Why doesn't my Muse side chat know what I told the main chat?

Side chats don’t share one live memory with the main chat. Members’ understanding, as of October 2026, is that each side chat has its own memory on top of your Muse’s long-term memory, and things move between them only when they are promoted, which takes time. Older side chats may never pick up a new instruction unless you tell them. Tell the side chat directly, or start a fresh one with a hand-off brief.

Side chats are the separate conversation threads you can open in the Muse app alongside the main chat. Keystroke called them “one of the only context mgmt primitive available to us,” so people use them a lot: in October 2026 Chonky had 12 and Nic had more than 50.

Steps

1. Nudge the old side chat. If you gave the main chat a new standing instruction, such as how to talk, go to the older side chat and tell it that its instructions changed. Nic says new side chats pick changes up and old ones usually don’t without that nudge.

2. Or replace it with a fresh one. Nic wrote a skill he calls /baton-pass: it reads a side chat’s whole history, writes a hand-off brief, starts a new side chat from that brief and archives the original. He uses it to refresh older side chats after big changes in the main chat.

3. Make Muse check before it answers. gumdean noticed that in voice calls, which go to his main chat, Muse sounded as if it knew what was happening in his build side chat when it hadn’t looked. His fix was a working rule, not a setting:

   • Each side chat is its own room.
   • Asked about work that lives in another room, Muse names that room, then reads it (and the day’s notes) before answering.
   • If it hasn’t read it, it says so.
   • A handoff isn’t done until the receiving room confirms it arrived.

   After adding the rule, his main chat went and read the other room and came back with what was actually there. “If your Muse ever sounds a little too seamless about stuff from another chat, that’s the tell.”

4. Name side chats for what they do. Nic and gumdean both said clear names help Muse say where something lives before it goes to read it.

5. Shape what gets promoted. Keystroke curates what moves up to main memory himself, and says you can influence the automated promotion with a line in your agent and memory files about what clears the bar.

Nic added one more trick: specific side chats can schedule a task to check in with each other by name, to coordinate related work.

Why it happens

Chonky’s description, from experimenting, with the caveat “could be wrong tho so pls take with grain of salt”:

a Muse has long-term memory, and each side chat has access to that plus a siloed memory of its own. every now and then some of that side chat-specific memory gets promoted to long-term memory.

So something said in the main chat can take a while to reach a side chat, and the other way round.

Nic added the piece about instructions, which he says Muse explained to him. A side chat starts with its own copy of SOUL.md, the file that holds the Muse’s persona and standing directives. Updates to the main chat’s SOUL.md don’t flow down to existing side chats, so new ones follow your latest directive and old ones keep the version they started with.

Keystroke expects an old side chat to catch up at its next context compaction, when the conversation is condensed. The feature he wanted most from the Muse team was a way to trigger compaction by hand and to see how full the context window is.