The archive · AI & Models · Product decision · 2023–2026
Clipto bets AI needs a local memory layer; #1 on Product Hunt
Founder who sold his text-to-video startup to Tencent builds fully local AI memory for TBs of personal media — $15M ARR, profitable, valued at $250M.
Clipto
What the business is
Clipto is a privacy-first, on-device AI memory platform: it indexes videos, audio, images and documents locally, then lets users search and query them in natural language — with an MCP so AI agents can access the library too.
Starting capital：$15M all-equity round at $250M post-money (Aug 2026) from HSG (ex-Sequoia China), GL Ventures, EnvisionX Capital, Palm Drive Capital, Hans Tung, Lu Zhang and 522 Ventures; preceded by Pre-A, Pre-A+ and Pre-A++ rounds since July 2025 (total ~$16.5M per PitchBook)
How it started
Clipto launched in 2023, founded by Henry Kang, a Carnegie Mellon robotics and computer-vision PhD who interned at Microsoft Research Asia in 2004, founded text-to-video platform Zhiying (ZenVideo) in 2017, and led Tencent's AIGC products after Tencent acquired it in 2020. His thesis: AI built world models but no user models — every agent conversation starts from zero, because personal data sits untapped on local devices. 'A smart person, without memory, is just an amnesiac.'
What happened
Clipto hit break-even within three months and topped $10M first-year recurring revenue; by early 2026 it passed $15M ARR and net profitability with about 20 employees and small teams in Hong Kong and Singapore. Back-to-back Pre-A rounds from July 2025 pushed its valuation above $250M by January 2026. In late May 2026 its Mac launch took #1 Product of the Day on Product Hunt. In August 2026 it disclosed the $15M round at $250M post-money, launched Windows and Android apps plus an MCP for AI agents, and reported 30M+ users across 150+ countries with hundreds of thousands of paying subscribers.
How it ended up
Still running and scaling: profitable at $15M+ ARR, expanding platforms (Windows, Android, MCP agent access) and positioning memory as an infrastructure layer; 30M+ users as of August 2026.
Background
Clipto, founded in 2023 by Henry Kang (康洪文), is a privacy-first on-device AI memory platform. Kang — a Carnegie Mellon robotics PhD who led Tencent's AIGC products after Tencent acquired his text-to-video startup ZenVideo (Zhiying) in 2020 — argues that AI built world models but no user models: agents are smart but amnesiac, because personal data sits unindexed on local devices.
Clipto indexes videos, audio, images and documents entirely on-device using more than ten proprietary small models, then answers natural-language queries — 'the drone shot of vehicles crossing the desert' — in seconds, with no files leaving the machine. The bet is that memory, not generation, is AI's next infrastructure layer, and that local-first processing plus the rise of Apple M-series and NVIDIA AI PCs makes it viable.
The company hit break-even within three months, passed $15M ARR and net profitability by early 2026, and in late May 2026 its Mac launch took #1 Product of the Day on Product Hunt. In August 2026 it disclosed a $15M round at a $250M post-money valuation from HSG (ex-Sequoia China), GL Ventures, EnvisionX, Palm Drive and others, launched Windows/Android apps and an MCP for AI agents, and reported 30M+ users across 150+ countries with hundreds of thousands of paying subscribers.
What has to be true
- Content generation exploded while retrieval stayed manual: editors scrub timelines and lawyers re-read depositions, so organizing existing media became the real pain point.
- The user-model gap is structural: agents reason on world knowledge but can't see a person's history, so a persistent local memory layer is needed for agents to be useful.
- Hardware shifted — Apple M-series and NVIDIA AI PCs put enough compute on-device to index terabytes locally at near-zero marginal cost versus expensive cloud tokens.
- Privacy and latency make local processing a requirement for legal, healthcare and enterprise data, not just a preference.
- Kang's two decades on video understanding and generation meant he saw the content-organization bottleneck before it was fashionable — and had shipped products through two exits before.
What can be applied
When content becomes abundant, the scarce layer shifts from generation to organization: turning private local data into agent-readable memory builds a moat no model upgrade can displace.
Aftermath
As of 2026-08-31, Clipto is profitable and scaling: 30M+ users across 150+ countries, hundreds of thousands of paying subscribers, ~$15M ARR, and ~20 employees with teams in Hong Kong. It disclosed a $15M all-equity round at a $250M post-money valuation from HSG, GL Ventures, EnvisionX Capital, Palm Drive Capital, Hans Tung, Lu Zhang and 522 Ventures, and launched Windows and Android apps plus an MCP for AI agents to search private libraries. The company frames search as the first application of memory and aims to become an AI-native memory layer shared by agents; no IPO announced.
Sources
- 斩获 PH 全球榜首｜Clipto.AI：端侧多模态架构，补齐 AI 核心记忆层
- 当所有AI都在理解世界，这家公司在理解你
- Clipto Raises $15M to Build the Memory Layer for the AI Era
- Clipto's $250 Million Bet on Local AI Memory as Content Overload Hits Knowledge Workers
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card