v2 · 18 concepts · guides · by @korpus

Project Memory for AI Coding Agents

Give your AI coding agent a project memory: decisions, conventions and handoffs it reads first and updates last, so tomorrow starts where today ended.

ai-coding-agent-memory
persistent-memory
claude-code-memory
decision-log
session-handoff
project-context
mcp
ai-memory
agents-md

Use template Free. Sign in or sign up on the way.

How it works

What you get. A ready structure your agent can read in one pass:

context/ the project brief, architecture and glossary
decisions/ numbered decision records, including one that supersedes an earlier one
conventions/ code style, testing, git and review rules, definition of done
runbooks/ release, database migration and debugging procedures
handoffs/ where the last session stopped, journal/ incidents and lessons

The example project, Tideline, is made up. Replace it with yours and keep the structure.

Start in three steps.

Use this template, then edit the project brief and architecture (about five minutes).
Connect your AI over MCP (Claude, ChatGPT, Cursor or another app that speaks MCP) and ask it to read AGENTS.
Try the first prompt. Work as usual and end each session with a handoff.

How it grows. Your agent writes decisions, handoffs and lessons back into the bundle. Every change carries a one-line reason, and the log shows who changed what and why, so the next session can see how the project got here.

Does an AI coding agent remember my project between sessions? Not by itself. A chat forgets. A bundle keeps the decisions and conventions in one place that every session reads first, and that you can read and correct too.

Use template, connect your assistant, and ask the first prompt: it takes two minutes.

What you get

A bundle of 18 concepts, copied into your account. Changes are yours alone; the publisher never sees them.

Every concept in it18 paths · show

AGENTS
README

context/

architecture
glossary
project-brief

conventions/

code-style
definition-of-done
git-and-reviews
testing

decisions/

0001-postgres-over-document-store
0002-streaks-derived-from-events
0003-no-orm-query-builder-only
0004-grace-day-superseded

handoffs/

2026-09-28-grace-day-api

journal/

2026-08-14-streak-drift

runbooks/

debug-wrong-streak
release-checklist
run-a-migration

Example prompts

What to ask your agent once the template is yours.7 prompts · show

Read AGENTS and context/project-brief, then tell me in five lines what this project is and which decisions you must not undo.
Record a decision: we chose Postgres over a document store because reports need joins. Write it as the next numbered decision and link it to the architecture page.
Before you change the streak logic, read the decisions and conventions that apply and list them. Then implement it and add a line to the log of why.
We are stopping for today. Write a handoff: what is done, what is half-finished, the next step and the files involved, and remember it for the next session.
A streak came out wrong in production. Follow the debugging runbook, then add a dated entry to the journal with the cause and the fix.
Review this pull request against our code style, testing rules and definition of done, and point to the page each remark comes from.
Replace the Tideline example with my project: ask me five questions, then rewrite the project brief, architecture and glossary from my answers.

Screenshots

How it looks in use.2 screenshots · show

Project Memory for AI Coding Agents bundle tree in Korpus: context, decisions, conventions, runbooks, handoffs and journal folders plus AGENTS and README

A decision record for AI agent memory: context, choice and consequences, with links to the related handoff and the decision it supersedes

Make it yours

One click after sign-in, and your agents can start working in it.

Use template