An assistant that remembers
A research assistant that picks up where it left off. It remembers the accounts it has already looked into, how your team likes things done, and what it concluded last week, so you don’t have to explain it all again every time.
For example: last week it learned that a customer’s financial year starts in July. This week you ask it to “carry on with the Acme analysis”, and it already knows.
What you need
- An assistant (in almyty, an “agent”: an AI model with instructions and the tools it may use).
- Nothing else: almyty keeps the memory for you. You can connect an outside memory service instead if your company already uses one.
How to build it
- Open Agents, open your agent and click Edit. In the Memory section, turn on Remember between conversations.
- Under Whose memory, choose who the memory belongs to:
- Each person has their own: it remembers each person it talks to separately.
- This agent’s own: one memory, just for this assistant.
- Shared by all agents: your organization’s memory, which every assistant set to share can read and add to.
- Under What it saves, choose Facts it learns, Whole conversations or Only when asked, and under How long it keeps memories, how long. Use Never save to list things it must never write down, such as card numbers or passwords.
- Save. From now on it writes down what is worth keeping while it works, and looks up what is relevant the next time.
- To see or correct what it remembers, open the agent’s Memory tab, or Memory in the sidebar for everything your organization keeps. Add memory lets you write something down yourself, for example a company policy all assistants should know.
How memory is kept
Every memory has a length of life: Short-term notes from recent work (tidied into lasting facts every hour), This project, Long-term facts and preferences, and Shared by all agents. When a fact changes, almyty keeps the old version too, so you can see what an assistant knew at any past moment.