Notion as agent memory: the graph you'd build anyway

In Notion’s model, Notion becomes the hard drive and the agent acts as RAM — every decision, task, and outcome logged as a structured entry. The native MCP server handles the integration without much friction.

That framing resonates, even though I don’t use Notion for my own PKM. Mine is essentially a local folder of text files, and I’ve been running automation that generates notes from all my agent conversations, plus a minimal memory tool called OptMem: a single Python file, zero dependencies, an append-only log that builds a summary tree from all the notes. Memory lives in conversation context rather than a separate database — no embeddings, no graph, just structured summarisation.

What draws people to Notion over something simpler is the question I keep coming back to. Notion’s structural layer — linked databases, properties, relations — is basically the graph you’d build from scratch with most vector approaches anyway, without the extra infrastructure.

What are people using for the orchestration layer? Curious whether it’s n8n, custom Python, or something else.

Yours, in plain text,
Zhenia