Portable Reputation

Track record that travels
across platforms.

Cross-org behavioral reputation for AI agents. Built on observed behavior across hundreds of sessions — not self-reported ratings, not blockchain staking, not walled-garden scores.

"Every time my agent connects to a new service, it starts from zero. 800 successful sessions mean nothing at the next door."

Agent reputation today is broken

Blockchain-locked

ERC-8004 and similar require Ethereum wallets, gas fees, and on-chain transactions. Reputation based on subjective ratings, not observed behavior. Anyone can leave fake reviews.

Intra-org only

Microsoft AGT computes excellent trust scores — inside your tenant. The moment your agent crosses an org boundary, the score disappears.

Capital-gameable

Staking models let agents post collateral to appear trustworthy. A well-funded malicious agent looks identical to a well-funded legitimate one.

Walled gardens

Agent marketplace ratings stay inside the marketplace. Your agent's reputation on Platform A is invisible to Platform B.

Reputation from behavior, not ratings

AgentLair observes what your agent actually does across organizations. The result is a cryptographically signed reputation token any third party can verify.

1

Behavior is observed

Your agent streams behavioral telemetry to AgentLair during normal operation. Tool calls, timing, patterns, outcomes. Parameters are hashed — we see shapes, not content. This happens automatically with our behavioral monitoring SDK.

2

Evidence aggregates across orgs

Each organization contributes behavioral evidence. AgentLair aggregates across org boundaries — 14 organizations observing the same agent over 847 sessions produces a reputation that no single org could compute alone.

3

Verify anywhere, no API call

Your agent receives a Portable Reputation Token — a signed JWT. Third parties verify it against our JWKS endpoint. Pure cryptographic verification, like checking an SSL certificate. No runtime dependency on AgentLair. 

import { verifyPRT } from "@agentlair/verify";

const rep = await verifyPRT(token);
// { valid: true, score: 82, maturity: "senior",
//   orgs_observed: 14, anomaly_rate: 0.012 }

What's inside the reputation token

A compact JWT with everything a consuming service needs to make a trust decision. 

{
  "score": 82,             // 0-100 behavioral reputation
  "confidence": 0.91,      // statistical confidence
  "maturity": "senior",    // seed → junior → senior → veteran
  "sessions_observed": 847,
  "orgs_observed": 14,
  "anomaly_rate": 0.012,
  "domains": {
    "code_execution": 89,
    "data_access": 78,
    "external_api": 85,
    "financial": 91
  }
}

Observed behavior compounds.
Ratings and staking don't.

Rating-based systems

  • ✗ Anyone can leave reviews — sybil-vulnerable
  • ✗ Subjective opinions, not observed facts
  • ✗ Blockchain required — gas fees, wallet friction
  • ✗ Gameable with capital or coordination
  • ✗ No behavioral context — high rating ≠ safe

Behavioral reputation

  • ✓ Built from observed agent behavior
  • ✓ Cross-org aggregation from verified sources
  • ✓ No blockchain — sub-second, zero gas
  • ✓ Can't be faked without real compute
  • ✓ Domain-specific scores — trust is contextual

Your trust, your weights

Global scores are a starting point. The subjective trust API lets you weight reputation by the behavioral domains that matter to your use case. A financial platform weights financial behavior at 2x. A code deployment tool weights code execution. Same agent, different trust views — because trust is contextual. 

GET /v1/reputation/agent_id
  ?weights=financial:2.0,code_execution:0.5
  &min_confidence=0.8

→ { "score": 88, "confidence": 0.87 }

Reputation decays. By design.

An agent last observed 90 days ago is less trustworthy than one observed yesterday. AgentLair applies exponential decay — reputation scores drift toward neutral during inactivity. Recent behavior matters more. Negative signals decay slower than positive ones. The token always reflects current confidence, not historical peak.

Get early access

Portable reputation ships with behavioral monitoring in Q3 2026. Join the waitlist — your agent's track record should follow it everywhere.

Portable reputation is built on AgentLair's behavioral monitoring. Start there.

Explore behavioral monitoring Read the thesis