After Jeremy Herzog’s informative presentation at SOTM US on HIFLD Next and its preparations for LLM usage, and some helpful follow-up discussion, I’ve decided to implement llms.txt and human- and LLM-readable body messages in the HTTP errors returned from my Overpass container build. The changes are in a working branch at the moment but will be merged into the main branch and the container image soon.
The llms.txt file contains a default message that explains that the endpoint is not intended for use by LLMs, AI agents, or high-volume scripts and that high-volume users should run their own local Overpass instance. The llms.txt file can be overridden by mounting a replacement file in the container at startup.
The HTTP 429 Too Many Requests response body recommends exponential backoff, and the HTTP 504 Gateway Timeout response body recommends adjusting the query timeout, simplifying the query, and backing off if those changes aren’t effective. These responses are currently hardcoded in the nginx configuration but could be moved to static files if there’s a reason to override them in the future.
I haven’t posted the exact wording here because the GitHub repo is a better place to discuss it if you have feedback.
But I wanted to share the idea here in case other Overpass instance builders might consider doing the same thing to help shape LLM and bot traffic.