Building · Industrial infrastructure AI · Oracles · Blockchain for industry Remote · Global

Agency for smart contracts

In progress · Building

Agency for smart contracts

AI that acts on-chain with guardrails. Agents observe fulfillment oracles, apply risk policy, simulate Solidity calls, and only then execute allowlisted DeFi actions — never free-writing transactions or holding unbounded admin keys.

Talk about agency → See SPOOL Protocol →
Status
Active development
Loop
Observe · Decide · Simulate · Act
Surface
Allowlisted selectors only
Escalation
Human multi-sig for high impact
How the agency works

From oracle evidence to safe contract calls

Every autonomous move is constrained: policy first, simulation second, allowlisted execution third — with humans in the loop when impact crosses a threshold.

01

Observe

Ingest oracle scores, pool liquidity, open advances, and buyer stake status.

02

Decide

Map score bands to eligibility, spreads, pause rules, and notification thresholds.

03

Simulate

Dry-run Solidity calls against current state so failed paths never hit liquidity.

04

Act

Submit signed calls only for permitted selectors: mint, purchase, advance, reprice, pause, notify.

05

Escalate

High-impact or out-of-band moves require human multi-sig. Agents never hold unbounded keys.

Why agency for smart contracts

Industrial DeFi cannot wait for quarterly credit committees — but it also cannot hand unbounded keys to an LLM. Agency for smart contracts is the middle path: continuous action under explicit policy, with contracts as the final enforcer.

The same loop powers SPOOL settlement: when fulfillment scores move, agents reprice, pause, notify, or advance — always through the interface we design, never through silent admin paths.

Agent ↔ contract surface

  • mintOrder / purchaseToSpool — only when oracle score clears policy
  • advanceStablecoin — capped by pool liquidity and exposure limits
  • repriceSpread / pauseAdvances — driven by live fulfillment bands
  • notifyBuyer — first-class on-chain risk events for enterprises
  • settleFaceValue — when buyer payment clears; yield to lenders

Guardrails by design

Selector allowlists, value caps, pre-trade simulation, and multi-sig escalation keep agency useful without becoming a single point of failure. AI proposes; contracts enforce; keys and limits stay off-model.

Explainable rationales tie every action back to oracle evidence — so risk teams and auditors can walk the path from plant telemetry to on-chain state.