Memory for IA Node in WhatsApp Workflows

Hi guys! I’m sharing my feedback on using the AI node in Workflows for WhatsApp, in case it’s useful for your roadmap.

Context: I have a WhatsApp workflow with the AI node connected to my own Anthropic API key. It works really well, but I’ve run into one limitation: the AI node doesn’t have conversation memory.

The userPromptTemplate only receives the latest message, so in a multi-turn conversation the AI forgets what came before. It loses the name, the context, and even says things like “we don’t have any history, this starts here.” That makes the conversation feel unnatural. I haven’t found any memory/history toggle in the node.

I solved it with a workaround: set_variable nodes that accumulate a transcript in a variable called historial, adding “Lead: ...” and “Soma: ...” before and after each turn, and then passing that to the AI through the userPromptTemplate.

It works, and the AI now remembers. But it has two drawbacks: (1) it requires extra nodes and is a bit fragile to maintain, and (2) I resend the entire history on every turn, so token costs increase as the conversation gets longer.

The idea: since the AI is already using my own API key, could you add a native conversation memory option, enabled/disabled with a toggle inside the AI node?

Ideas for how you could implement it — you already store the thread, and GET /conversations/{id}/messages exists:

• A toggle called “include conversation history” that automatically injects the last N messages from the thread as role-based messages — user/assistant — into the LLM, instead of sending a single userPrompt.

• A configurable window — last N messages or last X tokens — to control cost.

• Optional: an automatic rolling summary for longer conversations, so costs don’t spiral.

Thanks, and happy to give you more detail about the use case if it helps!