LLM Reference

easymysql.com/llms.txt is the whole 0.2 API in one Markdown file: the rules, an API map, every operation with its signature, the error table, and the differences between the backends. It is written for a coding assistant's context window, not for reading top to bottom. 

curl -O https://www.easymysql.com/llms.txt

easymysql.com/llm.md redirects to the same file. It also ships in the repository as llm.md.

What is in it

Section Contents
1. Rules Five things to always do, and thirteen patterns not to generate, each with the wrong and right form
2. API map Every public method, what it returns, and the section that covers it
3–4. Import, connect, lifecycle Both backends, driver options, threads, and the reconnection contract
5–7. Reading, writing, raw SQL Signatures and behaviour for select, upsert, bulk writes, expressions and batches
8. Transactions and session state Savepoints, thread ownership, the pending-work guard, resetSession()
9–11. Conditions, clauses, builders Every constructor, the keyword clauses, and the (sql, params) functions
12–14. Types, errors, backends Type mapping, the full exception table, and the MySQL/PostgreSQL differences
15. Worked example Five functions covering the common operations

Using it

Paste it into the assistant's context, or drop it in the project so the assistant reads it with the rest of the repository.

Tool Where it goes
Claude Code llm.md at the project root, referenced from CLAUDE.md
Cursor .cursor/rules/easymysql.md
GitHub Copilot .github/copilot-instructions.md
Chat interfaces Attach the file, or paste it before the question

A one-line pointer is enough when the file is already in the project: 

Use easymysql 0.2. The API is described in llm.md; follow section 1 exactly.

Scope

The file describes 0.2.0.0. Every example in it is executed against MySQL 8.0.46 and PostgreSQL 16.14 as part of the release check, on Python 3.9 to 3.14.

The v1 documentation is not a source for this version

/docs/v1 describes 0.1.9 and its examples do not run on 0.2. An assistant that was trained on it, or that finds it first, will generate calls that raise — update() and delete() without a condition, resetCache(), and conditions built by string concatenation. The file states this in its header.

What it changes in generated code

The behaviours that a model trained on older wrappers gets wrong, and that section 1 of the file states first: