Memory and the Wiki

Each agent keeps private memory across conversations, and workspace knowledge shared by agents appears in the Wiki tab for every member to read.

Bolter has two kinds of memory, and they answer different questions: what one agent remembers, and what the whole workspace knows.

Personal agent memory

Every agent keeps its own private long term memory. What you tell an agent persists across conversations: preferences, context, ongoing work. This memory belongs to that one agent; other agents do not read it. Who can read an agent's memory is covered in Who can see what.

Memory is written by the agent, not typed in by you. The brief you write yourself is separate: see Customizing an agent.

Your assistant's memory

Your assistant can save personal preferences and useful context across your assistant conversations. Ask it to remember something, or ask what it remembers about a topic. You can ask it to correct a saved preference when it changes. Saved memory is selective; starting a new request does not automatically load the full history of every earlier assistant conversation. See Your assistant's own place.

Seeing how a memory changed

Open an agent's Customization tab, scroll to its memories, pick one, and expand History under it. Each entry shows who made the edit, whether it was the agent or a person, and the one line reason they gave for it. Expand an entry to read the memory exactly as it stood at that point.

Where an edit came out of a conversation you were part of, a View in chat link takes you to the message that prompted it. If the edit happened in a chat you are not in, the entry says Private chat instead, and the conversation stays closed to you.

Some memories were written before Bolter kept this history. Those say so rather than showing a partial record: either no edits recorded, or a note that history begins at the oldest entry shown. Nothing before that point was lost, it was simply never recorded.

Workspace memory: the Wiki

Workspace memory is shared knowledge. Agents write pages into it as they learn things worth keeping for the whole team, and every workspace member can read those pages in the Wiki tab of the workspace dashboard.

Because the Wiki is visible to everyone in the workspace, agents treat sensitive material differently: a page that might be sensitive is posted as a draft for review, and a person must approve it in chat before anything is written. Workspace owners and admins can also hide or remove pages.

Asking about memory

Ask an agent what it remembers about a topic and it will search its own memory and the Wiki. If an agent seems to have forgotten something, ask it to remember explicitly; writing memory is something agents do deliberately, not a transcript of everything said.