Using these docs with AI

This documentation is also published as plain text written for language models. Point an AI assistant at it and its answers about Spherum come from the current pages rather than from whatever it remembers.

The files

Both are rebuilt with every release of the documentation:

File	What it holds
llms.txt	An index: what Spherum is, and a link to every part below
llms-full.txt	Every page, in full, as one file — about 20,000 tokens, which fits most assistants

Each guide — the Administrator guide, the Operator guide, the Guide for your people, Tokens and the API — is also published as a file of its own, linked from llms.txt, for when only one part matters.

In a chat assistant

For a single question. Paste the address of llms-full.txt into your message and ask the assistant to answer from it. This works only when the assistant can open links — if it can’t, attach the file instead.

For a project you come back to. Download llms-full.txt and add it to the project’s files. An uploaded file is a copy: when the documentation changes, it does not. Replace it when the changelog shows something new.

In a coding tool

Building an integration. The reference is generated from the API specification, and the specification is the most precise thing to give a tool that writes code:

https://api.spherum.ae/help/openapi.json

Give it that alongside llms.txt — the specification says what each call accepts and returns, and the guides say what the calls mean.

Keeping it in the project. Most coding tools read a rules or instructions file from the repository. One line there is enough for every session to know where to look:

Spherum documentation for LLMs: https://docs.spherum.ae/llms.txt

Editors that index documentation sites can also be given https://docs.spherum.ae as a source.

Working in your workspace, not only reading about it. These files give an assistant the documentation. To let it search, invite, re-send and archive in your own workspace, connect it over MCP: Connect an AI assistant.

Caution

Keep your API key and your people’s details out of the conversation. An example request needs the shape of the data, not real corporate email addresses. The key belongs in your integration’s secret store or in your AI assistant’s connection settings — never in a chat, a prompt, or a rules file committed to a repository.

When an answer disagrees with the pages

The pages are the source. An assistant can misread them, mix them with an older version of the API, or fill a gap with something that sounds plausible. When an answer matters — an irreversible step, or what the API returns — check it against the page it came from, and the changelog for what has changed recently.