Memory

Memory is, in the official README's words, "a basic implementation of persistent memory using a local knowledge graph. This lets Claude remember information about the user across chats." It solves one of the oldest pains of working with AI: every new conversation starting from scratch.

The official model has three pieces: entities (graph nodes with a name, a type, and observations — e.g. "Juan_Garavito" of type "person" with observation "prefers Tailwind"); relations (directed connections between entities — e.g. Juan_Garavito → works_at → NexFlowAI); and observations (atomic facts as strings, one per data point).

What you can ask

• "Remember that my preferred tone for client emails is formal but warm, no emojis."
• "What did we decide last time we discussed the checkout redesign?"
• "Note that the 'La Mariposa' restaurant has issues with the vegan menu — keep it in mind next time we write to them."

For the agent to get the most out of this, the official README recommends a short system prompt instructing it to "identify the user and retrieve memory at the start of the conversation; record new facts at the end." That way the notebook accumulates without you asking every time.

What you need to configure

Nothing. No token, no external account, no folders. Memory lives on your own computer as a JSON file and the agent reads/writes it via the MCP tools.

The file path is configurable via the MEMORY_FILE_PATH env. If you don't set it, the file is born in the process cwd — better to pin it to an absolute path so it persists across sessions.