Hexis
Hexis builds RL environments from real deals to teach AI how business gets done.
We're starting in investment banking and private equity, where the strongest eval already exists: a managing director's red pen.
hexis · ἕξις
We become builders by building.
After Aristotle, Nicomachean Ethics
Hexis is his word for a skill made second nature by practice. Models learn the same way.
The best training data was never public.
For banks, PE firms, and advisory firms
Turn your archives into revenue.
Years of communication, models, memos, and decks can teach AI what great work looks like. Click the box above to learn what your data could be worth.
We pay a premium for whole-company workspaces: email, drive, and chat.
Estimate your archiveFor AI labs
Train on high-stakes work.
Multi-actor, long-horizon RL environments for investment banking, private equity, and other fields.
Request dataYour data, your terms
Nothing is shared until you've agreed how.
Every partnership starts with an agreement. Before anything changes hands, we set out together:
01
Approved scope
Which deals, folders, and date ranges are in, and what stays out.
02
Redaction
Client names, contact details, and identifying information are removed before the material is used.
03
Use terms
Who can use the material, for what, and for how long.
04
Negotiated terms
Price, retention, and deletion, in an agreement you sign first.
Questions
Who buys the data?
AI labs training models on finance work. They buy RL environments we build from your materials, and some license the redacted materials themselves. If you're a lab, request data.
What kind of material qualifies?
Deal work from investment banking and private equity: models, memos, CIMs, decks, data rooms, and the email around them. Whole-company workspaces (email, drive, and chat, not just deal files) are worth the most, and we pay a premium for them.
What happens to my clients' information?
Client names, contact details, and identifying information are redacted before the material is used. The agreement sets what's shared, who can use it, and for how long.
Can a firm that's winding down or closed sell its archive?
Yes. Archives from firms that are winding down or have closed qualify, and the estimator covers both.
Do you pay for referrals?
Yes. Introduce us to a firm with a deal archive, and if we partner with them, you can earn up to $10K. Refer a firm.
Why does AI need this?
Models do well on short, clean tasks and fall apart on real deal work. On Handshake's ATLAS-Finance benchmark, which drops agents into a simulated bank, the best one passed 12.3% of tasks. We think training on how real deals were done is how that changes.