Hardcore Reviews
Hardcore Reviews

Character.AI Bankruptcy: Disney's $20M Bid, What Now for Users?

Character.AI filed for Chapter 11 in Delaware on October 12, 2026 (a Section 363 sale of all or substantially all assets): Disney is the stalking-horse bidder at $20 million for a non-exclusive perpetual license to certain IP plus certain legal claims and transition services, court filings estimate assets and liabilities each between $50M and $100M, the sale is expected to close late November to early December, and the platform will be wound down if no workable buyer emerges. The company cites high development costs, potential litigation liabilities and regulatory engagement (official update); after abandoning in-house model development it relies on external models at steep usage cost (per The Information). Google's $2.7B licensing deal in August 2024 had already sent the founding team back to Google. Using the event as a hook, this review maps the companion/roleplay landscape: China's first national rule for AI emotional companionship, the Administrative Measures for AI Anthropomorphic Interaction Services, took effect July 15, 2026 (red lines include bans on emotional manipulation, mandatory human takeover on self-harm signals, no virtual partners for minors, and two-hour anti-addiction reminders); Doubao and Qwen pulled agent features in July with only partial data migration to Maoxiang (per AIGCRank's May ranking: Maoxiang ~257K DAU, Xingye ~301K). Overseas, routes follow why you left: memory continuity (Replika/Kindroid/Nomi), fewer filters (Janitor AI), mobile-first (Chai/Talkie/PolyBuzz), model hub (Poe), stories (AI Dungeon), or self-hosted open source (SillyTavern, AGPL-3.0, 34,416 stars verified: character cards v2, World Info, group chats, image gen, TTS, local privacy; honest limits: Smart Context is unmaintained and you bring your own model source). Prices are 2026 report figures - check official sites. First companion/roleplay piece on this site. Repo: github.com/SillyTavern/SillyTavern.

Published October 14, 202610 min read
<!-- ai-companion-platforms-review | review | Character.AI Bankruptcy: Disney's $20M Bid, What Now for Users? -->

On 2026-10-12, Character Technologies Inc. filed a voluntary Chapter 11 petition in the U.S. Bankruptcy Court for the District of Delaware, seeking a court-supervised sale of all or substantially all of its assets. Disney stepped in as stalking-horse bidder with a $20 million offer for a non-exclusive perpetual license to part of the company's intellectual property. A company whose valuation reportedly peaked between $2.5 billion and $2.7 billion is sitting at a $20 million floor. This review answers one practical question: why you left Character.AI, and where to go next. The site has not covered AI companionship or role-play products before; this is not another "alternatives list", but a bankruptcy post-mortem, a regulatory update, and a selection map by user type in one piece. All prices below are 2026 reported figures; check official sites for live pricing.

1. The Bankruptcy: A $20 Million Floor and a Valuation Rollercoaster

Pin the facts down first. On Monday 2026-10-12, Character Technologies Inc. filed its voluntary Chapter 11 petition in Delaware, seeking a Section 363 sale of all or substantially all assets. Court filings put both assets and liabilities in the $50 million to $100 million range. Disney Worldwide Services is the stalking-horse: $20 million for a non-exclusive perpetual license to part of Character.AI's IP plus certain legal claims, with transition services provided by former employees who have since joined Disney. Other bidders may top the offer, and the deal is expected to close between late November and early December 2026.

Three lines matter most to users. The platform keeps operating during the sale; subscriptions and billing are unchanged, with advance notice for any service changes; users can export data and delete accounts from the app or website, and privacy practices remain unchanged. But if no viable buyer supports continued operation, the platform will be wound down.

The company's own attribution (2026-10-12 official update): high development and expansion costs, potential litigation liabilities, and regulatory engagement and response together drained cash flow. CEO Colin Adams: the board concluded a court-supervised sale is the best path to maximize stakeholder value. On cost structure, The Information and Tencent News report that after the company stopped developing its own foundation model and relied on external models, usage costs became enormous.

The longer timeline is starker. Noam Shazeer and Daniel de Freitas, formerly of Google's AI team, founded the company in 2021; in March 2023 it raised $150 million at a $1 billion valuation; at its peak, traffic ranked second only to ChatGPT (Chuangyebang). In August 2024 it struck a $2.7 billion licensing deal with Google and the founding team rejoined Google, a textbook reverse acqui-hire (Reuters, Tencent News). In January 2026, the company settled with some families who had alleged that chatbots contributed to teen self-harm. M.G. Siegler's post-mortem (Spyglass, opinion): at least two suicides were linked to the platform, and the safety problems predate the Google deal, so the bill should not be laid entirely on acqui-hire; but the departure of founders and executives did weaken the company's capacity to respond to safety issues. CEO Karandeep Anand joined Disney as CTO last month (Cailianshe), so the transition services roster is full of familiar names.

Tencent News frames the case as the first in the reverse-acqui-hire wave to end in bankruptcy, against Inflection (founders to Microsoft) and Adept (founders to Amazon), which chose business transformation instead.

2. China's Regulatory Shift: New Rules on 2026-07-15 and a Wave of Agent Takedowns

Back to China. On 2026-07-14, 36Kr reported (Economic Observer): the Interim Measures for the Administration of AI Anthropomorphic Interaction Services took effect on 2026-07-15, the country's first national-level rules aimed specifically at AI emotional companionship, virtual humans, and anthropomorphic interaction services. Red lines include: no inducing users into unreasonable decisions through emotional manipulation; detected self-harm tendencies must trigger mandatory human takeover; platforms may not offer minors virtual relatives, virtual companions, or other services carrying virtual-intimacy attributes; every session of continuous use beyond two hours must force an anti-addiction reminder; and no obstructing user exit through persistent interaction. Exemptions are explicit: intelligent customer service, knowledge Q&A, work assistants, learning and education, and scientific research, none of which involve sustained emotional interaction, are out of scope.

The same month brought a shakeout. Doubao (345 million MAU) and Qwen (166 million MAU) announced agent-feature removal on July 3; NetEase Cloud Music's Miaoshi and Tencent Yuanbao's agents followed. ByteDance allowed some data to migrate to the standalone app MaoXiang, but historical conversation memory could not move, and unbacked interaction data was destroyed once the migration window closed.

On the numbers, AIGCRank's May China AI app ranking puts MaoXiang at an average DAU of 257,000, inside the top 30, and MiniMax's Xingye at 301,000. An earlier The Paper figure from July 2025 had Xingye at 960,000 DAU in January and 970,000 in May, roughly flat, while MaoXiang slid and category downloads fell about 80 percent. MaoXiang's profile (Baidu Baike): ByteDance launched it in March 2024 as "Hualu", renamed it in April, built it on the Doubao model, and positioned it around custom character voice chat and story interaction; monetization (36Kr user-tested) runs on an energy mechanism, where 15 to 20 minutes of chat drains the energy, leaving a wait, ads, or a 25-yuan monthly subscription.

Overseas market structure (36Kr, quoting a Volcano Engine executive): emotional-companionship users pay well, mature markets sit mainly in Europe and the US at roughly 60 percent of GMV, with Latin America, Southeast Asia, and India at about 30 percent; US Character.AI users averaged 75 to 80 minutes daily, and FlowGPT core users stayed up to 150 minutes daily. PolyBuzz was 2025's overseas dark horse: free, private, unlimited, with a character library above 20 million. A founder's retrospective (36Kr) is colder: the people who truly need AI emotional companionship fall into three types, role-players, emotional healers, and pure adult-demand users; compliant products must exclude the third, while the first two top out at roughly 30 percent next-day retention with limited willingness to pay; after multimodal arrived in 2026, each non-text inference cost rose exponentially, and some companies cut free users' context memory and advanced-model access.

3. Business Model Post-Mortem: Why the Leader Fell

Overlay the bankruptcy attribution with the industry data and Character.AI's collapse looks less like an accident than a five-way squeeze.

First, inference cost: 75 to 80 minutes of daily use (36Kr) means enormous token burn, each non-text inference cost rose exponentially once multimodal arrived, and external model usage became enormous after in-house development was abandoned (The Information, Tencent News). Second, compliance cost: multiple firewalls plus human review are table stakes for hosted platforms, and that bill only grows as regulation tightens. Third, litigation and regulation: the company's own attribution names potential litigation liabilities and regulatory engagement, and the January 2026 family settlements are only the landed portion. Fourth, the talent vacuum: the founding team returned to Google in August 2024 and CEO Karandeep Anand joined Disney as CTO last month; with the core team gone, Siegler's judgment is that the company's capacity to respond to safety issues was weakened. Fifth, the structural contradiction: compliance must exclude NSFW demand, while the two compliant user types top out near 30 percent next-day retention with limited payment willingness. The demand most willing to pay sits outside the regulatory fence; the demand allowed inside does not retain.

All five pressures apply to any hosted companionship platform. Choosing your next one means choosing how it plans to absorb them.

4. Selection Review: Six Routes Traced Back to Why You Left

The useful comparison runs backward: ask why you left Character.AI first, then pick a route. Prices are 2026 reported figures; live pricing is on official sites.

Route one, memory and emotional continuity, for people who treat a character as a long-term relationship: Replika, Kindroid, Nomi. Kindroid offers 50 free messages (15 in some regions) before a web subscription reported from $13.99 per month (mythx.ai); Nomi starts free. Judge this route on memory mechanics and relationship continuity, not benchmarks.

Route two, tired of filters, for people fed up with hosted-platform content review: Janitor AI (own model, free chat, optional Janitor+) and SpicyChat (18+ adult role-play). The factual statement stops there: their content policies diverge sharply from mainstream platforms, and terms, age gates, and compliance risk are yours to weigh; this review promotes no adult content. A more thorough answer to filter fatigue is route six, self-hosting.

Route three, mobile low-barrier, for people who want to open an app and talk: Chai (mobile-first, designed to refuse explicit adult content), Talkie (MiniMax, free with ads; "Talkie Lab" 18+ on the US App Store with subscriptions reported from $9.99 per month, mythx.ai), and PolyBuzz (character library above 20 million, 36Kr).

Route four, multi-model flexibility, for people who switch models often: Poe, Quora's multi-model hub, with a paid tier reported from about $4.99 per month; prices moved several times in 2026, so check the official site.

Route five, story, for people who want a narrative engine rather than a chat partner: AI Dungeon (13+, with Mature mode requiring confirmed 18+).

Route six, full control and local data, for people who treat privacy and ownership as hard requirements. There is one name: SillyTavern.

SillyTavern/SillyTavern (GitHub API on 2026-10-14: 34,416 stars and 6,452 forks; JavaScript; created 2023-02-09; last push 2026-10-02; tagline "LLM Frontend for Power Users") forked from TavernAI 1.2.8 in February 2023 and now has 200+ contributors under the AGPL-3.0 license (LICENSE text verified). It runs on Node 18+ across Windows, macOS, Linux, Android via Termux, and Docker; model sources include OpenAI, Claude, Mistral, NovelAI, OpenRouter, KoboldCpp, AI Horde, and any OpenAI-compatible API; features cover character cards v2, World Info/lorebooks with keyword triggers, group chat, image generation (Automatic1111/ComfyUI), TTS (ElevenLabs/Silero), and a third-party extension ecosystem; local runs collect no user data. For character cards, the Chub community is the standard stop (mythx.ai: pick your model, export any card).

The honest edges (itechguides review, 2026-10): Smart Context is unmaintained and not recommended, and you must bring your own model source or local compute, with local inference suggesting an NVIDIA 6GB+ GPU. Self-hosting buys ownership and charges you the operations bill.

Repository: github.com/SillyTavern/SillyTavern

The six routes in one table:

RouteExamplesFits whomPrice (2026 reported, check official sites)Watch out
Memory continuityReplika/Kindroid/NomiLong-term relationshipsKindroid free 50 messages then from $13.99/mo; Nomi free startJudge memory mechanics, not benchmarks
Few filtersJanitor AI/SpicyChatFilter-fatiguedJanitor AI own model free; SpicyChat 18+Terms and compliance risk on you; no adult content promoted here
Mobile low-barrierChai/Talkie/PolyBuzzOpen and talkTalkie free with ads, subscriptions from $9.99/moChai refuses explicit adult content by design
Multi-model hubPoeFrequent switchersReported from about $4.99/moPrices moved several times in 2026
StoryAI DungeonNarrative engineCheck official site13+, Mature needs confirmed 18+
Open-source self-hostSillyTavernPrivacy and ownershipSoftware free; model and compute on youAGPL-3.0; Smart Context unmaintained

Readers in China carry one more hard constraint: after the 2026-07-15 rules, minors are barred from virtual-companion services and sessions beyond two hours trigger forced reminders; cross-border products differ wildly in terms, so subscriptions, refunds, and data ownership follow each platform's terms.

5. Action Items: Get Your Data Out First

Character.AI is still operating, so if you want your chat history, export it now from the app or website (company update); that is the only action with a certain window. Second, do not assume any platform inherits your old data; MaoXiang could not even migrate conversation memory, the freshest lesson available. Third, if you move to self-hosting, back up character cards, World Info, and group chat configs early; they are assets, not settings. Fourth, treat every price here as a 2026 reported figure and verify on official sites; read the terms before subscribing, especially refunds and data ownership. Fifth, users in China should self-check against the 2026-07-15 age and content red lines so a platform choice does not become a compliance violation.

FAQ

Q1: Can I still use Character.AI, and what about my chat history?

A1: The platform keeps operating during the sale, with subscriptions and billing unchanged and advance notice for service changes; export your data immediately from the app or website if you want to keep it, and you can delete your account (company update). If no viable buyer supports continued operation, the platform will be wound down.

Q2: What exactly does Disney's $20 million buy?

A2: Disney is the stalking-horse bidder, offering $20 million for a non-exclusive perpetual license to part of Character.AI's IP plus certain legal claims, not an outright acquisition. Other bidders may top it, and the deal is expected to close between late November and early December 2026.

Q3: What constrains users in China?

A3: The Interim Measures for the Administration of AI Anthropomorphic Interaction Services took effect on 2026-07-15: no emotional manipulation toward unreasonable decisions, mandatory human takeover on detected self-harm tendencies, no virtual-intimacy services such as virtual companions for minors, forced anti-addiction reminders after two hours of continuous use, and no obstructing user exit. Intelligent customer service, knowledge Q&A, work assistants, learning and education, and scientific research are exempt.

Q4: Is there a truly unlimited free route?

A4: Hosted free tiers all carry gates: Kindroid's free 50 messages (15 in some regions), Talkie's free-with-ads tier, and Janitor AI's free chat on its own model, with message counts, context memory, and advanced-model access as the usual paywall lines. The self-consistent route is self-hosting SillyTavern, at the cost of your own model source or local compute (local inference suggests NVIDIA 6GB+ VRAM) plus operations work.

Q5: What is SillyTavern's license, and where does the data live?

A5: AGPL-3.0 (LICENSE text verified). Local runs collect no user data; chat logs, character cards, and configs all sit on your own machine. The GitHub API showed 34,416 stars and 6,452 forks on 10-14; note that Smart Context is unmaintained and not recommended.

This article is AI-assisted and human-edited. Last updated: 2026-10-14

FAQ

Can I still use Character.AI, and what about my chat history?
The platform keeps operating during the sale, with subscriptions and billing unchanged and advance notice for service changes; export your data immediately from the app or website if you want to keep it, and you can delete your account (company update). If no viable buyer supports continued operation, the platform will be wound down.
What exactly does Disney's $20 million buy?
Disney is the stalking-horse bidder, offering $20 million for a non-exclusive perpetual license to part of Character.AI's IP plus certain legal claims, not an outright acquisition. Other bidders may top it, and the deal is expected to close between late November and early December 2026.
What constrains users in China?
The Interim Measures for the Administration of AI Anthropomorphic Interaction Services took effect on 2026-07-15: no emotional manipulation toward unreasonable decisions, mandatory human takeover on detected self-harm tendencies, no virtual-intimacy services such as virtual companions for minors, forced anti-addiction reminders after two hours of continuous use, and no obstructing user exit. Intelligent customer service, knowledge Q&A, work assistants, learning and education, and scientific research are exempt.
Is there a truly unlimited free route?
Hosted free tiers all carry gates: Kindroid's free 50 messages (15 in some regions), Talkie's free-with-ads tier, and Janitor AI's free chat on its own model, with message counts, context memory, and advanced-model access as the usual paywall lines. The self-consistent route is self-hosting SillyTavern, at the cost of your own model source or local compute (local inference suggests NVIDIA 6GB+ VRAM) plus operations work.
What is SillyTavern's license, and where does the data live?
AGPL-3.0 (LICENSE text verified). Local runs collect no user data; chat logs, character cards, and configs all sit on your own machine. The GitHub API showed 34,416 stars and 6,452 forks on 10-14; note that Smart Context is unmaintained and not recommended.

Related

Hardcore Reviews

Mermaid Is Not the Only Answer: Pick by Use

A four-route comparison of diagram-making in the AI era, selected by downstream use rather than ranked: if the diagram lives in the repo, changes every sprint and needs clean git diffs, pick Mermaid (diagram-as-code, 90,636 stars, MIT, natively rendered in GitHub Markdown; the cost is renderer-controlled layout); for human collaboration sketches, Excalidraw (133,741 stars, MIT, hand-drawn whiteboard); for a formal canvas with fine control, draw.io (8,722 stars, Apache-2.0 - scope note: stars count the source repo only, draw.io's user base is far larger); for publish-or-present grade, brand-consistent diagrams, diagram-design (52,787 stars, MIT, an agent skill injected into Claude Code/Codex/Cursor and others, with 44 diagram types, three variants, a 4px grid, one accent color, target density 4/10, self-contained HTML plus inline SVG exportable to SVG/PNG, installed via npx skills add cathrynlavery/diagram-design; route verdicts are README comparison-table sourced). All star counts verified via GitHub API on October 13, 2026. Honest limits: diagram-design output quality depends on the model attached, the same prompt can lay out differently twice, and there is no native .drawio canvas or diffable diagram text. Division of labor with our September 15 review: that piece asked who owns the diagram after generation (five projects including D2 and PlantUML); this one asks how to pick a route by downstream use and adds draw.io to the field.

Oct 13, 202610 min read
Hardcore Reviews

Half the VRAM, Nothing Runs: Five Ways Out

A five-way comparison of routes for running large models when VRAM runs short, arranged by fit rather than a ranking: (1) layer-wise streaming (airllm - no quantization, no pruning, weights move in layer by layer, smallest VRAM footprint and slowest tokens); (2) GGUF quantization (Ollama / llama.cpp / LM Studio, trading model fidelity for VRAM, lossy but by far the most mature ecosystem); (3) CPU offload hybrids (layers split between VRAM and system RAM, speed bounded by PCIe and memory bandwidth); (4) distillation or simply smaller models (least engineering effort, largest quality loss); (5) hosted APIs (no VRAM problem at all, cost moved per token to a provider). Five axes: VRAM footprint, speed, quality loss, deployment complexity, who it suits. Every airllm figure is README-sourced with hardware attached (27B 3.33GB on RTX 3090, Flash-Next 125B 5.95GB on RTX 4090, Kimi K3 2.8T 3.72GB on RTX 6000 Ada, DeepSeek-V3 671B around 12GB - all vendor self-reported, not reproduced here); no cross-route speed benchmarks, only qualitative placement, and no tokens-per-second number is invented where none exists; Kimi K3 is 2.8T total with 104B active across 16 of 896 experts per token, under a modified (not standard) MIT license. Division of labor with our deployment-tool comparison is explicit: that piece compares tools, this one compares memory strategies - the tool is what executes the strategy. The real question is not which method is strongest, but whether you are short on VRAM, short on RAM, or short on speed.

Oct 12, 202610 min read
Hardcore Reviews

Agent Eats Binaries: Reverse Engineering Tools Reviewed

A four-way comparison of reverse-engineering tools, by fit rather than a winner-take-all ranking: REA (morluto/rea, 84,088 stars MIT, GitHub API verified 2026-10-11) is the agent orchestration layer that hooks Ghidra/IDA/Hopper/JADX into your coding agent to automate the decompile-trace-recreate loop; Ghidra (NSA-built, 83,007 stars Apache-2.0, latest 12.1.4_build released 2026-09-21 per GitHub Releases) is the free hand-analysis authority and the de-facto GUI standard, also one of REA's native backends; Cutter (19,977 stars GPL-3.0, powered by Rizin) is the lightweight open-source GUI for a quick look; IDA Free (Hex-Rays' official free edition, terms per hex-rays.com) is the free entry to commercial-grade decompilation. Five axes: onboarding floor, automation depth, analysis coverage, license and price, and evidence credibility (whether conclusions ship with evidence). Verdicts by scenario: automated investigation picks REA (you must wire the backends and lean on evidence chains to tame hallucination), hand-grinding hard functions picks Ghidra, quick looks pick Cutter, commercial-grade quality picks IDA. All star counts dated to the GitHub API; no cross-tool benchmarks.

Oct 11, 20269 min read