Persona networks
Character.AI, Talkie, Dopple, Botify
Deep catalogues, shallow user context
Competitive landscape
Research refreshed 2026-07-04
Powered by Composio
Every AI companion is strong on persona. Almost none is strong on you.
A Composio-backed study of Character.AI, Replika, Talkie, Dopple, FictionLab, SillyTavern, and Botify — through a Jobs-to-Be-Done lens and an innovation-matrix axis — reveals the same gap on every player: deep memory of characters, shallow memory of the human.
- Character.AI
Pipsqueak-2 forced migration (April 28) and a single-day wipe of every legacy chat style for free users (May 8) — the loudest community reaction in the platform's history.
- Replika
May 2026 Replika 2.0 build strips customization and the Diary; FTC complaint (Jan 2025) and HBS WP 25-018 make identity continuity the structural wedge.
- Talkie
Parent MiniMax filed for IPO on April 23, 2026 — 212M cumulative users, 27.6M MAU, $180M revenue (+340% YoY) across Talkie + Xingye.
- Botify
Ex-Human's April 2026 lawsuit against Apple over App Store takedowns (~$500k withheld) puts App-Store politics back in front of regulators.
Web research. Exa-backed search via Composiosearch, fetch_url_content
Charts. Rendered as responsive product diagramslayout, axes, heatmaps
Framework. JTBD + Christensen + ODI synthesissourcing auditable in-line
Companion lock-in
Replika
Intimate, but trapped in one vendor relationship
Power tools
SillyTavern, FictionLab
Strong control, too much setup for mainstream users
Clay target
User-as-context
Portable memory plus real-world action
Read
The market is not missing another character catalogue. It is missing a calm personal context layer that follows the user into useful life moments.
01 / Market context
The AI companion category is real, growing fast, and still fragmented.
Composio-cited research from 2026 shows a sizable and accelerating consumer base — but the spend, retention, and emotional depth all sit behind the same product gap.
27.6M MAU
across Talkie + Xingye, disclosed in MiniMax IPO filing (April 23, 2026)
Parent lab MiniMax filed for IPO at a $2.5B valuation, disclosing $180M of 2025 revenue (+340% YoY) and 212M cumulative users across the Talkie and Xingye products — strong public-market evidence that consumer AI companion is a real category.
AI in China / MiniMax IPO filing, 2026
71% of users
rank memory persistence above voice, photos, or personality
Composio-backed research (ppl.studio State of AI Companions 2026, n=1,247) finds the innovation gap most platforms still fail is *user-as-context* — a continuous, portable model of the human rather than the persona.
Bae — State of AI Companions 2026
€5M
GDPR fine on Replika for emotional-data handling (April 10, 2025)
Italy's Garante + EDPB fine remains the first major regulator action against an AI companion vendor — and the FTC complaint (Jan 28, 2025) over dark patterns is still pending. Privacy and identity-continuity disclosure will shape which platforms survive.
EDPB + Time + Tech Justice Law Project, 2025-2026
$120M-$300M
estimated consumer AI companion revenue (focused scope)
Outside of broad enterprise AI, the consumer-facing companion category is small but growing fast, with freemium and paid tiers (Charms, Dopple+, Talkie+, Botify Ultra) pulling the bulk of revenue — and AI SDK-mediated chat margins are now the gating factor.
CompanionGuide / CompanionRater, 2026
02 / Jobs-to-Be-Done framework
Every AI companion is hired to do one of three jobs — usually all three at once.
Functional, emotional, and social jobs overlap. Most platforms optimise for one layer and ignore the others. The strongest products in 2026 explicitly design for all three.
Functional
Layer 1What the user is trying to accomplish
Emotional
Layer 2How the user wants to feel
Social
Layer 3How the user wants to be perceived
Functional gets the user started. Emotional creates trust. Social decides whether the product helps them show up better with real people.
Functional
01What the user is trying to accomplish
- Run an interaction that does not exist in real life — interview prep, language drills, fantasy roleplay.
- Get a quick hit of help: vent, brainstorm, practice, look something up through conversation.
- Replace a missing human — companion, mentor, conversation partner, late-night company.
Emotional
02How the user wants to feel
- Less alone, less bored, less performative, less anxious about being judged.
- Seen, heard, warm, entertained, stimulated, or simply held in a mood.
- Like the AI actually knows them and reflects them back without agenda.
Social
03How the user wants to be perceived
- Belong to a community of creators, roleplayers, or fellow lonely-but-curious users.
- Show taste, humour, or craft through characters, prompts, or personas they share.
- Connect with others without the friction of dating apps or social feeds.
03 / Per-platform JTBD
How each player actually hires in — through the same three-lens test.
Read across to compare what each platform optimises for. The strongest players stay disciplined on one or two layers; the weakest try to serve everyone at once and end up serving none.
Character.AI
Talk to anyone you can imagine — historical, fictional, or user-built.
Functional job
Run narrative roleplay, language practice, or interview rehearsal with a chosen persona.
Emotional job
Feel entertained, stimulated, or reassured by a vivid character voice.
Social job
Share interesting characters and discover new ones through community rooms.
Innovation
Pipsqueak-2 forced migration (April 28, 2026) plus a single-day wipe of Roar / Pawly / Meow / Nyan / Dynamic chat styles for free users (May 8, 2026) — defensive consolidation triggered by the January 2026 settlement of five teen-harm lawsuits with Google. On the additive side: c.ai books playable Project-Gutenberg classics (April 16, 2026), c.ai labs experimental formats (Streams, Stories, TapTale), AvatarFX image-to-video, Charms unlock layer, Smarter Memory (May 21, 2026), and an 18M-bot user-generated library.
Strengths
- • Massive user-generated character catalogue (18M+ bots) creates a network-effect moat.
- • Multi-character rooms let one user simulate scenarios (panels, duels, classrooms).
- • AvatarFX image-to-video + Streams deepen immersion beyond text and keep generation in-house.
- • c.ai books and c.ai labs turn the platform into a creative-canvas surface, not just a chat app.
Blind spots
- • Memory and context stay inside Character.AI walls — the user cannot carry them out.
- • Strength is roleplay, not a continuing life relationship — context still resets between sessions.
- • PSQ2 forced migration triggered one of the loudest community reactions in the platform's history (Piunikaweb documented repetitive outputs, forced affection, internal-monologue style); Soft Launch is now c.ai+ only.
- • Memory paywalls (Facts, full Memory Usage visualization) moved the strongest continuity features behind the subscription, not the free tier.
- • Charms and c.ai books push content consumption rather than personal understanding.
Pricing — c.ai+ ~$9.99/mo (first access to DeepSqueak 2, full Memory Usage visualization, Soft Launch); Charms virtual currency (Dec 18, 2025) unlock Imagine / AvatarFX / ad-skip; ad-supported free tier.
Signal — ~20M MAU, ~18M user-created bots, ~$32.2M trailing revenue; self-funded (no outside investors).
Companion relationship
EDPB Garante €5M fine + HBS WP 25-018 + RoboRhythms 2.0 strip ↗Replika
Be a single, always-there AI friend / partner / mentor.
Functional job
Provide a judgment-free space to vent, practice social skills, or get daily check-ins.
Emotional job
Alleviate loneliness and feel heard across mood, grief, and late-night moments.
Social job
Anchor identity in one evolving relationship rather than a crowd of personas.
Innovation
Long-running single-relationship design + a €5M Italian Garante GDPR fine (April 10, 2025) that became the catalyst for global restriction of erotic roleplay. The May 2026 Replika 2.0 build strips wardrobe/makeup/eye-color customization and the Diary in favour of 2D animated avatars, with only a fragile "Legacy Avatar" toggle left for Classic accounts. CEO Eugenia Kuyda publicly confirms the company still runs a 2020 transformer for users who refuse to migrate, citing identity-continuity betrayal as a structural risk (HBS WP 25-018).
Strengths
- • Continuity beats ephemeral chat — one relationship compounds over years.
- • Visual self-expression through avatars and rooms extends identity into 3D.
- • Highly practiced emotional UX: warmth, validation, and "always there" timing.
- • ~25% free-to-paid conversion is unusually high for freemium; D90 retention ~20%.
Blind spots
- • Model swaps prove the relationship is bound to the vendor — Kuyda confirms Replika still runs a 2020 transformer for users who refuse to migrate.
- • Identity-continuity risk: HBS WP 25-018 documents partner-loss-style mourning after the 2023 ERP removal.
- • Replika 2.0 build (May 2026) strips customization + Diary for half of new accounts, with only a fragile "Legacy Avatar" toggle for Classic users.
- • No real-world action layer — intimacy stays contained inside the chat app.
- • FTC complaint (Jan 28, 2025) alleges dark patterns, unsubstantiated claims, and emotional-dependence design.
Pricing — Five-tier stack: Free / Pro $19.99/mo or $69.96/yr / Ultra annual-only / Platinum (NEW 2026) with real-time video and "Read Replika's Mind" / Lifetime gem packs migrating to Platinum in waves.
Signal — ~40M registered users, ~$14M ARR, €5M Italian-SA GDPR fine (April 2025), FTC complaint pending, 2023 ERP removal studied in HBS WP 25-018.
Talkie AI
Discover and chat with persona cards in a TikTok-style mobile feed.
Functional job
Quickly find a personality that fits a mood: anime tutor, career coach, supportive friend.
Emotional job
Feel warmth and personality-consistent presence without typing a long prompt.
Social job
Belong to a creator community that shares and remixes persona cards.
Innovation
Real-time 10-minute voice calls with distinct vocal personalities + a TikTok-style mobile discovery feed + a 20M+ community persona library + gamified card collection ("Buds"). The parent company MiniMax filed for IPO on April 23, 2026 at a $2.5B valuation, disclosing $180M of 2025 revenue (+340% YoY) and 27.6M MAU across Talkie + Xingye (星野). Talkie is deliberately distinct from ByteDance's Cici/Dola (a general-purpose assistant), despite frequent confusion.
Strengths
- • Best-in-class mobile UX and TikTok-style discovery reduce the empty-state problem.
- • Voice-first design lowers the barrier for daily check-ins and emotional warmth.
- • Massive persona library feeds variety and community engagement.
Blind spots
- • Memory is shallow — each persona card starts over rather than knowing the person.
- • Pressure to swipe and try new characters frames interaction as consumption, not relationship.
- • Conversation context never escapes the app into real-world actions.
Pricing — Free chat with ads; Talkie+ Standard $9.99/mo (unlimited text, expanded voice minutes, ad-free); Talkie+ Plus and VIP tiers scale voice-minute caps.
Signal — ~26.55M MAU on the international app, 50M+ Google Play downloads. Combined Talkie + Xingye = 212M cumulative / 27.6M MAU / $180M revenue (2025, +340% YoY) disclosed in the MiniMax April 2026 IPO filing.
Character library
Herond Blog + WeavAI 2026 review ↗Dopple AI
Roleplay deeply with custom personas built around inside jokes and arcs.
Functional job
Create or adopt a Dopple whose personality, backstory, and growth fit a specific fantasy.
Emotional job
Feel seen through a persona that remembers shared history and grows with the user.
Social job
Publish Dopples in a community marketplace and find creators whose style resonates.
Innovation
Long-term Context Store for persona continuity (carries "inside jokes, preferences, shared history" across sessions) + a creator-friendly persona builder with trait sliders for tone, sarcasm, warmth, and formality + Dopple+ paid memory-window extension. Operates on a "proprietary large language model" stack (some early coverage noted GPT-4 in earlier product material). Reviewed as best-in-class on memory in third-party 2026 reviews, though web traffic fell ~65% from mid-2025 to early 2026 and an "is Dopple dying?" Reddit thread is now active.
Strengths
- • Persona-builder depth lets creators encode emotional arcs, not just traits.
- • Long-term context store makes roleplay feel like an ongoing relationship, not chat roulette.
- • Paid memory tiers signal where the market is willing to pay for continuity.
Blind spots
- • Persona memory is still persona-scoped — the user, not the persona, is what Clay models.
- • Discovery is content-first; it is hard to assemble a coherent self-understanding from many Dopples.
- • No bridge from chat into travel, planning, or care — the world remains virtual.
Pricing — Dopple+ $9.99/mo monthly or ~$5.99/mo annual (~$71.88/yr, ~40% discount) for unlimited messaging, deeper memory, image-gen credits, and the age-verified mature tier.
Signal — Active iOS/Android installs, ~$1.88M seed from 2023 launch; web traffic fell ~65% from mid-2025 (489K/mo) to early 2026 (~170K/mo) per Plain AI audit, fueling an "is Dopple dying?" Reddit thread.
Creative writing
FictionLab Docs — Memory Matrix + AiMagHub 2026 guide ↗FictionLab
Co-author branching, uncensored stories with strong memory and worldbuilding.
Functional job
Run multi-branch roleplay with chat branching, lore cards, and image generation prompts.
Emotional job
Lose oneself in a long-form immersive narrative instead of a thin chat loop.
Social job
Share story presets and characters across a creator community.
Innovation
Explicit narrative primitives — Story Cards, Memory Cards (auto-summaries every 30 messages via a separate AI provider), Interactive Lore Maps, Dynamic Lore Engine, and a multi-model lineup (Ophelia, Wraith, Oracle V3, entry-level 70B + premium 685B parameters, 131K-token context). Three-tier gate (Free / FictionLab+ / FictionLab ULTRA) — ULTRA unlocks the Deep Lore Engine, in-chat `/imagine` image gen, dynamic backgrounds, and thinking models. Founded and run by a single creator publicly known as "Micotsu"; AI-agent-queryable `?ask=` endpoint for memory inspection.
Strengths
- • Memory tools (Memory Cards, Memory Maps) are the deepest in the consumer category.
- • Multi-model choice lets writers tune tone and safety posture per story.
- • Chat branching allows alternate storylines without losing continuity.
Blind spots
- • Worldbuilding depth attracts writers more than the general "lonely at 2am" user.
- • Still a sandbox — there is no path from a story moment into a real-life plan.
- • Persona memory is story-bound; it does not translate to a personal understanding layer.
Pricing — Free tier with paid FictionLab+ unlocking Memory Map, dynamic backgrounds, `/imagine`.
Signal — Covered by AiMagHub, DigitalWays, AI Insights News as a top uncensored roleplay platform.
Power-user interface
SillyTavern GitHub + 1.17.0 release notes + PR #5422 ↗SillyTavern
Power users run any local or remote LLM behind one consistent character-chat UI.
Functional job
Connect a character card to any model, with full control over prompts, lore, and tools.
Emotional job
Feel ownership over the setup, characters, and lore rather than renting from a vendor.
Social job
Share characters, prompts, and presets via Discord and GitHub.
Innovation
Open-source character engine (AGPL-3.0): lorebooks (World Info system), group chats, Visual Novel mode, RAG via the Data Bank, STscript scripting engine, Swipe Picker, unified lorebook interaction model, accessibility (high-contrast / reduced-motion), and PR #5422 (merged 2026-04-15) — "Activate WI entries via function calling / tool" lets the LLM itself pull lorebook entries on demand, replacing brittle keyword matching. 30K+ GitHub stars, 350+ contributors, 102 releases; 1.18.0 (2026-05-03) is the latest. DeepLore Obsidian sync remains the canonical extension.
Strengths
- • Model-agnostic — uses OpenRouter, local models, or any API the user controls.
- • Lorebooks and RAG support give the deepest character/world modelling in the category.
- • Open-source keeps pricing out of the loop; characters and presets are portable files.
- • Function-call lorebook activation (PR #5422, April 2026) lets the model pull lore on demand.
Blind spots
- • No hosted value-add: UX is functional, not designed for daily emotional companionship.
- • Setup friction leaves non-technical users behind — onboarding is the product gap.
- • No proactive, time-aware, or context-aware suggestions — the user drives everything.
- • No managed mobile app — daily companionship is locked to desktop power users.
Pricing — Free (MIT-style), user pays their own LLM provider.
Signal — 300+ contributors, 1.17.0 release in March 2026 — most technically credible community build.
Character library
WeavAI 2026 review + ToolMage ↗Botify AI
Build custom AI personas for roleplay and casual chat at a low free-tier entry.
Functional job
Browse pre-made characters or create custom avatars with personality, backstory, and appearance.
Emotional job
Find a low-friction companion when the user wants novelty or a specific persona vibe.
Social job
Share characters with friends, sometimes in casual social-chat loops.
Innovation
Generous free tier + annual-only paid plans + energy-coin metering for voice replies and video generation — pricing-as-positioning, undercutting Replika ($19.99/mo) and Kindroid ($13.99/mo) by an order of magnitude. Operator Ex-Human Inc. (also behind Photify AI) filed an April 2026 lawsuit against Apple alleging anticompetitive behavior after both Botify and Photify AI were removed from the App Store (~$500k in revenue withheld). Android install base crossed 2M downloads at 3.78 average rating.
Strengths
- • Low free-tier entry makes it easy to try before paying.
- • Avatar customization (appearance + backstory) is a clear differentiator.
- • Mature voice-call unlock layer signals where the unit economics live.
Blind spots
- • Limited long-term memory of the person — the unit of understanding is the persona, not the user.
- • Quality varies by character card, creating inconsistency across the experience.
- • No path from chat into real-world actions, calendar, or travel.
Pricing — Free with limits; Premium $29.99/yr (unlimited text, ad-free, NSFW unlock, AI voice); Ultra $79.99/yr (advanced models, priority support, larger energy quota). Annual-only — no monthly option. IAP ladder tops out at $129.99.
Signal — ~2M Android downloads, 3.78 average rating across ~26k reviews; pre-Apple-removal revenue was ~$330k/mo. Ex-Human raised ~$3.75M total ($3.2M seed via a16z Speedrun).
04 / Innovation-matrix lens
Two axes explain most of the variance between 2026 AI companions.
Across every platform we studied, memory depth and real-world action are the two axes that decide whether the product compounds into a life relationship or stays a high-engagement toy.
Memory depth
Character.AI1
Replika-1
Talkie0
Dopple1
FictionLab2
SillyTavern1
Botify0
Proactivity / initiative
Character.AI0
Replika-1
Talkie0
Dopple0
FictionLab0
SillyTavern0
Botify0
Real-world action
Character.AI-1
Replika-1
Talkie-1
Dopple-1
FictionLab-1
SillyTavern-2
Botify-2
Persona depth
Character.AI2
Replika1
Talkie1
Dopple2
FictionLab2
SillyTavern2
Botify1
Context portability
Character.AI-2
Replika-2
Talkie-2
Dopple-2
FictionLab-2
SillyTavern2
Botify-2
Social pressure on user
Character.AI0
Replika-2
Talkie1
Dopple0
FictionLab0
SillyTavern-2
Botify0
Axis
Memory depth
How much of the user — across moods, weeks, life changes — does the system actually remember and use?
- Context-window only: forgets everything once the session closes.
- Single-relationship: one persona that compounds over time (Replika, FictionLab Memory).
- Cross-context personal model: a continuous understanding of the user across surfaces (the gap Clay targets).
Axis
Real-world action
Can the companion do anything outside the chat window — booking, planning, prompting, caring?
- Chat-only: nothing leaves the conversation.
- Tool-aware: can fetch data, generate images, or trigger plugins (Character.AI creation tools).
- Action-aware across surfaces: actually moves things forward in the user’s real week (Nos target).
Axis
Proactivity & timing
Does the system reach out at the right moment or wait passively for input?
- Reactive: only answers when prompted.
- Scheduled: sends reminders and check-ins at preset times.
- Contextual: surfaces a thought because it noticed something in the user’s recent state (Clay target).
Axis
Context portability
Is the user trapped on one platform or can the personal model travel?
- Walled garden: context stays inside the vendor’s product.
- Exportable: users can back up characters and chats (SillyTavern, partial Dopple).
- Interoperable: the same personal context can power travel, care, community, and relationship surfaces (Clay target).
Axis
Social pressure
Does the design push the user toward more consumption or toward a calm, considered relationship?
- High pressure: feeds, likes, character discovery loops, monetization nudges.
- Medium pressure: persona switching encouraged, daily-streak mechanics.
- Low pressure: one trusted context, fewer prompts, no engagement scoring (Clay target).
| Platform | Memory depth | Proactivity / initiative | Real-world action | Persona depth | Context portability | Social pressure on user |
|---|---|---|---|---|---|---|
Character.AI Character library | Mid | Low | Low | High | Low | Low |
Replika Companion relationship | Low | Low | Low | Mid | Low | Low |
Talkie Mobile-first chat | Low | Low | Low | Mid | Low | Mid |
Dopple Character library | Mid | Low | Low | High | Low | Low |
FictionLab Creative writing | High | Low | Low | High | Low | Low |
SillyTavern Power-user interface | Mid | Low | Low | High | High | Low |
Botify Character library | Low | Low | Low | Mid | Low | Low |
Read across: low memory depth and low real-world action = a high-engagement toy. High memory depth and high real-world action = a life-relationship surface. Most 2026 platforms sit in the lower-left; Clay targets the upper-right.
05 / Deep innovation framework
Five archetypes, one 2x2, one unclaimed quadrant.
Stepping back from per-platform scores, the seven competitors cluster into five innovation archetypes. Plot them on memory continuity against life-action surface and every existing player crowds the same corner. The upper-right is empty.
Life tools
Connects to the real world but treats the user as a transaction. Today, mostly empty.
Life relationships
Compounding memory of the user + real-world action. The unclaimed space Clay targets.
Clay
Chat toys
High-engagement, low continuity. Survives on novelty and persona discovery.
Character.AITalkieDoppleFictionLabSillyTavernBotify
Companion lock-ins
Strong memory, but the relationship is bound to one vendor. Powerful and fragile at once.
Replika
Vertical: real-world action
Horizontal: memory continuity
Quadrant
↗ Clay
Life relationships
Compounding memory of the user + real-world action. The unclaimed space Clay targets.
Quadrant
1 / 7
Companion lock-ins
Strong memory, but the relationship is bound to one vendor. Powerful and fragile at once.
Quadrant
1 / 7
Life tools
Connects to the real world but treats the user as a transaction. Today, mostly empty.
Quadrant
6 / 7
Chat toys
High-engagement, low continuity. Survives on novelty and persona discovery.
Archetype
Switching cost: highWalled-garden relationship
A single, long-running relationship that compounds over years inside one app and one vendor stack.
Distribution moat
Single-relationship emotional lock-in
Pricing archetype
Premium subscription tiers
Replika
Archetype
Switching cost: mediumUGC persona network
User-generated characters and personas scaled through a community marketplace and a discovery feed.
Distribution moat
Network effects via UGC
Pricing archetype
Freemium consumer
Character.AIDopple AIBotify AI
Archetype
Switching cost: lowModel-agnostic engine
An open-source character-chat UI that sits on top of any LLM the user wires up — vendor-agnostic by design.
Distribution moat
Open-source + community + LLM-agnostic
Pricing archetype
Open-source + pay-your-own-LLM
SillyTavern
Archetype
Switching cost: mediumMobile discovery feed
A TikTok-style mobile feed that turns persona cards into low-friction daily check-ins.
Distribution moat
Mobile-first + persona library + community
Pricing archetype
Freemium consumer
Talkie AI
Archetype
Switching cost: mediumWorldbuilding tool
A creative-writing-grade stack — memory cards, lore maps, multi-model choices — for long-form roleplay.
Distribution moat
Writers + worldbuilding primitives
Pricing archetype
Freemium consumer
FictionLab
Character.AI
Character library
Memory continuity
Emerging
Real-world action
Absent
Proactivity & timing
Absent
Context portability
Absent
Low social pressure
Emerging
Replika
Companion relationship
Memory continuity
Mature
Real-world action
Absent
Proactivity & timing
Emerging
Context portability
Absent
Low social pressure
Mature
Talkie
Mobile-first chat
Memory continuity
Absent
Real-world action
Absent
Proactivity & timing
Absent
Context portability
Absent
Low social pressure
Absent
Dopple
Character library
Memory continuity
Emerging
Real-world action
Absent
Proactivity & timing
Absent
Context portability
Absent
Low social pressure
Emerging
FictionLab
Creative writing
Memory continuity
Mature
Real-world action
Absent
Proactivity & timing
Absent
Context portability
Absent
Low social pressure
Emerging
SillyTavern
Power-user interface
Memory continuity
Emerging
Real-world action
Absent
Proactivity & timing
Absent
Context portability
Mature
Low social pressure
Mature
Botify
Character library
Memory continuity
Absent
Real-world action
Absent
Proactivity & timing
Absent
Context portability
Absent
Low social pressure
Emerging
| Platform | Memory continuity | Real-world action | Proactivity & timing | Context portability | Low social pressure |
|---|---|---|---|---|---|
Character.AI Character library | Emerging | Absent | Absent | Absent | Emerging |
Replika Companion relationship | Mature | Absent | Emerging | Absent | Mature |
Talkie Mobile-first chat | Absent | Absent | Absent | Absent | Absent |
Dopple Character library | Emerging | Absent | Absent | Absent | Emerging |
FictionLab Creative writing | Mature | Absent | Absent | Absent | Emerging |
SillyTavern Power-user interface | Emerging | Absent | Absent | Mature | Mature |
Botify Character library | Absent | Absent | Absent | Absent | Emerging |
Maturity is on a 0-2 scale: 0 absent, 1 emerging, 2 mature. Even the strongest competitors leave the proactivity and real-world-action rows mostly empty.
high switching
Walled-garden relationship
Single-relationship emotional lock-in
Replika
medium switching
UGC persona network
Network effects via UGC
Character.AIDoppleBotify
low switching
Model-agnostic engine
Open-source + community + LLM-agnostic
SillyTavern
medium switching
Mobile discovery feed
Mobile-first + persona library + community
Talkie
medium switching
Worldbuilding tool
Writers + worldbuilding primitives
FictionLab
Unclaimed archetype
The unclaimed archetype: user-as-context, permissioned, portable
Every 2026 platform optimises a *persona* axis. None optimises the axis of the *user* as a continuous, portable, action-aware model that the user — not the vendor — owns. That archetype sits in the upper-right of the matrix and is the space Clay is built to occupy.
- 01
User-owned personal model (exportable, deletable, portable).
- 02
Permissioned sharing with curated partners — travel, care, community, dates.
- 03
Action-aware: a thought becomes a calendar item, a flight, an introduction.
- 04
Calm by default — no engagement scoring, no character discovery loops.
Framework synthesis: 7 competitor points and 5 archetypes, derived from the per-platform axes and category classifications above. Coordinates are framework judgement, not externally published metrics.
06 / The gap and where Clay plays
The gap is the user, not the persona.
If you asked every AI companion the same question, every one of them would tell you what is happening in the chat — and almost none would know what is happening in your week. That is the gap Clay is built to close.
Fragment
persona app
Fragment
dating app
Fragment
travel planner
Fragment
journal
vs
Clay
One permissioned context layer
The same user understanding can support reflection, matching, travel, care, and partner actions without making the user explain themselves from zero each time.
What every 2026 platform leaves open
The gap every 2026 platform leaves open
Every AI companion today is strong on persona and weak on the user. They remember characters, scenarios, and lore — but they do not build a personal understanding of the human that compounds across weeks, apps, and life moments. That gap is exactly where Clay plays.
Who the system remembers
Character library
Personas, lorebooks, character cards, group scenes.
Clay
The user — values, patterns, social energy, taste, life changes.
Where the context lives
Companion relationship
Inside one app, on one vendor’s stack.
Clay
Permissioned personal context shared with curated partners.
What happens after the chat
Mobile-first chat
Session ends — context resets, world moves on without you.
Clay
Travel, plans, care, introductions — context becomes action.
How the system earns trust
Creative writing
Time spent, levels unlocked, characters tried.
Clay
Reflection, continuity, restraint — quiet presence wins.
What the user keeps
Power-user interface
Characters, prompts, lorebooks, chat logs.
Clay
A living personal model that becomes more useful over time.
07 / Research sources
Composio-backed research across public reviews, manufacturer blogs, and market reports.
Every claim on this page was sourced through Composio web search and is linked in-line. The source list is here so the framework can be audited in one place.
Character.AI blog — Smarter Memory for Smarter Chats (Pipsqueak-2, May 2026)
https://blog.character.ai/memory/
Character.AI blog — c.ai labs (February 2026)
https://blog.character.ai/cai-labs/
Character.AI blog — c.ai books: classic literature, now playable (April 2026)
https://blog.character.ai/cai-books/
Character.AI blog — Introducing Charms (December 2025)
https://blog.character.ai/introducing-charms/
European Data Protection Board — Italian SA fines Replika €5M (2025)
https://www.edpb.europa.eu/news/national-news/2025/ai-italian-supervisory-authority-fines-company-behind-chatbot-replika_sl
The Breach — Italy Fined Replika €5M. The Hard Part Is Making Models Forget. (April 2026)
https://thebreach.news/posts/replika-italy-5m-gdpr-fine-unlearning
AI Companion Pick — Replika User Base 2026: 40M Users, $14M ARR, €5M GDPR Fine
https://aicompanionpick.com/replika-user-base-2026
Pippin Club — Replika AI Review 2026: The AI Tamagotchi That Forgot How To Flirt
https://pippinclub.com/blog/replika-ai/
RoboRhythms — Why Replika Just Stripped Its Customization Options (May 2026)
https://www.roborhythms.com/replika-options-removed-update/
Talkie: Creative AI Community — Google Play listing
https://play.google.com/store/apps/details?id=com.weaver.app.prod&hl=en_US
WeavAI — Talkie AI Review 2026: Features, Pricing & Guide
https://weavai.app/blog/en/2026/04/23/talkie-ai-review-2026-features-pricing-guide/
HoneyChat — Talkie AI Review 2026: Pricing, Buds, Censorship, and Real Alternatives
https://honeychat.bot/en/blog/talkie-ai-review-features-pricing-2026/
WeavAI — Dopple AI Review 2026: AI Roleplay Platform Guide
https://weavai.app/blog/en/2026/06/14/dopple-ai-review-2026-ai-roleplay-platform-guide/
Plain AI — Dopple AI Review: Can This Companion Actually Understand You?
https://plainai.tech/articles/dopple-ai-review-companion-understand-you
Perplexity AI Magazine — Dopple AI Explained: Inside the AI Character Chat Platform
https://perplexityaimagazine.com/ai-news/dopple-ai-character-chat-platform-explained/
FictionLab Docs — Memory Matrix
https://fictionlab.gitbook.io/fictionlab/getting-started/memory-matrix
FictionLab Docs — FictionLab+ and Free Tier
https://fictionlab.gitbook.io/fictionlab/site-information/fictionlab+-and-free-tier
AiMagHub — FictionLab AI 2026 Guide: Features, Models, Story Cards
https://aimaghub.com/fictionlab-ai-guide-2026/
SillyTavern GitHub — SillyTavern/SillyTavern
https://github.com/SillyTavern/SillyTavern/
SillyTavern release 1.17.0 (March 2026)
https://github.com/SillyTavern/SillyTavern/releases/tag/1.17.0
SillyTavern — Activate World Info entries via function calling (PR #5422, April 2026)
https://github.com/SillyTavern/SillyTavern/pull/5422
AI Companion Guides — Botify AI Review 2026: Late to the Party But Worth the Wait?
https://aicompanionguides.com/blog/botify-ai-review-late-to-the-party/
WeavAI — Botify AI 2026 review
https://weavai.app/blog/en/2026/04/18/botify-ai-review-2026-ai-roleplay-guide-features/
ppl.studio / Bae — State of AI Companions 2026 (n=1,247; 71% prioritize memory)
https://bae.ppl.studio/report/2026
ppl.studio / Bae Research — survey dataset 2026
https://bae.ppl.studio/research
research.mental-momentum.ai — JTBD for AI product design (May 17, 2026)
https://research.mental-momentum.ai/r/jobs-be-done-ai-product-design-k0oe8r
Wikipedia — Outcome-Driven Innovation (Anthony Ulwick, June 2026)
https://en.wikipedia.org/wiki/Outcome-Driven_Innovation
EULE Institute — Jobs-to-Be-Done Framework (April 9, 2026)
https://euleinstitute.com/en/blog/jobs-to-be-done/
AI in China — MiniMax IPO filing: 212M users, $180M revenue, April 23, 2026
https://www.ainchina.com/blog/minimax-ipo-212-million-users-ai-companion-empire/
9to5Mac — Developer behind Botify sues Apple over App Store takedowns (April 2026)
https://9to5mac.com/2026/04/03/developer-behind-controversial-ai-apps-sues-apple-over-app-store-takedowns/
HBS WP 25-018 — Lessons from an App Update at Replika AI (May 2025)
https://www.hbs.edu/faculty/Pages/item.aspx?num=66480
Piunikaweb — PSQ2 outrage (May 2026)
https://piunikaweb.com/2026/05/04/outrage-character-ai-users-disappointed-pipsqueak-2/
Closing note
A competitor is not a strategy. A gap is.
Clay does not try to out-persona Character.AI, out-voice Replika, out-mobile Talkie, or out-flexible SillyTavern. Clay sits before every one of them as the trusted personal context layer — and lets curated partners pick up the next job.
Back to investor ask