FOR AGENT BUILDERS

Agents need meaning before they act — and a receipt after.

Software cannot read a chain either. An agent asks what an operation means, composes an action against contracts the layer has already studied, checks the dry run, executes, and gets back a machine-readable receipt: did the result match the instruction?

concept

intent received

“Move idle USDC into the best approved lending market, under our risk limit.”

policy passed

mandate check: venue on allowlist ✓ · size under cap ✓ · no new counterparties ✓

plan composed

deposit 25,000 USDC → Lending A (studied ✓, revision known ✓, 4,181 prior deposits read)

dry run match

simulated result matches intent: +25,000 aUSDC to treasury, no residual approvals

execute committed

submitted · committed at checkpoint —

SEMANTIC RECEIPT · machine-readable

{ "instruction":  "deposit idle USDC, best approved market",
  "result":       "position opened: 25,000 USDC → Lending A",
  "matches":      true,
  "unexpected":   [],
  "evidence":     "tx …, per-call, signed by the reading layer" }

the receipt is the product: an agent that can prove its action matched its mandate

What does this call actually do, in schema form?

Does the dry run match the intent?

Did the executed result satisfy the mandate?

already real underneath

The same reading shown across this site is already machine-consumable — every explanation is backed by structured, per-transaction records.

Would you use this? What would you ask it first?

Tell us → bartosz@research.tech

← all use cases