memory

To maintain context continuity across sessions, you can use a layered memory model that stores daily session logs in memory/YYYY-MM-DD.md files and centralizes long-term learnings in MEMORY.md, preventing cold starts when conversations restart.

Preventing cold starts is achieved by automatically logging session-start context, heartbeat updates, and session-end summaries into daily memory files, allowing seamless resumption of previous work without losing context.

Persistent memory storage works by writing daily entries to dated markdown files, aggregating durable learnings into a central MEMORY.md file, and optionally ingesting this content into a searchable knowledge base using RAG and ChromaDB.

Yes, you can make long-term memory searchable across sessions through optional RAG and ChromaDB ingestion, which allows you to retrieve stored context and knowledge base entries for later reuse across different missions.

You do not need a knowledge base to store session context and heartbeat updates, as the system writes daily entries directly to markdown files and a central MEMORY.md file, while the RAG knowledge base is an optional feature for searchable retrieval.