# llms.txt for API docs

> llms.txt for api docs is a short Markdown map in front of a long reference. The format, and how RingKit's two files split the job.

Published 2026-11-15 · by RingKit · tags: ai receptionist, setup. Canonical: https://www.ringkit.com/blog/llms-txt-for-api-docs

llms.txt for api docs is a Markdown file an agent can read before it calls an API. It is not the API. It is a map: a name, a short summary, and pointers to the pages that actually specify each call. RingKit publishes two of them for the phone API. `https://api.ringkit.com/llms.txt` is the short map. `https://api.ringkit.com/llms-full.txt` is the long reference. Fetched on 15 November 2026, the short file was 4,454 bytes and the long file was 73,651 bytes. The point of the split is that an agent should not load every field on every turn.

The format those files are aiming at is the [llms.txt proposal](https://llmstxt.org/) by Jeremy Howard. The page, updated 10 August 2026, says the file is Markdown with a fixed order so a parser can read it, not only a model. This page uses RingKit's two API files as the worked example, including the places they do not match the proposal line for line.

## What the proposal requires

The proposal says the only required section is an H1 with the name of the project. After that, a blockquote holds a short summary. Then come zero or more blocks that are not headings, for notes. Then come H2 sections. In the proposal, each H2 is a list of links: a markdown link, then an optional colon and a note. A section titled Optional is the convention for links an agent can skip when it needs a shorter context. The file itself stays small. The detail sits behind the links and is fetched when a question needs it.

The same page is clear about what the file is not. It is not a sitemap. A sitemap lists pages for a search index and is often too large to put in a context window. llms.txt is a curated list for an agent that is about to do a task. It also is not `robots.txt`. That file says what an automated crawler may fetch. llms.txt is read on demand, when an agent needs the map.

## The short file is the order of operations

RingKit's API `llms.txt`, opened the same day, starts the way the proposal asks. The H1 is `RingKit`. The blockquote says what the product is: an AI receptionist, pay for what you use, set up over the API or MCP, and the human step is paying. Under that, before any H2, a bullet list gives the URLs an agent needs next: the MCP server, the REST base, the OpenAPI document, and the long file.

The one H2 is `Lifecycle (MCP tool names; REST in /openapi.json)`. It is a numbered list of twelve steps, from create through an optional hand-over. It is not a list of links. That is the gap with the proposal, which wants H2 sections to be link lists. The steps are still the useful part of the file. They tell an agent the order: create, teach, voice, routing, a billed text simulation, fund, number, a verified test call, then calls, webhooks, keys, and an invite. Funding sits before the number because ordering a number needs $2 of paid credit. The file says there is no free signup credit. There is no free trial.

`ownerPhone` is not in the create step. The file's create line says to keep the API key from `ringkit_create_business`. The key is shown once. The full field list for each of those steps is not in this file. That is deliberate. The short file points at `llms-full.txt` for it.

## The long file is the reference

`llms-full.txt` repeats the idea in one document an agent can load. Its H1 is `RingKit: full reference for AI agents`. The H2s are Base URLs, Authentication, Conventions, Pricing, The lifecycle, Hand-over, Webhooks, a route index, and MCP. Under the lifecycle, each step has the REST method and path, and notes on when to stop and ask a person. The [docs](https://www.ringkit.com/docs#agents) tell a plain tool-calling loop to put this file in the system prompt and to give the model one HTTP tool, next to the OpenAPI document at `https://api.ringkit.com/openapi.json`.

Those are three different artifacts. OpenAPI is the schema: paths, parameters, response shapes. The long text file is the order and the judgment: which call comes first, which error hint to follow, which step waits for a person. The short file is the index you fetch when you do not want 73 KB in the prompt yet. The proposal's rule fits that use. Fetch the map. Follow the link when the step needs a field.

The marketing site has its own `llms.txt`, and that one is closer to the proposal's link lists. It has an H1, a blockquote, and H2 sections (Docs, Product, Blog, Optional) whose items are links to Markdown pages. It describes the website. The API file describes the phone API. An agent setting up a line should open `https://api.ringkit.com/llms.txt`, not the homepage file. The lifecycle is what it should read before it POSTs. The site file is for a reader.

## How to use the pair

Start the agent with the short file, or with the sentence "read `https://api.ringkit.com/llms.txt` and follow `next`." Every MCP tool result includes a `next` field. The short file and that field are the same idea: do not dump the whole surface. When the model has to form a body, give it the matching section of `llms-full.txt` or the OpenAPI operation. Do not paste both files, plus the blog, plus the website, into one prompt and hope the order survives.

A few limits are in the short file and are easy to skip. Keys are `manage` or `read`. A read key cannot create. POSTs take an `Idempotency-Key`. The same key within 24 hours returns the stored response. A text simulation is billed from the wallet at the AI rate. The number is $1.99 a month and talk time is $0.20 a minute, first 10 seconds free, then 6-second increments, with a $10 minimum top-up. The [pricing page](https://www.ringkit.com/pricing) is that rate card for a person. The text file is the same numbers, written so an agent does not have to scrape HTML.

## What the file will not do

It will not place a call by itself. It will not hide the fact that a person pays and a person reads back a verification code. It is not an SDK, and the docs say there are no SDKs yet. The tool list an editor calls is the MCP server, described in [MCP phone receptionist](https://www.ringkit.com/blog/mcp-phone-receptionist). The HTTP calls under those tools are [AI receptionist API](https://www.ringkit.com/blog/ai-receptionist-api). llms.txt is the map in front of both.

If you publish your own file for a different API, the proposal's minimum is the H1. The useful version is the H1, a blockquote that states the trap (here, do not order the number first), and links that stay small enough to load. RingKit's short file spends its one H2 on the twelve steps instead of on a link list. Use it for the order. Use the long file, or OpenAPI, when you need the fields.

---

Try Ringkit: $0.20/min, $1.99/mo per number, no plan. Get started at https://app.ringkit.com/onboard
