agent-shared-memory

To share memory across multiple AI agents, use a persistent, append-only shared memory ledger like memory.yml. It records decisions, open questions, and agent history, providing continuity between independent Claude, Codex, and Gemini sessions that lack shared conversation history.

An append-only audit trail for multi-agent coordination is a persistent ledger that logs decisions, open questions, and artifacts sequentially. It ensures project continuity by recording agent history across multiple runs without modifying past entries.

To initialize shared memory for a multi-agent project, create a per-project scoping directory named .coord. The system generates a memory.yml file inside this directory to start logging decisions and artifacts.

Yes, you can safely append new decisions to memory.yml while multiple agents are running. The system uses atomic writes secured with a lock to prevent data corruption when logging entries across concurrent multi-agent sessions.

To generate a digest of the current project state from multi-agent memory, use the read mode. This function processes the append-only memory.yml ledger and summarizes existing decisions, open questions, artifacts, and agent history.

You need a persistent blackboard for multi-agent workflows when coordinating separate AI sessions on a single project. It bridges the gap when agents like Claude, Codex, and Gemini cannot natively access each other's conversation histories.