Currently, using the Search API requires separate pay-per-use credits at 2.5ยข/search, even if you're on an unlimited plan. This makes it prohibitively expensive for personal use cases like integrating Kagi into your own tooling.
The feature would allow subscribers on Professional or Ultimate plans to generate an API token that draws from their existing subscription allowance rather than requiring separate billing. It would be rate-limited to prevent abuse - something that approximates normal human search behaviour rather than automated bulk usage.
This wouldn't affect the existing pay-per-use API at all. It'd just give personal users a way to access Kagi programmatically without being double-billed for searches they're already paying for.
The main use case is personal tooling and integrations - things like MCP servers for LLMs, Alfred/Raycast extensions, or custom search integrations. A user on the Professional plan wants their local LLM to be able to search the web using Kagi, so they set up an MCP server that calls the Search API. Right now that costs extra on top of their subscription. With this feature, those API calls would just consume from their existing unlimited allowance.
A natural two-tier model emerges from this:
- Subscription-backed token - for personal use, rate-limited to prevent bulk/automated abuse, consumes from existing plan allowance
- Pay-per-use token - current pricing, higher rate limits, for businesses and production workloads
The rate limits naturally segment the two audiences - anyone building something commercial will need the higher limits of the pay-per-use tier regardless. The subscription tier is just for people who want Kagi in their own personal setup.
It's also worth noting that API responses are cheaper to serve than full page renders - no HTML, CSS, JS, or images to deliver. So for personal usage volumes, this shouldn't add meaningful cost.
