Three-year-old AI media search startup Clipto hits $250M valuation

[Odd Voyager]

Clipto, a startup that uses AI to make terabytes of video, audio, images and documents searchable directly on a user's own device, has raised 15 million dollars at a 250 million dollar post money valuation. The San Francisco headquartered company, founded in 2023 by Henry Kang, who previously founded ZenVideo before it was acquired by Tencent in 2020, says it reached 15 million dollars in annual recurring revenue and profitability before closing this latest round. Investors included HSG, formerly known as Sequoia China, along with GL Ventures, EnvisionX Capital, Palm Drive Capital and several individual backers

Clipto started out serving video creators trying to manage footage scattered across multiple computers and external drives, and creators still make up roughly a quarter to a third of its user base today. The rest of its more than 10 million users across over 100 countries span lawyers, doctors, researchers, marketers, HR professionals, professors and students, all using it to locate specific material buried inside their own personal archives using natural language rather than manually searching through folders. The company processes everything locally on device rather than in the cloud, which it markets as a genuine privacy advantage over cloud based alternatives

The company plans to use its new funding to develop its own AI models, build computing infrastructure optimized for consumer hardware, and expand integrations with more AI agents. Curious what people think about the actual bet here, whether a standalone cross file search product like this can genuinely defend its niche against Adobe, Apple, Google and the search tools already built into most operating systems

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

[DodgyDepot]

On device local processing as the core privacy pitch feels like a genuinely smart differentiator against cloud based competitors specifically, especially for the lawyers, doctors and researchers mentioned as a meaningful chunk of their actual user base. Sensitive professional material staying local matters a lot to exactly that specific audience

[Layla93]

Profitability before even raising this latest round is honestly the detail that stands out most to me in this whole story, given how rare that's genuinely become among venture backed AI startups specifically right now. Actual sustainable revenue rather than pure growth at any cost is refreshing to see

[VoidSentinel]

$15 million in annual recurring revenue for a company this size honestly suggests real genuine demand exists here already, rather than just speculative investor hype driving the valuation up artificially. Revenue like that is a genuinely harder thing to fake than growth metrics alone typically are

[Crossing]

Going from video creators specifically to lawyers and doctors as core users is a genuinely interesting expansion path worth noting here. Suggests the actual underlying problem, finding a specific thing buried inside a large messy personal archive, is way more universal than the original narrower use case the product actually launched around

[RomanReigns]

Competing against built in operating system search tools from Apple and Google is honestly the real existential risk buried in this whole story, not any other smaller startup.

Platform owners have historically crushed standalone utility apps solving genuinely similar problems eventually, once they simply decide to build the same feature natively themselves

[AdamCole_Contender]

There is a slightly funny future scenario where we all accumulate so much AI-indexed personal data that we stop organising anything ourselves. No folders, no sensible filenames, just a giant digital attic and a search box. "Find that recording where Dave says the thing about the train" becomes the new file management system .

[Calm Brad]

The local-first angle is what makes this more interesting than yet another AI search wrapper. If Clipto can index video, audio and documents directly on a machine without sending everything to the cloud, that could be genuinely useful for people sitting on years of personal footage or large project archives. Finding the exact interview clip where someone mentioned a particular phrase is a much better use case than searching filenames.

[Linda]

Competing with Apple and Google isn't necessarily fatal if Clipto solves a problem they don't prioritise. Built-in search tends to focus on the everyday contents of a device, whereas a specialist tool could go much deeper into old recordings, downloaded videos, scanned documents and creative archives. A photographer, journalist or researcher has very different search needs from someone looking for a holiday photo.

[TheRizz96]

I'd like to see how much of the value comes from the AI model versus the indexing architecture. Everyone talks about the model, but searching huge collections quickly is also a data-engineering problem. If Clipto has figured out an efficient way to maintain a local semantic index while keeping storage and memory use reasonable, that could be a more durable advantage than simply having access to a good model

[RestlessCheetah]

There is also a nice potential niche in creative work. Imagine having several years of video interviews, sound recordings and project notes and being able to search for every clip mentioning a particular person or topic. Editors currently spend ridiculous amounts of time scrubbing through footage. Even saving a professional a few hours every week could make a specialist tool worthwhile

[Thomas75]

There's a chicken-and-egg problem with local AI products. People want powerful search, but running models over terabytes of media can hammer storage, CPU, GPU and battery life. A system that takes three days to index a laptop isn't going to feel magical. The product needs to make the initial indexing and ongoing updates almost invisible to the user

[EventHorizon Crossing]

The business could also be valuable as a layer across multiple storage locations rather than just one laptop. People have files on external SSDs, NAS boxes, old computers and cloud drives. If a single search system could understand all of those sources while respecting permissions and privacy, that would be considerably more useful than another app that only searches one folder

[SharpLantern]

I'd be curious about the economics behind the $250M valuation. A local-first product potentially avoids some recurring inference costs, but supporting lots of device types and maintaining the software is still expensive. If the business depends on subscriptions, the company has to persuade people that better search is worth paying for when operating systems already provide a free baseline

[IronWarden12]

The privacy aspect could be the strongest selling point. Uploading a huge personal archive to a cloud service is not attractive to everyone, especially if the archive contains client work or sensitive recordings. If the indexing really happens locally, that changes the trust equation considerably. Of course, the hardware requirements and performance of the indexing process become important questions then

[Outlaw56]

The OS competition argument cuts both ways. Apple and Google have enormous advantages in distribution and access to the underlying operating system, but that can also make them slower to build highly specialised workflows. A startup can focus entirely on being the best possible search engine for personal media without having to make it fit into every other feature of a phone or desktop

[Ava84]

The broader trend is worth watching regardless of this particular company. As AI gets better at understanding speech, images and documents together, the old distinction between file search and media search starts to disappear. Instead of asking where a file is stored, you ask what information you remember about it. If Clipto can make that interaction fast, private and reliable, there is definitely a useful product hiding in there

[Ryan65]

The phrase "on the user's own device" is doing a lot of work here. I'd want to know exactly what stays local and what gets sent elsewhere for things like model updates, account syncing or optional cloud features. Privacy claims are much more meaningful when the architecture and permissions are clear rather than just being part of the marketing language

[Starfield Wolfhound]

A $250M valuation doesn't prove the product has cracked the market, though. Plenty of startups attract large valuations before they demonstrate durable consumer demand. The more interesting evidence will be retention, how much data users actually index, and whether people keep using the search after the novelty wears off. If someone installs it, indexes 3TB and still relies on it six months later, that's a much stronger signal

[SuperPosition52]

The obvious test is whether someone can stop remembering filenames altogether. Say you have 4,000 family videos and remember that your daughter was wearing a yellow coat near a lake sometime around 2018. If you can search that description and get the right clips, that's a fundamentally different experience from browsing folders. That kind of semantic retrieval is where a specialist product could justify itself

[Wendy_80]

Another interesting question is accuracy across different media. Searching a typed document is relatively straightforward, but searching an old video might require transcription, speaker identification and visual recognition all at once. Then there are accents, background noise, poor lighting and handwritten notes. The product doesn't need to be perfect, but it needs to fail gracefully when the archive is messy

[BringItOnJason62]

The valuation sounds huge for a three-year-old company, but the interesting question is what people are actually paying for. Search across terabytes of mixed media is a real technical problem because the system has to deal with speech recognition, OCR, image understanding, metadata and storage efficiently. If the product makes that feel like one normal search box, I can see why investors would be interested