Idempotent order creation with client idempotency keys, structured machine-readable errors, and no silent retries. The same request produces the same result. |
| Policies in the payload | Cancellation terms, refundability, and fee rules ship inside every offer as structured data, not as a link to prose. |
| Offers expire honestly | Every offer carries its expiry. A stale offer re-quotes rather than booking at a moved price. |
| Human approval by default | The agent assembles, a person approves payment. Autonomous agent payment comes when the rails and liability rules are settled, not before. |
| Webhooks for changes | Cancellations, schedule changes, and refunds arrive as events your agent can subscribe to, so post-booking servicing is automatable too. |
03What exists today
The guides: dated, source-linked references on travel APIs and MCP servers for agents, what assistants can actually book, agentic payment rails, and the fare-watch pattern. A reference fare-watch implementation (TypeScript, mock mode plus Duffel test mode, human approval enforced in code) ships alongside the guides. Machine-readable site summary at /llms.txt.
When the booking API and MCP server open, this page becomes their documentation. Until then, hold us to the table above.
04Short answers
Can my AI agent book a hotel on Levelfare today?
Not yet. The booking API and MCP server open when our supply platform integration goes live. What exists today: the guides, a reference fare-watch implementation, and /llms.txt. This page states the contract we are building to.
Will agents get a different price than humans?
No. There is exactly one all-in fare per room per stay, with our 3 to 5% markup already inside it. The price an agent reads from the API is the price a human sees on the page.
Will agents be able to pay autonomously?
The default is human approval: the agent assembles the booking and a person confirms payment. Agent-initiated payment rails (Visa Trusted Agent Protocol, Mastercard Agent Pay, Stripe Shared Payment Token) will be supported as they mature and as liability rules settle.