Tell us the data your model needs. We go get it.
Compound Labs sources proprietary training data directly from the companies and people who own it.
Codebases. Enterprise records. Expert workflows. Audio. Documents. Real-world environments. We negotiate the rights, build the dataset and deliver it ready for your training pipeline.
What can we actually get that you can't?
DATASET / 001AUDIO
10,000 hours
Spanish insurance calls
10,000 hours · Spanish · Insurance · Audio
- Proprietary source
- AI-training rights
- Transcribed + structured
DATASET / 002CODE
200
Production Rust repositories
200 repos · Full commit history · CI/CD
- Production code
- Commit history
- Issues + CI data
DATASET / 003AGENTS
RL environment
Hospital billing workflow
Real enterprise workflow · Agent interaction
- Real enterprise workflow
- Agent interaction
- Evaluation environment
What we source
Code
- Production repositories
- Commit histories
- CI/CD
- Issues
Enterprise
- ERP
- CRM
- Transactions
- Workflows
Human
- Expert demonstrations
- Calls
- Meetings
- Screen recordings
Documents
- Legal
- Medical
- Financial
- Technical
Agents
- RL environments
- Verifiers
- Real workflows
- Evaluations
On the desk now
ORDER / 01ACTIVE
Enterprise .NET codebases
40 production repos with years of history
ORDER / 02ACTIVE
Pharmacy distribution ERP
65M rows across 10 years, 250 stores and 15K SKUs
ORDER / 03ACTIVE
Production SaaS codebases
Full-stack platforms with real usage
From impossible dataset to delivered dataset.
Find
We identify who owns the data.
Acquire
We approach the owner and negotiate access and rights.
Transform
We extract, clean, structure, annotate or simulate the data as required.
Deliver
You receive the dataset and the relevant rights / provenance documentation.
Your name stays out of the deal.
We approach the owner and negotiate in our name. Nobody learns what you're buying, including the seller, until you decide to tell them.
DEAL RECORD
We've done this before.
exits between the founders
data deals negotiated
recruiting partners for doctors, lawyers and engineers
We hire and manage the specialists ourselves.
You pay when the data passes your review.
A small retainer starts the hunt.
If we can't find the data, that's all you pay, and you get a report of who we contacted.
Success fee
Due when the dataset passes your review. Both numbers are in the quote before you commit.
Start with a pilot: one narrow order, first samples in under 7 days.
Need data? / Have data?
FAQ
They sell labor and platforms they already run. We take a specific order and source it from wherever it sits, including companies and countries they don't cover. Many teams use both.
It comes directly from the companies and people who own it, under agreements that explicitly permit AI training. You can see the chain for every asset.
You pay a small retainer to start and a success fee when you accept the dataset. Both numbers are in the quote before you commit.
Start with a pilot. One narrow order is enough to see whether the data is good.
The retainer covers the work and there is no success fee. You get a report of who we contacted and why it didn't close.
Exclusive means exclusive, and it's written into the contract. If you choose non-exclusive, it costs less and the contract says who else can buy.
Licensing terms that explicitly permit AI training, negotiated per asset. Every asset ships with a data card and the rights document, with a paper trail back to the original owner.
You get the rights you ordered, and they are written into the contract. Exclusive orders are yours alone. Non-exclusive orders cost less and the contract says who else can buy.
We reply within 48 hours with a sourcing plan and first samples land in under 7 days. Full delivery depends on the spec, and the sourcing plan gives you a timeline before you commit.
We sign an NDA before you share the spec if you want one. Data is de-identified before it leaves the owner.
You can ask for a sample deliverable, which is an anonymized data card and the rights document for one asset.