Install¶
mathema needs Python 3.10 or newer and has two required dependencies, sympy for the symbolic route and pyyaml for the record store. Install it into a virtual environment rather than a system Python:
python3 -m venv .venv && source .venv/bin/activate
pip install mathema
or, in a project managed with uv:
uv add mathema
That is the whole of what most people need: mathema check, verify,
audit, lock, the derive route and the record store all come with the core
install.
Optional extras¶
Each extra adds one capability without making it everyone's dependency.
| Extra | Adds |
|---|---|
numpy |
array-shaped claims, matrix structure checks and array parameter synthesis |
smt |
z3 as an additional decision procedure for the derive route (a native library of roughly 100 MB) |
mcp |
mathema mcp serve, which exposes mathema's checking tools to a coding agent |
coverage |
reading a native .coverage report, so the tests you already run count toward the implementation score (a coverage.json export works without it) |
symbology |
conventional notation for parameter and function names when claims are rendered |
all |
numpy, smt, mcp and coverage together |
pip install "mathema[all]"
Offline by design¶
mathema makes no network calls. There is no account, no API key and no
telemetry, so it runs against a private codebase without source or claims
leaving the machine. The one command that fetches anything,
mathema init --agents, does so only when you run it explicitly.
Next¶
The quick start takes five minutes and one function from a falsified claim to a proven one.