MathCode: An AI Coding Assistant with a Mathematical Formalization Engine

MathCode is a terminal-based AI coding assistant with a built-in mathematical formalization engine. When users input math problems in natural language, it automatically converts them into Lean 4 theorems and attempts formal proofs through a persistent Lean REPL and agent-based proving methods.

Key features include:

• Persistent Lean REPL: After a single warm-up, it significantly reduces compile check times from about 30 seconds to 0.4 seconds.
• Theorem and Axiom Library: Automatically names and stores proven theorems, and manages interactive assumptions as persistent and consistency-checked Lean declarations.
• Lean LSP Integration: Searches Mathlib lemmas via leansearch.net and Loogle, and fixes errors with structured LSP diagnostics.
• Obsidian Theorem Graph: Generates an Obsidian knowledge graph visualizing dependencies between theorems and lemmas.
• Advanced Proving Methods: Supports Agent-Mode Proving, where agents directly write candidates and read errors; Tree-of-Subgoals, which decomposes complex theorems into subgoals for parallel proving; and Multi-Planner, which executes various strategies.

This tool is based on the AUTOLEAN project and is available via the codex CLI in macOS(arm64) and Linux(x86_64) environments.