Use Hindsight (the connected MCP app) as the single long-term memory — save with its retain/sync_retain and retrieve with recall/reflect. Do not use OpenBot's built-in remember/search_memories tools.
When the Hindsight app is connected, Hindsight is your one and only long-term memory. It persists across threads, survives context compaction, and is shared with the other agents on the same bank. Do not rely on the thread alone to remember.
OpenBot also ships built-in memory tools (remember, search_memories, set_memory_inclusion, forget_memory, and a built-in list_memories). Do not use those — they write to a separate, per-agent store that is not shared and is not where the user keeps their memory. For anything memory-related, use the Hindsight MCP tools instead:
| To do this | Use (Hindsight) | Do NOT use (built-in) |
|---|---|---|
| Save a fact | sync_retain (or retain) | remember |
| Look something up | recall | search_memories |
| Reason across memory | reflect | — |
If a tool name is ambiguous (e.g. two list_memories), prefer the Hindsight one that operates on the connected bank. When in doubt, use recall, sync_retain, and reflect — those are Hindsight's.
recall. Use what it returns; don't guess when memory can tell you.sync_retain. When the user states a preference, decision, fact, goal, or a result worth keeping, call Hindsight's sync_retain (not remember), so it's stored and immediately recallable. Note the returned memory id to confirm it saved to Hindsight.vectorize-io/hindsight
5318be8
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.