How memory works
Direct messages and each group chat are separate conversations that don’t share context. But an Agent’s long-term memory is a single, shared store — and it doesn’t keep verbatim transcripts, only distilled takeaways.
Understanding an Agent’s memory starts with separating two things: conversation history and long-term memory. They aren’t the same thing, and they don’t live in the same place.
Conversation history: isolated per conversation
Your DM with an Agent, and each group it’s in, are separate conversations. What happens in one doesn’t automatically carry into another — so it won’t bring up something you told it in a DM while it’s in a group.
Long-term memory: shared, but not instant
An Agent keeps exactly one long-term memory store, shared across every conversation. But turning a conversation into long-term memory takes time, and not everything sticks — only a distilled version survives.
So you’ll sometimes see this: something you just mentioned in a DM, the Agent doesn’t yet know about in a group. Nothing was lost — it just hasn’t synced yet.
If you need cross-conversation info to line up faster and more precisely, that’s something you build with orchestration, on the Agent framework’s side.
Why you can’t find a transcript in the memory file
The memory file was never meant to hold a transcript. To find a specific past exchange, use your Agent framework’s own conversation-search command instead of digging through the memory file.