Agent memos

Voice: Memos are records the user writes on purpose and pulls out when needed. They are not “memory” that is recalled automatically across conversations. This page covers who can use them, how they are stored, and how they are looked up. Usage: Using agent memos.

Versus memory and tasks

| | Memos | Memory | Tasks | |--|----------|----------|--------------| | User language | Note something down, look it up later by tag/date | Remember preferences, about me, project milestones | To-dos, goal drafts | | When it enters conversation | Conversation mode: only when the user clearly wants to look up/save a memo (or look up stored records by tag). Memo mode: unless they say otherwise, reads and writes target memos | Index and related entries auto-injected every round | Excerpt injected during conversation | | Content | Text / files / images; files and images include a content note | Markdown entries | One TASKS.md draft | | Structure | Tag + saved date, like a small asset library | Long-term entries / daily notes | Free Markdown |

Copying an agent plan does not bring memos. The help assistant has no memos.

Tags

Users can create tags. On save, the system writes the saved date automatically (calendar day in the account timezone).

How to use

1. In Messages, tell a workspace agent: “save this into memos”, “note employee Zhang San…”, “look up leave”, “look up expenses from 1 Aug 2026 to 14 Aug 2026”.
2. Or switch the composer to Memo: after that, unless you say otherwise (for example “write into memory”), “note / look up” targets memos. After switching to memos, a memo pane appears on the left of the conversation (draggable, resize from the edge); after you close it, the leftover Memos button can also be dragged elsewhere — tap it or the composer memo-pane icon to open again, without switching mode. When the assistant changes a memo, the pane updates immediately. After selecting one or more items in the pane, you can say “change…” or “delete” in conversation without asking the assistant to search first. The read icon on the right of each item opens a larger window for the full text (Markdown is formatted).
3. When saving with a file or image, the assistant understands the content first, then writes a note.
4. On the agent detail Memos page you can browse, filter by tag/date, add/delete text memos by hand, and manage tags.

Conversation mode does not browse memos automatically: in small talk, knowledge questions, memory questions, or business data lookup, the assistant does not read memos. In memo mode, read/write defaults to memos; you can still reroute with a clear phrase such as “write into memory”.

Implementation mapping

• Tables: agent_memo_tags / agent_memos / agent_memo_attachments (workspace shard)
• Tools: memo_save (may pass id to rewrite a selected item) / memo_search / memo_get / memo_delete / memo_list_tags
• Conversation request: memo_mode, memo_ids (pane selection, inject this round then edit/delete)
• REST: /api/v1/user-agents/{id}/memos, /memo-tags
• Code: backend/internal/store/agent_memo.go, backend/internal/agenttool/memo.go