MathCode, Mathematical Coding Agent

Aug 17, 2026 01:17 AM - 2 hours ago 1

Overview

MathCode is simply a terminal AI coding adjunct pinch a built-in mathematics formalization engine. Give it a mathematics problem successful plain connection and it will automatically person it into a Lean 4 theorem and effort a general impervious — pinch a persistent Lean REPL, reusable theorem and axiom libraries, agentic proving, and an Obsidian knowledge graph.

Quick Start

Requires macOS (arm64) aliases Linux (x86_64), positive the codex CLI for the default backend.

git clone https://github.com/math-ai-org/mathcode.git cd mathcode bash setup.sh codex auth login mathcode

setup.sh prepares the merchandise checkout, downloads the bundled runtime and Lean toolchain, and installs a user-local mathcode launcher. Try it with:

mathcode -p "prove that the quadrate of an moreover number is even"

Outputs are written to LeanFormalizations/. A browser UI is disposable via ./run webui.

Features

Persistent Lean REPL

A persistent Lean connection server brings compile checks to ~0.4s aft a one-time warmup, alternatively of ~30s.

Theorem Library

Every proved theorem is auto-named, stored, and made importable truthful the prover and planner tin reuse it.

Axiom Library

Store conversational assumptions arsenic persistent, compile-checked, consistency-reviewed Lean declarations.

Lean LSP Integration

Searches leansearch.net and Loogle for verified Mathlib lemmas and uses system LSP diagnostics for repairs.

Obsidian Theorem Graph

Generates an Obsidian vault that visualizes theorem-to-lemma limitations arsenic a knowledge graph.

Agent-Mode Proving

Each impervious becomes an interactive convention wherever the supplier writes candidates, sounds errors, and recompiles.

Tree-of-Subgoals

Decomposes analyzable theorems into independent subgoals and proves them successful parallel, past stitches them back.

Multi-Planner

Runs aggregate planners successful parallel for divers impervious strategies; the prover picks the champion approach.

Citation

If you usage MathCode successful research, please cite:

@misc{mathcode2026, title = {MathCode: A Frontier Mathematical Coding Agent}, writer = {Team Math-AI}, diary = {math-ai-org.github.io}, twelvemonth = {2026}, period = {April}, url = {https://github.com/math-ai-org/mathcode} }

The mathematics formalization and proving pipeline is based connected the AUTOLEAN project.