LLMs

If you are working in a plain chat window (ChatGPT, Claude, or similar), the model does not really know the Lemma language. It will guess, and the guesses are often close but wrong. To fix that, paste this link into the chat and tell it to read the reference first: lemma.run/llms.txt (opens in new tab) .

It is a single page that describes the whole language (specs, data fields, rules, vetoes, imports) in a format models read well. Once the assistant has it, the Lemma it writes actually matches the real syntax.

If your assistant supports LemmaBase MCP, use that instead. It connects straight to LemmaBase, so the assistant can search, read, and run your published specs, not just recall the language. Reach for llms.txt when MCP is not an option and you just need the model to get the syntax right.

New to Lemma itself? Start with the guide on Lemma.run (opens in new tab) .