Kled V3 is a rebuilt version of the company's iOS app that pays people to capture photos, video, audio and text that AI labs need for training data. The main change in V3 is the home screen: the previous app was built around a single button for uploading photos already in a contributor's camera roll, while structured tasks sat in a separate tab most contributors rarely opened, which meant Kled "could only collect the data people already had, not the data labs actually needed." V3 replaces that screen with a task discovery feed where every live task is ranked per contributor, lab-priority tasks appear first, and the pay for each task is shown upfront, with a recommendation system that learns where each contributor's submissions get accepted by modality, task type and region.
Under the hood, tasks span image, video, audio, text and annotation, built from 108 templates covering camera, microphone, screen recording, touch and keyboard input, with each template carrying instructions plus examples of accepted and rejected submissions. Kled puts its network at more than 500,000 opt-in contributors across the US, Southeast Asia and Europe, collecting over 8 million datapoints a day across 38,000 structured datasets , and the company says this lets it build any dataset a lab can specify into the app and deploy it within 72 hours. The release also ships FD 1.0, a fraud detection model that runs on every upload and screens for duplicates, AI-generated images, content taken from the internet, edited files and behavioural fraud, scoring each submission for task relevance and quality before accepting it.
The launch matters because frontier labs are running out of easy web-scraped data and need structured, licensable collection from the physical world, with provenance attached. Co-founder and CEO Avi Patel positioned V3 as "the data collection layer for agi," and the app routes payouts to contributors through PayPal, Venmo, or a Solana wallet once they reach a $25 balance, typically within three to five business days, with earnings depending on a submission's uniqueness, diversity, quality and current buyer demand. For investors and operators tracking the data supply side of AI, Kled is worth watching as a bet that a consumer app and ranked task marketplace can out-collect traditional vendor pipelines.
Comments (11)
Sign in to join the discussion.
P
Priya Vatsa12d ago
"Solved data collection for AGI" is a sentence that will age like unrefrigerated milk. Bookmarking this tweet for the 2027 postmortem.
Z
Zara El-Khoury12d ago
Been watching this space from the fund side and the licensing angle is the only moat that matters right now. Everyone else is scraping and praying.
M
Michael Okafor12d ago
72 hours for ANY consumer dataset is a wild claim. What happens when I ask for 500k hours of people assembling IKEA furniture in their underwear, asking for a friend.
G
Greg van Leer12d ago
Is this basically Mechanical Turk with better branding? Genuinely trying to understand what makes 2024 me care when 2019 me already saw this movie.
L
Linnea Hallberg12d ago
The jump cut at 0:14 where the voiceover says "physical reality" and the drone shot kicks in. Chef's kiss editing, whoever cut this earned their invoice.
T
Takashi R.12d ago
Tagline rewrite, no charge: "The world's camera roll, licensed." You can Venmo me later.
D
Dev Ganapathy12d ago
Need to know if that live collection demo was pre-recorded or if there's an actual orchestration layer behind it. Thread doesn't say and the suspense is killing me.
B
benny12d ago
Reporter here, would love 10 min to understand how the collector network actually gets compensated. DMs open, promise not to be weird about it.
R
Ritu Shenoy12d ago
Genuine question from someone new to this: how do you verify the people submitting data actually have the rights to license it? Like what stops me from filming my neighbor's dog.
K
Kosta P.12d ago
How many people are on this team? Because if it's under 15 I'm going to need to lie down.
A
Amara Okonkwo12d ago
Building adjacent in synthetic data and honestly the real-world capture angle makes our pitch harder this week. Respect, mildly annoyed, mostly respect.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Kled
Sourcing the largest licensable datasets.
Sourcing the largest licensable datasets.
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Kled V3 is a rebuilt version of the company's iOS app that pays people to capture photos, video, audio and text that AI labs need for training data. The main change in V3 is the home screen: the previous app was built around a single button for uploading photos already in a contributor's camera roll, while structured tasks sat in a separate tab most contributors rarely opened, which meant Kled "could only collect the data people already had, not the data labs actually needed." V3 replaces that screen with a task discovery feed where every live task is ranked per contributor, lab-priority tasks appear first, and the pay for each task is shown upfront, with a recommendation system that learns where each contributor's submissions get accepted by modality, task type and region. Under the hood, tasks span image, video, audio, text and annotation, built from 108 templates covering camera, microphone, screen recording, touch and keyboard input, with each template carrying instructions plus examples of accepted and rejected submissions. Kled puts its network at more than 500,000 opt-in contributors across the US, Southeast Asia and Europe, collecting over 8 million datapoints a day across 38,000 structured datasets , and the company says this lets it build any dataset a lab can specify into the app and deploy it within 72 hours. The release also ships FD 1.0, a fraud detection model that runs on every upload and screens for duplicates, AI-generated images, content taken from the internet, edited files and behavioural fraud, scoring each submission for task relevance and quality before accepting it. The launch matters because frontier labs are running out of easy web-scraped data and need structured, licensable collection from the physical world, with provenance attached. Co-founder and CEO Avi Patel positioned V3 as "the data collection layer for agi," and the app routes payouts to contributors through PayPal, Venmo, or a Solana wallet once they reach a $25 balance, typically within three to five business days, with earnings depending on a submission's uniqueness, diversity, quality and current buyer demand. For investors and operators tracking the data supply side of AI, Kled is worth watching as a bet that a consumer app and ranked task marketplace can out-collect traditional vendor pipelines.
Comments (11)
"Solved data collection for AGI" is a sentence that will age like unrefrigerated milk. Bookmarking this tweet for the 2027 postmortem.
Been watching this space from the fund side and the licensing angle is the only moat that matters right now. Everyone else is scraping and praying.
72 hours for ANY consumer dataset is a wild claim. What happens when I ask for 500k hours of people assembling IKEA furniture in their underwear, asking for a friend.
Is this basically Mechanical Turk with better branding? Genuinely trying to understand what makes 2024 me care when 2019 me already saw this movie.
The jump cut at 0:14 where the voiceover says "physical reality" and the drone shot kicks in. Chef's kiss editing, whoever cut this earned their invoice.
Tagline rewrite, no charge: "The world's camera roll, licensed." You can Venmo me later.
Need to know if that live collection demo was pre-recorded or if there's an actual orchestration layer behind it. Thread doesn't say and the suspense is killing me.
Reporter here, would love 10 min to understand how the collector network actually gets compensated. DMs open, promise not to be weird about it.
Genuine question from someone new to this: how do you verify the people submitting data actually have the rights to license it? Like what stops me from filming my neighbor's dog.
How many people are on this team? Because if it's under 15 I'm going to need to lie down.
Building adjacent in synthetic data and honestly the real-world capture angle makes our pitch harder this week. Respect, mildly annoyed, mostly respect.