n8n agents have six memory backends. Simple memory keeps only recent messages and doesn't persist. MongoDB, Postgres, and Redis each store data differently. Motorhead is an external API requiring no database setup. Exata is managed Postgres that trades control for less overhead. Zep uses semantic memory to synthesize information across conversations.
If you don't know what the different memories for your NAN agents actually do, then I'll break it down really, really quick. The simple memory. This is just a built-in memory store inside of NAN that allows your AI agent to remember the last few messages in the conversation. The memory isn't persistent between runs, so I literally just use it for testing. MongoDB chat memory. If you don't know what MongoDB is, it's just a database platform. Only really use this one if you're already using MongoDB for other things, but if not, don't even worry about it. Motorhead is just an externally managed memory service. So if you just want a quick memory API where you don't have to worry about setting up the database or managing it. The Postgres memory just stores your memory inside of a Postgres table. This is great if you want structured data that you can query later. The Redis chat memory just stores your memory in Redis, which you can self-host so you get full control over it. And also it's really, really quick. Exata, I think that's how you pronounce it, is similar to the Postgres one, but it's all managed for you, which means you don't have to worry about managing it, but also you have less control. And then the Zep memory, this is more semantic memory so your AI agents can actually piece together the different conversations.
Dynamic Workflows are crazy. This is how to actually use them (beginner's guide)
