Clipto uses AI to search terabytes of video and is now valued at $250M

Clipto, a three-year-old San Francisco startup, closed a $15 million all-equity funding round on August 31, 2026, valuing the company at $250 million post-money — a striking number for a 20-employee business that has never needed to build a data center. The pitch is deceptively simple: video libraries, meeting recordings, and photo archives are accumulating faster than anyone can manage manually, and finding one specific moment inside them remains a friction-heavy task even in 2026. Clipto indexes all of it on-device, and in mid-August it extended that index to AI agents via Model Context Protocol, making a user's local media searchable by Claude, ChatGPT, and tools built on either platform.

The Claim

Clipto's core product indexes video, audio, images, documents, and meeting recordings stored on a user's device — no cloud upload required. Users search using natural-language prompts: describe a speaker, a topic, or a visual scene, and the software retrieves the relevant segment. Founder Henry Kang told TechCrunch the product began with video creators who had years of footage scattered across computers and external hard drives, no reliable way to find specific clips, and no appetite for manual tagging. That original creator segment now accounts for only a quarter to a third of the company's users; the rest span lawyers, doctors, researchers, HR professionals, professors, and students — essentially any knowledge worker who generates more recordings than they can manually catalogue.

The $15 million round was all-equity, led by HSG — the firm formerly known as Sequoia China — with participation from GL Ventures, EnvisionX Capital, Palm Drive Capital, individual investors Hans Tung and Lu Zhang, and 522 Ventures. Clipto reports crossing 30 million users and generating more than $15 million in annual recurring revenue, with the company describing itself as profitable. Running those metrics through a 20-person headcount produces revenue-per-employee figures well above typical SaaS benchmarks, which at least partly explains the investor appetite at a 17× ARR multiple. Offices span San Francisco, Singapore, and Hong Kong.

The newest capability, and the one most relevant to developers, is MCP support. Clipto's implementation allows external AI tools — Claude and ChatGPT by name in the company's materials — to query the user's indexed media library. Multiple sources covering the announcement emphasize the same authorization design: access is user-scoped, per-request, and requires active approval. An AI agent cannot passively browse the index. The user must surface the request and explicitly authorize the lookup each time, keeping the agent's reach constrained to what the user consciously decides to share.

What We See

The MCP integration is the most technically significant part of this announcement, even if the valuation number gets the headline. What Clipto has shipped is a worked example of "bring your own data" at the tool-permission layer: the user decides what the AI can touch, on a per-query basis, rather than granting ambient, standing access to an entire file system. Our read is that this per-request authorization model is the right architecture for agent-connected personal data, and it deserves attention from any developer building MCP tools. The design question isn't just whether consent is required — it's at what granularity consent operates. Clipto's answer is per query, not per session or per integration.

Kang's background adds credibility that the company's age alone doesn't. Multiple reports indicate he spent close to 20 years approaching the same core problem from different angles — a PhD at Carnegie Mellon focused on robot memory systems, then ZenVideo, a video-creation startup that was acquired by Tencent before Clipto came together in 2023. The AI Daily Post makes the sharpest observation here: semantic video search isn't a product category that ChatGPT invented. It's an old, genuinely hard problem that became commercially tractable only when local compute — specifically Apple Silicon NPUs and the memory headroom to run capable multimodal models on-device — finally caught up to the ambition. Founders who stay on a problem long enough to be there when the underlying technology matures are a different kind of bet than first-movers chasing a hot market.

The cross-format, cross-ecosystem positioning also holds up under scrutiny. Adobe, Apple, and Google each address versions of this problem but remain anchored to their own storage and format silos. A single Clipto query can simultaneously return a video clip, a PDF, and a voice memo — a result set that none of the incumbents can assemble today without routing data through cloud infrastructure. That's a real, concrete differentiation.

Where It Falls Short

The "local-first, no cloud" framing requires a platform caveat. A separate source notes that on-device processing applies to the Mac version, which requires Apple Silicon and at least 16GB of RAM. The Windows version is still in beta and requires at least 12GB of RAM. The iOS, Android, and web versions are explicitly cloud-powered. That means a substantial portion of Clipto's 30 million users are processing data on Clipto's servers — the same arrangement as any conventional cloud service — not on their own hardware. Users attracted specifically by the privacy pitch should verify which version they're actually running before assuming the local guarantee applies to their device.

The valuation math also invites scrutiny that the announcement sidesteps. A $250 million post-money figure against $15 million ARR implies roughly a 17× multiple — aggressive even for a profitable, fast-growing AI company in 2026. The investor roster is serious, HSG carries genuine weight, and Hans Tung has backed durable consumer software companies before. But the multiple encodes assumptions about future growth and market expansion that Clipto hasn't made explicit. The most obvious competitive threat — Apple, Google, or Adobe shipping equivalent on-device search as a bundled system feature at zero marginal cost to the user — is real, not hypothetical, and none of the coverage from any source directly addresses what Clipto's moat looks like if that happens.

Sources