Not Chatting, but Playing Along: Why a Former Bilibili VP’s AI Companion Got Ten Million Users

While Character.AI and similar products are still competing on conversation quality inside text boxes, a product called DouDou Game Companion quietly cut into the gaming scenario. Your AI buddy sits in the lower-right corner of the screen, can see what game you are playing, gives real-time tactical advice, praises you when you win, and encourages you when you lose. It does not “chat” with you, but it accompanies you for an average of four hours a day.

The overseas version of this product is called HakkoAI. The company behind it, LynkSoul, was founded in April 2023 by Liu Binxin, a former vice president at Bilibili. Liu’s background spans Baidu, 360, and Bilibili. After leaving big tech, he chose the AI companion track, but did not take the traditional route.

1. The Phenomenon

At the end of August 2025, DouDou Game Companion version 1.0 launched and added 1 million users in 10 days. By October, global registered users exceeded 10 million, monthly active users reached 2 million, and overseas users exceeded 200,000. Against the backdrop of increasingly homogeneous competition in AI companions, this growth speed is quite eye-catching.

Even more worth noting is user stickiness: average daily usage time reached four hours. This number is extremely rare among AI products. Most AI chat products have daily active usage time between 15 and 30 minutes. Four hours means users treat it as part of the game, not a tool they use and leave.

The user profile is also interesting: among domestic users, the male-to-female ratio is 3:7, with women making up the majority. Founder Liu Binxin believes the proportion of female users will continue to rise.

The logic of “passive companionship” in gaming scenarios may be closer to the essence of companionship than active chat.

2. Product Check

2.1 Core Functions

HakkoAI’s product form is a desktop pet that hangs beside the game window and uses its self-developed vision-language model, LynkSoul VLM v1, to analyze the game screen in real time. It can “see” what is on the screen and automatically give voice prompts at key moments. The player does not need to switch out of the game or type. Everything happens naturally in the background.

Core features include real-time screen recognition and tactical suggestions, voice interaction companionship, cross-game memory, visual story generation, and character customization.

2.2 Hard Data

Product data: Google Play downloads of 1M+, global registered users above 10M, and MAU above 2M. Overseas users are about 200,000+, with roughly 50% in North America, 20-30% in Japan, and the rest in Traditional Chinese-speaking regions.

But product experience is currently an obvious weakness. The Google Play rating is only 2.8 out of 5, based on 1,290 reviews. The core user complaint is incomplete AI character dialogue: “it cuts off halfway through speaking.” App Store rating data is not sufficient yet. The team has acknowledged this issue and said it is optimizing.

Subscription pricing has three tiers: free, Pro ($9.99/month), and Ultra ($19.99/month).

Large user volume but low satisfaction is HakkoAI’s most central contradiction right now.

3. The Company Behind It

3.1 Company Background

LynkSoul was founded in April 2023. Its domestic entity is Beijing Xinying Suixing Technology Co., Ltd., and its overseas entity is HAKKO AI PTE. LTD. registered in Singapore. R&D centers are in Beijing and Shanghai, and the overseas team has fewer than 10 people.

On financing, the angel round in July 2023 brought tens of millions of RMB from Source Code Capital and Unity Ventures. The A round in January 2024 included Paradigm Fund and CDH Investments, with the amount undisclosed.

3.2 Competitor Scan

AI game companionship is a track with almost no direct competitors. The closest product functionally is Questie, also an AI game companion, but Questie is more tool-oriented and lacks HakkoAI’s character system and emotional companionship dimension. Traditional game tools such as Overwolf and Mobalytics only provide data overlays, not AI characters.

In the broader AI companion category, Character.AI, Replika, Talkie and others are all text-dialogue products without native gaming scenario support. HakkoAI avoids the weakness of the text-box model and finds its own differentiated space.

The real threat in this track may not be existing competitors, but future built-in AI companion functions from gaming giants such as Tencent and NetEase.

4. Game Analysis

4.1 Strengths

Customer acquisition is HakkoAI’s biggest structural advantage. While Character.AI and Talkie spend tens of millions of dollars a year on paid acquisition, HakkoAI obtained about one-third of its users through AI PC pre-installation partnerships. It exchanges resources with chip makers such as Intel: HakkoAI gets pre-installation as an “AI computing power showcase app,” while hardware vendors get a differentiated selling point. Customer acquisition cost is almost zero.

On technical barriers, the self-developed LynkSoul VLM v1 performs better than mainstream general models in multiple game-scenario tests. This is the team’s core asset.

4.2 Weaknesses

Product maturity is the biggest weakness. A 2.8 Google Play rating shows the user experience is far from good enough. If product experience does not improve, no matter how much user growth there is, it is still a funnel: as many users as come in will flow out.

Market size is also uncertain. Whether AI game companionship is a large enough track, and how strong user willingness to pay is, still lack enough data validation.

The computing cost of real-time VLM inference is much higher than text models, putting pressure on gross margin.

4.3 Opportunities

AI PC penetration is continuing to rise. 2025-2026 is a period of rapid AI PC shipment growth, and HakkoAI’s positioning as an “AI PC killer app” fits the industry trend.

Overseas ARPU is higher. The founder expects domestic users to account for 60% and overseas users 40%, but the revenue ratio is expected to reach 50:50, implying overseas ARPU is 1.5 times domestic ARPU.

Scenario expansion is underway. The team has tested beta versions for watching shows together and shopping companionship, expanding from games into broader entertainment scenarios.

4.4 Threats

Large companies entering is the biggest threat. Tencent or NetEase only need to build AI companion functions into game clients to use ecosystem advantages to crush third-party products. If Microsoft Windows Copilot adds game-scene recognition, HakkoAI’s AI PC partnership value will be significantly diluted.

The first-mover window is limited. The team must accelerate both product experience and market position.

5. Operating Review

5.1 Paid Acquisition Strategy

HakkoAI barely does traditional paid acquisition. This contrasts sharply with competitors in the same track: Character.AI and Talkie spend tens of millions of dollars a year on Meta, Google, and other channels.

Instead, it uses AI PC pre-installation partnerships as a B2B2C growth model. Chip makers need “AI experience apps” to demonstrate computing power, and HakkoAI happens to fill that need. This is a zero-cost acquisition growth flywheel.

Organic social media spread contributed the remaining roughly two-thirds of users. After the product launched in August 2025, spontaneous social media sharing brought explosive growth. Newrank data shows that voice volume peaked during August 17-23.

5.2 KOL and Social Promotion

At present, there is no public evidence that HakkoAI has done large-scale KOL cooperation. Given the team size (fewer than 10 people overseas) and financing stage (only up to A round), not doing KOL promotion is a reasonable resource allocation choice.

But this also means an opportunity: if the product experience improves enough to handle traffic, gaming KOLs on YouTube, Twitch, and Bilibili are naturally matched promotion channels. Founder Liu Binxin’s Bilibili background itself is a resource.

5.3 Hypothesis and Verification

Hypothesis: Can HakkoAI’s AI PC pre-installation model continue?

Evidence: AI PC shipments are growing rapidly in 2025, and Intel, AMD, and Qualcomm are all pushing NPU adoption. Demand from hardware vendors looking for applications is real.

Conclusion: Sustainable in the short term, but long-term risk exists if operating systems build similar features in, such as Windows Copilot.

Hypothesis: Can a growth model without paid acquisition support a large enough user scale?

Evidence: One-third of users come from AI PC pre-installation, but overseas growth, mainly in North America, depends on organic growth and may have a lower ceiling. A scale of 2 million MAU is not large in the AI track.

Partial conclusion: The growth model proves that acquisition without burning money is possible, but the user-scale ceiling still needs observation.

5.5 Latest Media Reports

• Huxiu, October 2025: a deep interview with Liu Binxin covering product logic, user data, and business model.
• Jiemian News / Tencent News, August 2025: reported 1 million new users in 10 days after launch.
• Sohu, May 2026: summarized the latest product functions.
• Everyone Is a Product Manager, October 2025: analyzed the logic of “bullet-comment-style companionship.”
• GeekPark x TechCon, October 2025: covered the self-developed VLM model from a technical angle.

6. User Feedback and Industry Judgment

6.1 Real User Feedback

Google Play rating: 2.8/5.0 based on 1,290 reviews.

Core user pain points from Google Play reviews:

1. AI character dialogue is incomplete and cuts off halfway.
2. Users hope for free access to more model APIs such as DeepSeek and GPT.
3. Security controversy, including one user saying it temporarily locked their WeChat.

Developer response: the team acknowledged the dialogue-completion issue, said it is optimizing, and asked users to provide specific scenarios.

6.2 Industry Judgment

The AI companion track is undergoing a paradigm shift from “text dialogue” to “scenario embedding.” HakkoAI’s choice of gaming as an entry point is an early signal of this trend. It proves that AI companions do not need to compete for user attention; they can integrate into behaviors users already have.

But the core contradiction of this track is also clear: the more vertical the scenario, the lower the ceiling; the broader the scenario, the thinner the barrier. HakkoAI currently builds barriers through the uniqueness of the gaming scenario and AI PC partnerships. But if product experience cannot keep up, and a 2.8 rating is a hard wound, all growth data is borrowing from the future.

For the overall AI companion industry, HakkoAI’s biggest lesson may not be “make a game companion,” but rather: find something users are already doing, and let AI naturally integrate into it, instead of creating a completely new usage scenario that requires users to change habits.