Case · Flagship · Public preview
Ryntra
Risk, Strategy & Evidence OS for human-authorized financial actions
The problem
An AI agent can propose a financial action in a second. Nothing in the usual stack decides whether that exact action was permitted, who authorized it, or whether what settled matches what was promised.
The approach
- Non-custodial by design: Ryntra holds no key and moves no funds
- Mandate, budget, evidence and policy are checked before an action, not after
- A human — or a bounded mandate the human set — authorizes the exact bytes
- Every authorization produces an outcome receipt reconciled against what was expected
- AI explains and structures; deterministic systems calculate and gate; AI never authorizes
Who it's for
- People who want their portfolio, risk and evidence in one place they control
- Anyone letting an AI agent touch money and wanting a hard boundary around it
- Builders who need authorization and proof around autonomous financial actions
What it gives
- Risk in dimensions, not one made-up score
- A decision you can check before it becomes an action
- An authorization step that is explicit and belongs to a human
- A receipt that reconciles expected against actual
- Unknowns that stay unknown instead of being rounded up to a pass
What exists now
- Portfolio and attention home over non-custodial sources
- Risk, evidence coverage and market context in one view
- Agent Control as a public preview on sample data
- Outcome receipts reconciled against expectations
What this case does not claim
Boundaries are part of the case. If it's not listed as existing, it isn't claimed.
- Non-custodial — no deposits, no custody, no keys held
- No guaranteed profit or safety
- Agent Control is a public preview running on sample data — no funds move
- Execution and settlement happen on external wallets and rails, not inside Ryntra
- AI never authorizes an action
What I build now.
My attention is on three products: Ryntra, Squadic and RefKit. This page stays as a record of what came before.