methodosView source ↗

● Open source · Python · Apache-2.0

Give your agent
a way forward.

Methodos is a typed procedural memory for LLM agents. It turns execution feedback into a graph of better next steps — inspectable, local, and built to stay out of your way.

Read the docs ↗See how it works ↓
procedural memorywhat to do next
QUERYABLELOCALIZEDVALIDATEDEVOLVING

The loop

Useful in the hot path.
Improving off the hot path.

Runtime guidance stays separate from offline evolution, so experimentation never touches the live graph.

01

Locate

Match the last action to the part of the graph that matters now.

locate(graph)
02

Guide

Turn the local neighbourhood into concise, cached guidance.

guide(graph)
03

Refine

Score runs, propose edits, and keep only validated improvements.

refine(graph)

A small core

One primitive.
Clear boundaries.

Read the architecture in the product docs. The repository is for contributors, not the primary user journey.

Start here

Keep the agent.
Add the method.

Bring your own solver. Methodos speaks through a small async protocol, so the graph can sit beside the agent you already have.

Open the getting-started guide ↗
$ pip install methodos

from methodos import PGAdapter

adapter = PGAdapter(...)