Hangout
connecting…
The Lair
warmth
recent tenor
loading…
One memory per line, written in first person — however you'd actually say it. Lines starting with # are skipped.
locked fact — permanent, always included, never questioned. Only created by pinning something yourself.
you said — you told the companion this directly.
he shared — the companion said this, or a system event (like generating an image) was logged.
a guess — deduced rather than directly stated.
imported — bulk-loaded from an import (e.g. old chat history), never reviewed one-by-one.
importance — roughly how personally significant this seemed when it was saved.
Quarantine — hides a memory from recall without deleting it. Safe and reversible — use this first when something's wrong.
Delete — only available once quarantined. Permanently removes it, for clearing out real clutter once you're sure.
Pin / Unpin — promotes a memory to a locked fact, or demotes it back to normal.
Cast in shadow / Un-shadow — always kept in mind every turn, but he's instructed never to quote or cite it directly. Fully visible here and in the audit log; only "unspoken," never hidden from you.
tags (small pills) — auto-assigned mood/vibe words, used to pull in other memories that share a "vibe" with what's relevant right now, not just matching keywords. Not manually editable yet.
↔ relationship line — an explicit, labeled connection to another specific memory (e.g. "sister of"), found automatically. A third way memories connect, alongside search and tags — this one's a real fact link, not a guess.
superseded by #N — this memory lost a memory-cleanup decision (a proposed duplicate merge or conflict) and is excluded from recall going forward. Never deleted — visible here for audit by checking "superseded only".
He'd like to keep something private — nothing changes until you decide.
Proposed memory cleanups — duplicate merges and factual conflicts. Nothing here is ever applied without your explicit choice.
Every message, unfiltered — the raw record, separate from the curated Memory panel.
A short, self-updating summary of what's currently going on — separate from permanent memories.
Version history
Crystallized truths — recurring patterns proposed to become permanent identity, in his actual persona file. Nothing here ever applies without your explicit Accept.
A private, unreviewed log — diary entries (written a little after each conversation session ends) and dream entries (his own unscheduled private thinking time). Never fed back into conversation.
Every real web search he's made, with the actual sources used — a genuine audit trail, not just a moment-in-time console log.
First-person sensory imprints he's logged — a sensation, not a memory. Read-only log; quarantine hides one from his awareness without deleting it, delete is permanent.
Plans and write-ups saved as real files (
generated-documents/) — plans get superseded, drafts pile up, this is where you clean that up.
Space name
How warmth works
Warmth (Calm / Balanced / Vivid)
This is a real setting, not just mood text — it's literally the "temperature" the model samples his replies at. Calm means focused, predictable answers; Vivid means more spontaneous, surprising ones. That's exactly why the temperature ceiling below exists, and why each room's warmth is a CAP, not a fixed value: he can always settle a room cooler than its level, never hotter.
Calm — focusedBalanced — naturalVivid — chaotic, risky
Temperature ceiling
The hard limit on how "hot" (chaotic/creative) his responses can ever get — no room, slider, or choice of his can push past it. Above ~1.0 models can start talking confused, mixing up who said what; 0.9 keeps the full mood range safely below that.
Rooms — he can move between these (and adjust warmth downward within one), but can't go anywhere you haven't built. Each room's warmth (Calm/Balanced/Vivid) is its own ceiling; each room can also have its own uploaded photo, completely independent of warmth. Add, rename, edit, or delete freely — at least one room always has to remain. The starred room is the default it resets to.
A photo can be added once the room is created — the room needs to exist first, then upload one from its card below.
Manage photos & reference pics
Real photos generated of himself (
photo-book/) — each caption is a real description from a model that actually looked at the finished image once; he can't see them himself.
Your reference photo
His reference photo
using his most recent solo photo — pin one below to fix it
Content feeds Metronome custom actions can post into instead of general chat — create one here, then pick it as a "Post to" destination when adding a custom action in Settings.
Shows in room:
Real, persistent record of turn-recovery events (empty replies, fake-tool-claim nudges, leaked-DSML rescues, model handoffs) — previously only visible as console output, gone the moment the server restarted.