Memory in the Autonomy Agents framework refers to the agent’s ability to persist conversational/contextual state across multiple turns. Rather than treating each prompt in isolation, memory lets the agent “remember” what’s been said or done previously: user preferences, earlier questions, the flow of dialogue, and anything else relevant to maintaining a coherent, multi-turn interaction. This makes conversation more natural, consistent, and contextually aware.
How the Example Works
In the demo below, the agent is able to respond to a follow-up prompt (“What have I asked you before?”) by recalling earlier user messages in the same conversation (“What is your name?”, “Where do you live?”). The example shows that you can manage multiple concurrent conversations by passing a conversation identifier: each distinct conversation ID has its own memory thread. Further, you can introduce a scope (e.g. per-user or per-customer) so that memory is isolated not just by conversation but also across user boundaries. This is useful in multi-user or shared-agent environments where you want to ensure one user’s context doesn’t leak into another’s.Isolating conversations
You can have multiple separate conversations with an agent at the same time by using a differentconversation identifier for each one:
Isolating conversations per user
You can add an extra layer of isolation by using a uniquescope identifier (like a customer ID) when conversation-level isolation alone doesn’t meet your needs in multi-user environments: