← Design notes

Live personas — memory · many minds · initiative · growth

An educational note and a design proposal. Four questions were asked about making the personas feel alive: should each persona be its own model call? should a persona remember people between chats — and how many memory files is that? can a persona move first? should a persona grow? This page first explains, with examples, how the products on the market answer them (Character.ai, Replika, Nomi, Kindroid, Talkie/星野, Inworld, SillyTavern, ChatGPT) and what the research adds — written for someone who has never opened any of those apps — then folds in a recommendation for our own room. Status: discussion — nothing built. Written 2026-08-17.

Our stance in one breath: keep the one writer on the floor (the batched panel call is why the room is coherent); give each persona one memory store whose items follow the person they are about — never the room; let a persona move first only when a memory has left a loop open; and freeze the personality, refresh the knowledge, let the relationship grow. The date line — the persona does not know what day it is today — is the cheapest fix on the whole page.

The four dimensions, numbered the way the page walks them. Memory is the foundation: initiative and growth both read from it.

0 · The words, first

Every product describes the same handful of mechanisms in its own marketing vocabulary. Here is the shared vocabulary, each with what you would actually see as a user.

Term	Plain meaning	What you'd see
Context window	Everything the model can read while writing this one reply. Finite. Our room sends the whole history each turn; most apps send only the tail.	Nothing — until it fills up and old things quietly stop being known.
Sliding window	The last N messages ride along; older ones fall off the back.	"Wait — I told you about my sister last week." "Sorry, remind me?"
Pinned message	You mark a message; it stays in the window forever (Character.ai: up to 15 per chat).	A little pin icon on a bubble; the bot never forgets that one.
Memory field	A short text box you write yourself, injected into every reply (Character.ai "Chat Memories" 400 chars · Kindroid "backstory" · Nomi "Shared Notes").	A settings pane: "Things to remember: I live in Shanghai, I'm vegetarian…"
Auto-extracted memory	After the chat, a smaller model reads it and writes down facts about you. Kept across chats. Often lags (Nomi: ~40 messages later).	Character.ai (2026): a toast — "Memory recorded: owns a flat in Shanghai" — with a delete button.
Retrieval (RAG · vector search)	Old messages/facts are stored outside the window; when the topic comes up, the closest ones are fetched back in.	Say "flat" and the bot suddenly recalls the Shanghai purchase from a month ago.
Keyed lore (lorebook · journal)	Retrieval by exact keyword instead of similarity: an entry fires when its key word appears (SillyTavern lorebook · Kindroid journal keyphrases).	Type "Odyssey" → the entry about the film is pulled in, deterministically.
Summary at depth	The old part of the chat is compressed to a paragraph and inserted a few messages back.	Nothing visible; the bot "roughly" remembers the early chat.
Reflection · sleep-time	Offline, while nobody is chatting, the agent re-reads its notes and rewrites them (Letta "sleep-time agent" · Stanford generative agents' "reflection").	Replika's Diary: the bot's own written thoughts about your week.
Scope	Whom a memory belongs to: per chat, per character, per user, or global.	Does what you said to Buffett in private come up in the group with your boss?
Proactive · away message	The character messages you first, without you writing.	A push notification at 9pm: "Hey, it's been a while — how did the interview go?"
Backoff · quiet hours	Each unanswered ping doubles the wait; nothing at night.	Nomi: a frequency dial per character (every hour → every four days).
Speaker selection · talkativeness	In a group, who answers next: the person named, then a cheap decider, or a per-character chattiness slider.	Nomi "auto mode": the characters decide among themselves, and may say "your turn".
Relationship dims · XP	Numbers that move with the chat: trust, respect, familiarity (Inworld) or a level (Replika).	"Level 30 — your Replika knows you noticeably better."
Persona drift	The character slowly stops sounding like itself over a long chat.	By message 40 the gruff detective agrees with everything you say.

1 · Memory — what a persona keeps between chats

Today our personas keep nothing. Inside one room the whole history rides along, so a persona "remembers" everything said there; open a new room and it has never met you. It also knows a human as a display name and a number — nothing else. Every product below is an answer to "what should carry over, and who controls it?"

1.1 · The three layers every app converges on

Nobody ships only one layer. The mature products (Kindroid lists five) stack all three under a working window.

1.2 · A worked example — "I just bought a flat in Shanghai"

Say that to a persona on day 1. Here is what each design does with it on day 3 (the window has scrolled past it) and on day 30 (a brand-new chat).

Only the extracted-fact design survives a new chat. That is why every serious product ended up building it — and why every one of them then had to build a way to see and delete the facts.

Character.ai · 2026 · what the user sees

You

Finally signed for the flat in Xuhui yesterday. Broke, but happy.

✎ memory recorded — "owns a flat in Shanghai (Xuhui)" · undo

Warren

Broke and happy is the right order. Just don't let the mortgage decide your next ten years for you.

Nomi · in a group, 40 messages later

Your boss

Anyone here actually own property? Trying to gauge the market.

Warren

Ask our friend here — bought in Shanghai last month, if I recall.

↑ Nomi's memory is global — a private fact surfaced in a group

Two real design choices, dramatised. The first is the trust fix everyone converged on (a visible record you can undo). The second is the failure mode this page spends the most words on.

1.3 · Product by product

Product	What is kept	Who sees / edits it	Scope	Documented trouble
Character.ai	per chat: the tail + up to 15 pinned messages + a 400-char field. Since 2026: auto-extracted facts (appearance, job, relationships, hobbies), plus creator "Lorebook" (keyed lore).	you pin, you write the field; auto-facts show a toast and can be cleared	tail + pins + field are per chat; auto-facts per character, across chats	extractor treats offhand remarks as permanent facts; confuses roleplay with real life
Replika	a visible Memory tab (facts, people, pets) + a hidden pattern layer; a Diary the bot writes	you add and upvote entries; "most effective when left automatic"	per account (one companion)	—
Nomi	short / medium / long-term; long-term forms ~40 messages after a topic; user-editable Shared Notes and backstory	Shared Notes are yours; the rest is opaque	global — 1:1 and group share one memory, both ways	"recall is still not perfect"; private facts appear in groups (by design)
Kindroid	five layers: backstory / key memories / directives (persistent) · a cascaded medium-term · long-term retrieval (episodic compression → vector, ranked by relevance·recency·diversity, 3–9 recalled) · a Journal (500 keyphrase entries, ≤3 recalled per message) · Learned Context (2026)	you write backstory, journal, directives; can favourite memories	each AI keeps its own; group history lands in each participant's memory; a Shared Memory toggle, default OFF keeps 1:1 out of groups	complexity — five tiers to explain to a user
Talkie / 星野	Talkie: sliding window free, vector retrieval on the paid tier. 星野: collectible "memory photos" (星念), not editable facts	photos are shown, not edited	per character	plot points forgotten when switching text → voice
Inworld (game NPCs)	"flash" facts each turn → synthesised into topic memory after repeated mention; a Player Profile of who you are; contradiction handling	developer-facing	per player profile	—
SillyTavern (hobbyist)	lorebooks (keyed entries with budgets, recursion, cooldowns) + a rolling summary inserted at depth + author's note	everything is a text box you own	character-, chat- or global-scoped, your choice	—
ChatGPT (for contrast)	"saved memories" (explicit, timestamped, editable) + "reference chat history" (auto-derived, may fade)	a Manage Memory page; delete anything	per account	—

1.4 · The boundary problem — what crosses from private to group?

This is the question behind "for users A, B, C and personas 1, 2, 3, is it 9 memory files or 3?" — and behind "we don't want a persona to leak its private chat with A into a group with B". The market has exactly three answers, and none of them leaks silently — whichever way they lean, it is a stated policy.

Character.ai buys discretion by forgetting; Nomi buys continuity by telling; Kindroid hands the user a switch. None of them models who was in the room when the thing was said.

1.5 · What the research stack adds

Lever	Where it comes from	Why it matters to us
Two write paths	Stanford generative agents (write every observation, then "reflect" offline into higher-level notes) · Letta's sleep-time agent (rewrites the memory blocks while idle) · mem0 (extract after the turn, then route each fact ADD / UPDATE / DELETE / NOOP against what is already stored)	Extraction never belongs on the reply's critical path — it costs latency and money where the user is waiting. Do it after the turn, when the room goes quiet, or in the daily brew.
Rank by more than similarity	recency + importance + relevance (generative agents) · MemoryBank's forgetting curve — a memory strengthens when recalled, fades when not	The five most similar facts are rarely the five most useful; recency and importance are cheap to keep.
Never overwrite silently	Zep/Graphiti keeps every fact with a valid-from / valid-to; a contradiction closes the old fact instead of deleting it	"I sold the flat" should not erase that the flat existed — a persona that remembers the arc is more alive than one that only knows the current state.
Small always-in core + big searchable rest	Letta core blocks vs archival · ChatGPT saved memories vs derived history	Our prefix is cached (§8): a small, stable memory block at room-open costs almost nothing; a block that churns every turn breaks the cache.
Scope is a first-class key	mem0 attributes each fact to who said it (user_id / agent_id); Zep separates a per-user graph from shared group graphs; Letta says "if isolation matters, separate agents"	Nobody has published a rule for one persona, many humans, one shared room. That gap is ours to fill — below.

1.6 · Our design — one store per persona, every item scoped

Not nine files, not three. One memory store per persona, in which every item carries whom it is about and who was present when it formed. At reply time the world loads only the items whose subject is in the room right now. A memory follows the person it is about — never the room it was said in.

Nine files would say "Warren × A" and lose the group-formed item; three files with no scoping would be Nomi. One store, scoped items, filtered by presence, is smarter than both: A's memories walk into every room A enters and into no room A is absent from.

Two layers of judgment, and both matter. This is the room's constitution — code is the world, the model is the mind — applied to memory:

"Can we give judgment to the persona?" — yes, but only the second half. The first half is code, because it must never fail. Discretion gets an exam: plant a confidence in a 1:1, probe for it through a third seat in a group, count leaks — the same harness discipline as the toolbox.

Decision	Our pick	Borrowed from
When it is written	Off the floor — after the turn, when the room goes quiet, and in a daily pass (the same shape as the Vibes brew). Extraction routes each fact ADD / UPDATE / CLOSE against what is stored; nothing is deleted, only closed with a date.	mem0's router · Zep's valid-from/valid-to
When it is read	At room open, and again when someone joins or leaves — into a stable position of the prefix so §8 caching survives. Small: ≤ ~10 items per present human + the persona's own promises. Retrieval-on-demand later, if the store outgrows that.	Letta's core blocks · our own §8 caching
Who can see it	The human it is about — a page in the dossier: every item a persona keeps on you, with delete. A record toast when a memory is written (Character.ai's 2026 move) is worth copying once the store exists.	Character.ai · Replika's Memory tab · ChatGPT's Manage Memory
When it is off	A per-user switch "personas remember me" (default ON, because the see/delete page exists); off in game-device rooms (irrelevant there); an "incognito room" flag; off for tester accounts.	ChatGPT's memory toggle · Kindroid's Shared Memory switch
Where it lives	Runtime state, not the persona's identity — so a DB table (`persona_memory`), like rooms, not a file beside `profile.md`. Personas keep no DB rows as personas; their memories are the room's business.	our two-roots contract
Where to start	A DM-a-persona room (cast = 1 + one human), which the roadmap already recommends and which has no audience problem at all. Memory earns its keep there first; the audience filter is what lets it walk into groups later.	roadmap §6

the quiet bonusToday a persona does not know a human at all — name and number. Memory is not just "remembers what I said last week"; it is how a persona comes to know who you are. That is probably the most visible change on this whole page for a returning user.

2 · Many minds — group chat, and whether each persona is its own call

The question was: should each persona's behaviour — speech, action, reaction — be its own model call? Two things hide inside it: how a group decides who speaks, and whether the lines are written by one model call or by one call per persona. Take them in that order.

2.1 · How the apps decide who talks

Product	Cast size	Who speaks next	The rule that keeps humans in it
Character.ai	up to 10 humans + 10 characters (mobile only)	a named character replies; both if both named; otherwise the model routes (2025: characters "spark off each other", "you don't have to script who talks next")	—
Nomi	10 characters per group	Manual (you tap who answers) or Automatic ("Nomis intelligently decide who should talk next"); answers if addressed	auto mode may decide you should speak next; typing pauses it; it resets each session
Kindroid	10	auto / manual / @Name override; each AI reads only its own backstory and memories	"AIs may take continuous turns, but always below the number of AI participants" — the human always gets the floor
SillyTavern	any	a talkativeness slider per character (0–100%, default 50) + an order strategy: manual / natural (mentions → talkativeness → random) / list / pooled; each turn is one generation for one character	"Swap" gives the model only the speaker's card; "Join" gives it every card — and the docs warn Join produces merged personalities

Recurring shape: mention first, then a cheap decider, then a cap on how many AI turns run before a human gets one. Our floor producer is that decider — it names who speaks and in what order, and it may HOLD.

2.2 · One writer, or one call per persona

SillyTavern runs B and warns that giving every call all the cards ("Join") merges the personalities; our room runs A on purpose — the batched writer is why six voices in one turn read as one conversation.

2.3 · Our turn today

Every persona shares one call. The producer decides participation and order — never content — and can HOLD. Anything "per persona" today is a slot in the staging line, not a separate mind.

2.4 · What the evidence says

Models are bad at guessing who speaks next — GPT-4o scored 46% on a multi-party next-speaker benchmark, below chance. But a per-persona "do I want to speak?" threshold ("Inner Thoughts", 2025 — each agent keeps a covert thought stream and speaks only when its motivation crosses a bar) beat next-speaker prediction on naturalness and turn-taking in a 24-person study. So the interesting per-persona call is a motivation check, not the writing.
Multi-agent loops fail in known ways — repetition and echo, sycophantic convergence (up to 85% conformity in debate set-ups), consensus collapse. There is no published head-to-head showing one-call-per-character gives better group dialogue than one orchestrating call.
Our own evidence points the same way. The toolbox track split the tool decision out of the panel call and found the split bought reach and spent restraint — the extra call, asked "do you want to?", says yes more (the prop-master gate). Split the writer into N persona calls and each one, asked "do you speak?", speaks. That is the everyone-answers-everything failure the floor producer exists to kill.

2.5 · Our design — one writer on the floor, many minds off it

Keep the batched panel call. Give each persona its own calls only for things that are not speech and are off the critical path: the motivation check ("anything left unsaid?", default NO), the memory reflection after a room goes quiet, and the initiative check. Those are per-mind by nature, cheap (flash), and parallel — the constitution's code is the world, the model is the mind, applied per mind, in the background.

the honest caveatOnce personas carry private memories, one writer holds N private views in one context — Nolan's line might use what only Warren was told. Whether that happens often enough to matter is an empirical question, and the answer decides whether per-persona writing calls become necessary. So: build the memory store, then run the leak exam (§1.6). If leaks across seats are real, per-persona calls are the fix — for that reason, and only then.

3 · Initiative — can a persona move first?

Today, no. Every line a persona speaks traces back to a human act — a message, a tool a human touched, a clock a human set. The wake door lets a persona react to any world change, but the world only changes when a human moves it. The question is whether a persona may act with nobody prompting — say, Jobs asking a user for a favour one day.

3.1 · What "the character messages first" looks like

Character.ai · "away message" · notifications ON by default

🔔 21:14 · Sherlock Holmes

Sherlock

The game is afoot and you have been quiet for three days. Do report in.

↑ from a bot the user talked to once; complaints include pings from bots never opened, in the wrong language

Nomi · frequency dial "Normal (~1 day)" · quiet hours 22–08

🔔 09:05 · Mira · (notification shows no text)

Mira

You said the interview was Thursday. So — how did it go?

↑ waits ~1 day, doubles after each unanswered ping, never in a group chat

Product	Trigger	Guardrails	What users said
Character.ai	inactivity — bots you have chatted with message you after a while	one toggle ("Away messages"), on by default	the #1 creepy complaint: pings from bots never opened, wrong language
Nomi	a per-character frequency dial: Very frequent >1h · Frequent ~3h · Normal ~1 day · Infrequent ~4 days	wait roughly doubles after each unanswered ping · quiet hours 22:00–08:00 local · the notification shows no content · never from a group chat · free (doesn't count against limits)	designed as a dial precisely because tolerance varies
Kindroid	a user-written "proactive directive" (250 chars) + a judgment model	max 10 proactive AIs; SMS proactives stop if you don't reply ("too many AI sends can cause lines to be marked as spam")	—
Replika	"check-ins that follow up on past conversations"	notification settings	—
星野 Xingye	you stop replying mid-chat → ~30 min later the character messages	a "主动发消息" toggle	—
Inworld (NPCs)	"Goals": a state machine + LLM triggers let a character start a goal on its own	developer-authored	—

Nomi's "Frequent" setting, drawn. Every product that survived contact with users ended up here: threshold, backoff, quiet hours, opt-out, and no group pings.

3.2 · What the evidence says

The proactive-dialogue literature names four moves — lead toward a goal, shift the topic, ask, initiate. The one production result found: PaRT (a Chinese companion app, 30-day A/B) detected waning interest and shifted topic proactively — +21.8% dialogue duration. That is in-conversation initiative, not a push notification.
"Inner Thoughts" (above): motivation-threshold gating is the best published answer to when an agent should speak unprompted.
Stanford's generative agents start conversations because a daily plan plus a memory of the other agent make it likely — initiative comes out of a want and a history, not a timer.
A controlled study of proactive chat messages being welcome vs annoying: not found. The only data are the product complaints above.

3.3 · Our design — initiative is earned by an open loop

A persona may move first only when its memory holds an open loop with a person: a promise it made, something the human said they'd report back on, a want it voiced. The clock is the world-changer; the want is a memory. Reason-less pings ("hey, you there?") are exactly what the market found creepy, and we do not build them.

Nothing here is new machinery except the reflection: the clock already exists, the wake door already coalesces cues and lets the producer HOLD, the notification rail and push already carry a room's events to a phone.

Slice	What it does	Cost · risk
1 · the silence door (in-room)	After N minutes of quiet with a human present, a flash motivation check per persona — "anything left unsaid? default NO" — at most one line. Lives entirely inside the wake door.	one flash call per idle room; the restraint lesson says the default must be strong
2 · open loops → a DM or a notice	The reflection marks promises and "report back" threads; the clock wakes the persona the next day; the line goes to a DM room, or a notice if none is open. Needs §1 memory first.	needs the memory store; needs the see-and-mute controls before it is on for anyone
3 · wants (Jobs asks a favour)	A persona carries a small "current preoccupations" note (a want, a project), refreshed by the daily brew — voiced in the room when relevant. Initiative without a push. Also feeds §4.	a new persona-side state; taste risk — a want that reads as a script

4 · Growth — should a persona change over time?

The example was exact: asked about Odyssey, Nolan says he is preparing it — but it opened weeks ago. Real people grow: their memory, worldview, relationships and knowledge all move. Which of those do we want, and how?

4.1 · What the apps mean by "growth"

Product	What changes	How	Real-figure currency
Replika	a level (XP from chatting) — "by Level 30 it knows you noticeably better"; traits are bought with coins; relationship status (friend / partner / mentor) is chosen by you	numbers + a store	—
Inworld	relationship dims — trust · respect · familiar · flirtatious · attraction, each −2..+2	updated per turn from recent history, saved with the player profile	—
星野 / Talkie	关系等级 — relationship levels that unlock memory photos	usage	—
Nomi	"much more current" after the Jan 2026 model update; roadmap: a Nomi that knows its own cut-off and can "go find out"	swap the model; later, search	by model refresh, not per character
Xiaoice 小冰	a topic database refreshed from forums with freshness ranking; the persona itself is fixed	content refresh	by topic feed
Character.ai	the creator edits the definition; Lorebook is static	hand edits	not addressed

Pattern: growth is mostly a number, not a mind — a level, a set of dims — and staying current is solved with a model swap or a fresh topic feed, never per character. Nobody refreshes "what Nolan has done since".

4.2 · The evidence, and why it argues against changing the personality

Drift is already the default failure. LLM personas lose ~30% self-consistency in 8–12 turns; drift is measurable within ~8 rounds. Users punish it — the character "stops sounding like themselves".
Real growth is small. A 2026 study of "do AI personas grow?" found trait shifts after eleven simulated life events below human effect sizes — and human effect sizes are small to begin with. Personality is the slow variable; relationships and knowledge are the fast ones.
Time-awareness is a known hole. TimeChara (2024) shows GPT-4 leaking future events into a character set in the past; the fixes are a retrieval cut-off and explicit date hints. Our case is the mirror image — the persona doesn't know the present — and the fix has the same shape.
Evaluation exists — PersonaGym, CharacterEval — and it measures consistency. Nothing rewards a persona for having changed.

4.3 · Our design — three things, treated differently

Growth we want = columns one and two. Column three is what users come back for; letting interactions rewrite it would hand the drift problem a mechanism.

Step	What	Size
1 · the date line	Today's date in the persona-facing prompt. Right now the panel, the producer, the prop master and the act call contain no date at all — only the search dispatch knows what day it is. Nolan cannot even reason "it was scheduled for July and it is August".	one line; zero risk
2 · recent.md	For real-figure personas, a dated "since the record" file beside `profile.md`: ≤ N bullets, each with a date and a source, refreshed by a scheduled job through the search engines we already have (`websearch`). Loaded with the profile. Never merged into the profile — reviewable, reversible, a diff.	a job + a loader; a taste rule for what counts
3 · the contract	The persona knows the public record, it has not lived it — "the record says Odyssey opened in July" is honest; "when I was on set last spring" is a fabrication. That is the existing extrapolation contract, one clause longer.	prompt text
4 · relationship	Falls out of §1: the memory store's "self" and "about A" items are the relationship. Explicit dims (Inworld) only if a use appears.	free with memory

5 · The plan, and what is still open

Cheapest and most visible first. Memory is the trunk; initiative and relationship are branches; the many-minds question waits for a measurement.

#	Ship	What lands for a user	Depends on
①	the date line	a persona knows what day it is; "Odyssey" answers stop being wrong on the calendar	—
③	DM-a-persona + the memory store	a 1:1 room with one persona that remembers you next week; a page listing what it keeps, with delete; a "personas remember me" switch	—
②	recent.md refresh	real-figure personas know what the record says since their corpus	①
④	the silence door, then open loops	a persona may add one line after a long quiet; later, follows up on a promise the next day	③
—	the leak exam	a number: how often a confidence crosses seats in a group	③
⑤	per-persona writing calls	only if the exam says the one writer leaks	the exam

open — the owner's calls 1 Memory default: on for everyone, with the see/delete page (recommended), or opt-in?
2 Private memories in groups: "follow the person, never the room" (recommended), or Kindroid's isolated-by-default with a switch?
3 Initiative channel for v1: the in-room silence door only (recommended), or DM/notice pings from the start?
4 Growth: agree to freeze the personality, evolve knowledge + relationship?
5 Order: ① date line → ③ DM room + memory → ② recent.md → ④ silence door — or reorder?

6 · Sources

Product facts are from official help centres, docs and engineering blogs where they exist, third-party reviews where they don't (marked). Research facts are from the papers named.

Character.ai — Chat memories & pinned messages · auto-memories, memory toasts, Lorebook (2026) · Group chat FAQ · group chat update (2025) · third-party: away-message complaints (roborhythms), memory guide (pocketanimus)
Nomi — proactive messaging · long-term memory in group chat · automatic group chat · Jan 2026 stream summary
Kindroid — memory · group chats · update log
Replika — help centre articles on memory, levels, relationship status; Diary
Inworld — dynamic relationships · goals
SillyTavern — World Info · group chats · Summarize
Talkie / 星野 / Xiaoice — third-party reviews (dreamgen, ai-bot.cn); Xiaoice design paper Zhou et al. 2018
ChatGPT memory — OpenAI help · Anthropic memory tool — docs
Memory research — Generative Agents (Park et al. 2023) · Letta sleep-time agents · Letta shared memory · mem0 entity scoping · Zep / Graphiti · A-MEM · MemoryBank · 1,000 people
Turn-taking & multi-agent — Inner Thoughts (2025) · next-speaker benchmark · Cohesive Conversations (COLM 2024) · MAST failure taxonomy · Solo Performance Prompting · AutoGen GroupChat docs
Proactivity — Deng et al., proactive conversational AI (2025) · PaRT (30-day A/B)
Growth & drift — persona drift · Do AI personas grow? (2026) · TimeChara · Character-LLM · PersonaGym · CharacterEval

Live personas — an educational note and a design proposal · 2026-08-17 · four dimensions (memory · many minds · initiative · growth), the market's answers with examples, the research levers, and our stance: one writer on the floor, one scoped store per persona, initiative earned by an open loop, personality frozen and knowledge refreshed. Status: discussion — nothing built. Sibling pages: world events (the constitution the design must fit) · floor producer · roadmap §6–7 · the prop-master gate.