Attive: on memories prompts and agents

This is a short excerpt on how to think about memories, prompts, and agents.

At first, you start from a blank slate and set up a couple of memories to improve the results you get in chat. The more you "train" Attive by writing memories, the better the results you’ll see.

Eventually, you’ll notice yourself asking the same questions again and again. That’s where prompts come in handy: you store those questions in prompts you can easily reuse.

At this point, as you may have noticed, memories are retrieved based on a semantic match between your questions and your memory library. For starters, or for trying out different questions, that’s perfectly fine. But the goal is to get high-quality results repeatedly.

This is where being explicit about which memories you want to tie to your prompt becomes relevant. That ensures the AI sees your question and only the relevant context. With that in place (as a stored prompt, for example), you’re one click away from that weekly or daily question you ask.

Now, you might have a set of prompts and memories that are all related to a use case, but with variations. It could be, for instance, deal-lost analysis, pipeline health, account health, etc. There are often multiple questions you’d ask in your day-to-day.

This is where you bring in Agents. They will have a set of memories and a set of prompts focused on a specific use case. Think of them as "fresh employees" on your team: you give them enough context ("memories") and tell them which tasks to perform ("prompts"). You can also provide an instruction to better accommodate the style of that "fresh employee"—how long or short their answers should be, for example.

All in all, you should be able to use Attive in its vanilla state with a few memories and by just asking in chat. But the more you invest in packaging your use cases into focused, trained agents, the better the outputs you’ll get. Having agents in place is also a great way to bring colleagues on board, since they can tap into the value you’ve already put in place for them.