portable-agent-memory

Give any AI coding agent a durable memory it can read, maintain, and improve.

Portable Agent Memory, or PAM, is a markdown-first protocol for turning an ordinary repository into an agent-readable memory workspace. It helps agents preserve decisions, facts, context, source notes, summaries, and follow-ups without depending on one vendor, runtime, database, or chat history.

The goal is simple: when a new agent opens the repo tomorrow, it should know where to look, what changed, what is trusted, what is uncertain, and how to keep the memory healthy.

Why This Exists

Most AI work disappears into chat history. The next session starts from a vague summary, stale assumptions, or a full reread of the repository.

PAM gives agents a durable place to put what matters:

• decisions and rationale;
• project facts and working assumptions;
• user or team preferences;
• troubleshooting discoveries;
• source summaries;
• open questions;
• chronological work history;
• generated wiki pages and indexes;
• maintenance reports.

It is not a SaaS product, a vector database, or a new agent framework. It is a portable contract plus a starter implementation that works with plain files.

What Is In This Repository

The important pieces are:

• memory/pam.version.json: explicit PAM memory format

and schema metadata for migration detection.

• memory/agent-memory/pam-runtime.md: the

compact graph-first runtime guide for everyday agent memory lookup.