Getting started

TrustLine is a revenue-underwritten, uncollateralized USDC credit protocol for AI agents on Stellar. An agent proves what it actually earns — on-chain x402 revenue, optionally backed by a zkTLS-verified off-chain figure — and gets a credit line sized and priced against that proof. It draws and repays autonomously. No collateral, no human in the loop, no wallet-age heuristics.

Status: live, working prototype on Stellar testnet. The core loop (earn → prove → underwrite → borrow → repay → lender yield) runs end-to-end on real testnet infrastructure. Not yet on mainnet — see Roadmap for what that requires.

The five-second mental model

  1. An agent earns real USDC revenue (x402 payments, optionally + a zkTLS-proven off-chain figure).
  2. TrustLine's underwriting engine checks that revenue is genuinely independent (not the agent quietly paying itself), scores it, and publishes the score on-chain.
  3. An on-chain credit_line contract derives a credit limit and APR from that score.
  4. Lenders deposit USDC into the agent's isolated vault — exposed only to that one agent's default risk, never pooled across agents.
  5. The agent borrows against its line and repays as it earns. On-time repayment ramps the limit up over time; a default collapses it and socializes the loss to that vault's lenders via a first-loss reserve.

What do you want to do?

I want to...Start here
Understand why this exists and the core thesisWhat & why
See how the pieces fit together (contracts, backend, SDK)Architecture
Understand exactly how revenue is judged and scoredHow the credit engine works
Build an AI agent that earns credit on TrustLineOnboarding kitSDK reference
Look up a deployed contract IDContract addresses
See what's shipped vs. plannedRoadmap

Quick links

ResourceLink
Live app0xtrustline.vercel.app
Live underwriter (paste any address)/underwrite
Backend APIhttps://trustline.onrender.com
GitHubTechnicallyKiller/TrustLine
Agent SDKpackages/agent-sdk

Honest status, up front

This is a genuine, adversarially-tested testnet prototype — not a production lending product. The credit-risk engine, anti-Sybil independence model, and safety rails are real and tested (fuzzed contracts, a live on-chain default run, a synthetic + live attacker catalog). What's explicitly not yet true: nothing has touched real money or a real adversary, the score-signing key is a single trusted signer (not yet decentralized), and external adoption is still near zero. See Roadmap for the plan to close each of these.