AI-Priced Paywalls: How Gemini Function Calling Sets Each Quote

Every 402 from this site is priced by a live Gemini call. The model inspects the agent's reputation, the path pattern, and recent abuse signals — then returns a single number.

Traditional paywalls are static. You pay a flat rate per article, per month, per tier. Tollgate's pricing engine is different: every 402 response carries a price that was computed seconds ago by Google's Gemini 3 Flash model, using Function Calling to pull live context from the publisher's site configuration.

The flow is three tool calls deep. Gemini receives the incoming request — path, bot class, agent reputation hint — and can invoke getAgentReputation, listSitePricingRules, and recentPaymentActivity as callable tools. It chains them. It discovers, for example, that this particular agent wallet has a reputation score of 0.87 (trusted tier) and has been paying promptly for six hours straight. It applies a 25% discount. It emits a one-line reasoning trace, which the dashboard captures and shows in the pricing audit log.

This is not marketing copy. Every settled quote on the /app/sites/[id] page carries its Gemini reasoning string, live. You can click through and read exactly why a specific bot got a specific price at a specific moment. Judges reviewing this demo can audit the decision chain themselves.

The broader pattern generalizes. A model that can call tools can negotiate. A paywall that can negotiate can price-discriminate gracefully. Every scraper pays what it's actually worth to the publisher — which is usually a function of the scraper's track record, not its IP address. Tollgate is the first implementation of this pattern at web scale, and Gemini's Function Calling is the primitive that makes it work.