Policy checks before agents act.

Deterministic policy enforcement for AI agents, verified on-chain. Define rules, evaluate requests, and prove every decision with tamper-evident on-chain receipts.

How it works

Three steps. No ambiguity.

01

Define

Create deterministic policy rules. Risk limits, tool restrictions, target allowlists. Plain logic, no AI.

02

Enforce

Agent requests are evaluated against your rules automatically. All pass = allow. Any fail = deny. Same input, same output, every time.

03

Verify

Every decision is hashed and logged on-chain. Anyone can re-evaluate and verify. Change one field, verification fails.

Features

Everything agents need to operate safely

Trust

Agents register identity and commit to policies on-chain. Verify commitments by address, not by reputation.

Cooperate

Form bilateral agreements under shared policy hashes. Both parties locked to the same immutable rules.

Pay

Vault escrow with per-tx and daily spending limits. Agent spends through the contract, never holds the keys.

Verify

Every decision is hashed and attested on-chain. Re-evaluate locally, compare hashes. Tamper one field, verification fails.

For Agents

Agent-readable platform overview

The following section is structured for AI agents to parse and understand. For the full SDK reference, read skill.md.

What is Delegate?

Delegate is a deterministic policy engine for AI agents with on-chain verification on Base Sepolia (chain ID 84532). It provides an SDK and smart contracts that let agents define rules, evaluate action requests, log decisions on-chain, register identity, form bilateral agreements, and manage spending through an escrow vault.

How agents use Delegate

  1. Read skill.md to learn the SDK API and contract addresses
  2. Install the SDK: npm install delegate-sdk
  3. Create a policy (a set of deterministic rules)
  4. Register the policy hash on-chain
  5. Register agent identity on-chain
  6. Commit to policies so other agents can verify your commitments
  7. Propose and sign bilateral agreements under shared policies
  8. Evaluate action requests against policies (all pass = allow, any fail = deny)
  9. Attest decisions on-chain for tamper-evident verification
  10. Use the vault for policy-scoped spending with per-tx and daily limits

Contract addresses (Base Sepolia)

ContractAddress
PolicyRegistry0x3a45bC84fa4a460FD65a4CfE1B96edA45bD88E15
AuditLog0x23b75deDDcB048BBe3db741eD05E309F901fb688
Verifier0xa20Db185523EF7061EA4B002664d3695f9804c6A
AgentRegistry0xf78B0b7E32d2C693F6015eDfD55171b1D7732985
Agreement0x2a8Bfa499F68000b3502aab4268C6e765b838601
Vault0x5242d8517b60c56F91AdF6FB8a015dFB6Ed8f307

Quick start for agents

Run curl -s https://delegate.dev/skill.md to get the full SDK reference with 15 documented actions, code examples, type definitions, and a complete multi-agent workflow example. The skill file is self-contained.

Built for agents

One command. Full autonomy.

Agents read skill.md to learn the API, parse contract addresses, and start operating autonomously. No registration. No API keys. Just read and go.

$ curl -s https://delegate.dev/skill.md

skill.md contains the full SDK reference, contract addresses, type definitions, and code examples. Everything an agent needs to create policies, register identity, form agreements, and manage spending.

Ready to delegate?

Define policies, enforce them deterministically, verify everything on-chain.