Data lab · Buyer · Broker
We buy the record of how real work gets done.
Anthroly is a data lab. We license operational data from US companies (the threads, tickets, calls and decisions behind a working business) and supply it to frontier AI labs as training corpora, RL environments and evaluations.
$900K+
in data bought, sold and brokered
100%
of supply sourced from US companies
Dana Whitfield9:14
Carrier says the pallet for Harlow Supply is stuck in Memphis. They want to reroute through Atlanta, adds two days.
Marcus Oyelaran9:16
Don't reroute. Their contract has a late penalty after Thursday. Split the order: ship the 40 units we have from Reno today, backfill the rest.
Dana Whitfield9:21
Done. Calling (901) 555-0148 to confirm the partial. Logging the exception on PO #88213.
- Source
- Logistics · US
- Signal
- Decision under constraint
- Kept
- Roles, timing, reasoning
Who buys from us
We sell to every major AI lab.
Data we've licensed sits in the training and evaluation pipelines of the labs below. Lab-specific terms stay confidential, as they should.
OpenAI
Anthropic
Google DeepMind
Meta
xAI
Microsoft
Amazon
NVIDIA
Mistral AI
Cohere
NewExpert human data
Expert human data for frontier AI.
Verified engineers, scientists, doctors and Indic-language specialists, ready in days.
- Software and coding
- Math and science
- Medicine
- Indian languages
- expert
- EXP-0417
- field
- Internal medicine · MD
- languages
- Hindi · Marathi · English
- identityGovernment ID verified
- credentialsDegree and licence confirmed
- skills testPaid task · passed senior review
- agreementsNDA · IP assignment signed
- accessManaged account · two-factor on
- calibrationGold-answer set matched
- verified experts on the bench
- 500verified experts on the bench
- hold a PhD
- 100hold a PhD
- to a staffed pilot team
- 10 daysto a staffed pilot team
- experts deployed within a week of scale-up
- 100experts deployed within a week of scale-up
01
Vetted
ID and credential checks, then a paid skills test in their own field.
02
Reviewed
Gold answers, two review layers, agreement scores and a strict no-AI-tools rule.
03
Secured
NDA and IP assignment, managed accounts, two-factor login, least access.
Start with a free or low-cost 1–2 week pilot. Scale to a dedicated team once the data proves itself.
How we vet, review and secure the workOne market, two sides
01What we buy
Your operating history is an asset. Most companies have never priced it.
We license the working record a business produces just by running. Not the polished output, but the back-and-forth that produced it. The more reasoning a record shows, the more it is worth.
Team chat
Slack, Microsoft Teams, Google Chat
Where decisions get argued out in real time, with the tradeoffs still visible.
Gmail, Outlook, shared inboxes
Long-running negotiations, approvals and handoffs between people and companies.
Support and service tickets
Zendesk, Intercom, Freshdesk, ServiceNow
Problem, diagnosis, resolution. The most naturally structured reasoning a business produces.
CRM history
Salesforce, HubSpot, Pipedrive
Multi-month deal cycles with notes on why things moved, stalled or died.
Calls and meetings
Gong, Zoom, Fireflies, call recordings
Spoken reasoning, objection handling and real conversational turn-taking.
Code and engineering
GitHub, GitLab, Jira, Linear
Pull request reviews, incident threads and the history of why code changed.
Documents and wikis
Google Drive, SharePoint, Notion, Confluence
Drafts and revisions that show how a finished document was argued into shape.
Finance and operations
NetSuite, QuickBooks, SAP, ERP exports
Structured records that ground the conversations in what actually happened.
Field and frontline records
Work orders, inspection notes, dispatch logs
Physical-world work that almost never makes it onto the internet.
Design and creative files
Figma, Adobe, version histories
Iteration with feedback attached, which is rare and very hard to fake.
We only buy from companies headquartered and operating in the United States. Data you hold on behalf of your own clients stays out of scope unless you have the right to license it.
“Finished work hides the reasoning. The record of how it got finished is the part the internet never saw, and the part models are missing.”
02For AI labs
Real work, structured for reinforcement learning.
Agents are now judged on whether they can do the job, not whether they can talk about it. That takes training signal from real workflows. We source it from US businesses and build it into the formats your post-training team already uses.
- environment
- ar_escalation.v3
- origin
- licensed operating record · US
- tools
- email · crm · ledger · calendar
- horizon
- multi-day, 40–120 steps
- reward
- verifiable: invoice resolved, policy kept
- exposure
- never published · zero contamination
RL environments
Sandboxed replicas of real workflows, built from licensed records, with tools, state and verifiable reward signals. Agents learn on work that actually happened.
Agent trajectories
Long-horizon, multi-tool sequences reconstructed from how people really completed tasks: days of context, interruptions, handoffs and recovery from mistakes.
Held-out evaluations
Private test sets drawn from records that have never been public, so a model cannot have memorized them. Scored against what the professional actually did.
Preference and SFT data
Expert demonstrations and pairwise judgments grounded in a domain, written by people who have done the job rather than by generalist raters.
Licensed corpora
De-identified enterprise text, audio and structured exports for mid-training and domain adaptation, with a clean chain of title for every record.
Custom sourcing
Tell us the capability gap. We find the US companies whose records close it, negotiate the license and deliver to your spec.
Tell us where your model breaks. We'll find the companies whose records fix it.
Request a sample pack03How a deal works
Weeks, not quarters. One counterparty the whole way.
Going direct to a frontier lab can mean a procurement cycle measured in quarters. Selling to a data lab that already supplies them collapses that, and keeps you dealing with one team from first call to final payment.
- 01
Intake
Tell us which systems you run and how far back the record goes. We ask for counts and date ranges, not files. Nothing leaves your systems.
- 02
Sample review
A small, de-identified sample is read by a person on our team. We never price a record we have not seen.
- 03
Offer and license
You get a firm offer and a bounded license: defined material, defined term, defined use. You keep ownership and keep operating as before.
- 04
Scrub and transfer
Names, contact details, account numbers and other identifiers are removed before anything moves. Transfer runs over encrypted channels.
- 05
Get paid
Payment follows delivery and approval. Money moves one way, from us to you. No listing, no auction, no fee to you.
- Logistics
- Healthcare administration
- Insurance
- Construction
- Legal operations
- Accounting
- Software
- Manufacturing
- Field services
- Real estate
- Financial services
- Retail
- Hospitality
- Staffing
- Energy
- Logistics
- Healthcare administration
- Insurance
- Construction
- Legal operations
- Accounting
- Software
- Manufacturing
- Field services
- Real estate
- Financial services
- Retail
- Hospitality
- Staffing
- Energy
04Safeguards
Built to be the buyer you'd trust with your own records.
Sourcing policy
We buy only from companies in the United States.
No exceptions. One legal framework means a clean, auditable chain of title for every record, and American operating data is what the labs ask us for most. If your company is headquartered outside the US, we're not the right buyer, and we'll tell you so on the first call.
Licensed, not sold
A license grants defined rights over defined material for a defined term. You keep the business, the data and the right to license it again.
Identifiers out first
Names, emails, phone numbers, account numbers and addresses are removed or replaced before anyone downstream sees a record. Roles, timing and the work itself stay.
Chain of title, verified
We confirm you have the right to license what you are licensing. Data you hold on behalf of your own clients stays out unless your contracts allow it.
Confidential on both sides
Sellers are never named to buyers. Buyer terms are never shared with sellers. Nothing is listed, auctioned or made public.
Bounded use
Every license specifies how the data may be used. Training and evaluation, yes. Re-identification, resale outside the license or targeting your customers, never.
Small circle of access
Only a handful of named people at Anthroly ever see a company identity alongside its data. Everyone else works with de-identified records.
05Questions
Asked on almost every first call.
Something else on your mind? Write to sharatth@anthroly.com.
Does Anthroly buy data, or broker it?
Both. Most of the time we are the buyer: we license the data directly from the company that produced it, pay for it ourselves and supply the AI labs. For some lab requests we broker a direct license between a US company and a lab, and we stay in the middle to run diligence, anonymization and delivery. Either way, you deal with one team.
Who do you sell to?
Frontier AI labs and the model teams inside large technology companies. We sell to every major lab. We do not disclose which lab receives which record, the same way we never disclose one seller to another.
Why only US companies?
One legal framework gives every record a clean, auditable chain of title, and US operating data is what the labs ask us for most. We do not buy from companies headquartered outside the United States.
What does it cost my company?
Nothing. There is no listing fee, no commission charged to you and no auction. Money moves one way, from us to you.
How much is our data worth?
Every record is priced individually, after we have read a de-identified sample. Depth of operating history, how much visible reasoning the record contains and how scarce the domain is move the number far more than raw volume does.
Is the data anonymized before anyone sees it?
Yes. Names, emails, phone numbers, account numbers, addresses and other identifiers are removed or replaced before anything moves. Roles, timestamps and the work itself are kept, because that is where the value is.
Do we lose ownership of our data?
No. A license is not a sale. You keep the business, the data and the systems, and keep using all of it exactly as before. The license covers defined material, for a defined term and a defined use.
What kinds of data do you buy?
The working record of the business: team chat, email, support tickets, CRM history, call recordings, code review and incident history, documents and wikis, ERP and accounting exports, and field records. The messier and more reasoning-rich, the better.
We are an AI lab. What formats can you deliver?
De-identified raw corpora, RL environments with tools and verifiable rewards, long-horizon agent trajectories, SFT and preference data, and private held-out evaluation sets. Everything ships with provenance and a clean license.
How do we get started?
Email us. Tell us your company, which systems you use and roughly how far back they go. Counts and dates are enough, so please do not send files. We read every message and reply personally.
Field notes
Plain answers about a young market.
One email to start
Find out what your company's record is worth.
Write to the people who make the offer. No fee, no commitment, and nothing leaves your systems.
or write to jaya@anthroly.com