Guide

Build with AI agents

llms.txt, OpenAPI, per-page markdown — docs your copilot can eat.

Most integrations are written with a copilot in the loop now, so the docs ship in formats a model can eat whole. Point your agent at any of these — they render from the same spec as the pages you're reading, so the two can never disagree:

llms.txt

The index: models, endpoints, prices, guides — one small file, llmstxt.org convention.

llms-full.txt

The whole site as one markdown document, blog articles included. Zero crawling.

openapi.json

OpenAPI 3.1 — feed it to codegen, Postman, or an agent's tool-use schema.

Any page + .md

Every docs page has a raw markdown twin at the same URL plus ".md". Blog posts too.

Paste-ready context

Drop this into your agent's system prompt (or a CLAUDE.md / AGENTS.md) and it knows where everything lives:

agent context

You are integrating the eroq API (https://eroq.ai).
Read these before writing code:
- https://eroq.ai/llms.txt          (index: models, endpoints, pricing)
- https://eroq.ai/openapi.json      (OpenAPI 3.1 schema)
- https://eroq.ai/docs/chat.md      (any docs page + ".md" = raw markdown)
Base URL https://eroq.ai/v1 · auth header "Authorization: Bearer $EROQ_API_KEY".
Credits are flat per call — check "usage.credits_remaining" in responses
and treat HTTP 402 as "wallet empty", not as an error to retry.

On every page

Copy page — the page you're reading, as markdown, on your clipboard.
View as markdown — the raw .md twin, linked with rel="alternate" for crawlers.
Ask AI — opens Claude or ChatGPT with a prompt already pointing at this page.

Why agents integrate this API fast

Request and response shapes follow the OpenAI convention wherever one exists, so a model's prior knowledge transfers: it already knows what messages, choices and usage mean. The delta worth teaching is credits — flat per call, refunded on failure, and HTTP 402 when the wallet is empty.