---
title: "The repo remembers so the next run doesn't relearn."
description: "Durable knowledge of how the codebase boots, where the gates are, and which patterns to follow. Versioned in the repo, loaded into every run, and corrected the moment a run proves it wrong."
url: "https://ara.so/product/repo-memory"
section: "Features"
---

# The repo remembers so the next run doesn't relearn.

Durable knowledge of how the codebase boots, where the gates are, and which patterns to follow. Versioned in the repo, loaded into every run, and corrected the moment a run proves it wrong.

## Three kinds of knowledge that compound.

### How it boots

Setup, build, and test commands captured once, so every run starts green.

### Where the gates are

The checks, owners, and protected paths a change must clear before it merges.

### Which patterns to follow

House conventions and prior decisions, so new code matches the code already there.

## Read it in, work the gates, write it back.

### Boot from memory

Every run loads the boot recipe, gates, and conventions before touching code. No cold start, no rediscovery.

_Environment_

### Work inside the gates

The agent follows the known patterns and respects the gates it already knows, so the diff lands close to mergeable.

_Implement_

### Write back what it learned

New decisions, fixed flakes, and changed conventions commit back to memory, so the next run inherits them.

_Repo memory_

## Stop re-explaining your codebase.

Point Ara at a repo. It learns how things work, then carries that forward into every run.