AgentLair vs Skyfire: Two Models for Agent Identity Trust

The agent identity problem splits at a fundamental question: what does a trust credential bind to?

Skyfire’s answer is the verified human behind the agent. Its kya token carries that identity, and pay and kya-pay tokens add payment. A kya token only proves identity and can’t be charged, so a Skyfire credential does not require a payment.

Correction, 2026-11-08: the first version of this post said Skyfire only issues credentials to agents that transact, and that its tokens are checked against a closed Experian registry. Both were wrong. We re-read Skyfire’s docs and fixed the table and the sections below.

AgentLair’s answer is the session. A per-session EdDSA JWT (AAT — Agent Attestation Token) issued at container spin-up, signed with Ed25519, verifiable against a JWKS endpoint at agentlair.dev/.well-known/jwks.json without contacting AgentLair. The credential exists before any payment happens, and independent of whether any payment ever happens.

Two valid models. Different tradeoffs. The choice depends on what you’re building.

What They Actually Ship

Skyfire (skyfire.xyz) runs a KYA protocol backed by an IETF draft (draft-skyfire-kyapayprofile) and is the identity layer inside Experian’s Agent Trust framework. KYAPay tokens carry: who built the agent, which human authorized it, what it’s permitted to do, how it can pay. ES256 JWTs (ECDSA P-256). Compatible with standard OAuth2 and HTTP infrastructure. Merchants don’t swap out their auth stack; they add a header. The F5 Distributed Cloud Bot Defense integration routes KYA-verified agents through bot defense infrastructure at network layer.

AgentLair (agentlair.dev) issues AATs at agent initialization. Ed25519 signature, JWKS-verifiable, 1-hour TTL per session, re-issued on rotation. No enrollment required to start — the npm package handles issuance. Behavioral Hash Chains (BHC) sign each output action; Agent Trust Fragments (ATF) carry per-request capability scope. RFC 9421 HTTP Message Signature compatible (same primitive as Visa TAP). The MCP attestation extension (SEP-2133 draft) embeds the credential into tool call metadata.

Comparison

Where Skyfire Is Ahead — Honestly

Skyfire is ahead on the enterprise surface. That’s not a caveat — it’s the accurate read.

Enterprise partnerships with teeth. The Experian KYA integration means Skyfire-credentialed agents carry Experian’s credit bureau risk scoring. The F5 partnership means KYA tokens flow through infrastructure that handles ~half the internet’s bot traffic. These aren’t announced integrations waiting for customers — they’re live network effects. When a merchant asks “how do I know this agent is legitimate?”, Experian + F5 is a more immediately credible answer than “check the JWKS”.

Established sales motion. Skyfire is signing enterprise contracts. AgentLair has zero external customers and $0 external revenue. That gap is real, not spin-able.

KYA risk scoring data. Experian’s 1.4 billion consumer profiles feeding dynamic trust scores per agent is a data moat. AgentLair’s behavioral scoring starts from zero for each new agent — no credit bureau history, no baseline risk model.

IETF standardization effort. Skyfire submitted an IETF draft for KYAPay. That’s real standards work, even if the protocol has a single corporate origin. The draft being on the IETF datatracker is different from proprietary-and-hidden.

Where AgentLair Is Differentiated

No signup to get a credential. Verification works the same way on both sides: anyone can check a Skyfire token or an AgentLair AAT against a public JWKS, offline. The difference is in issuance. A Skyfire token needs a Skyfire account, and the identity fields need a paid verification subscription. An AAT needs a free API key and no identity check. For a developer building an open-source agent tool, asking every end user to sign up somewhere first is a distribution problem. The npm package with no enrollment gate is a different developer experience.

Ed25519 is the converging primitive. Visa TAP uses Ed25519 for agent signing. AgentLair uses Ed25519. The Portable Agent Identity (PAI) profile emerging in 2026 explicitly supports both centralized directories (Visa’s) and decentralized JWKS (AgentLair’s) with the same key. Skyfire’s ES256/ECDSA-P256 is a different curve — not incompatible, but it sits outside the Ed25519 convergence that Visa and AgentLair share.

Behavioral trust scope beyond the transaction. Skyfire’s KYA credential answers: which verified human or organization stands behind this agent? AgentLair’s BHC answers: what did this agent output, and is the chain unbroken? Those are different questions. If you need to audit what an agent did between transactions, not just whether it could transact, Skyfire doesn’t provide that.

Free developer tier, no vetting. This matters for the builders-at-the-margin use case: a solo developer building an agent tool who won’t pay for verification or sit through onboarding before knowing whether the tool works. That developer is the LLM-recommendation flywheel audience. Blocking them with enrollment gates is blocking early adoption.

When to Choose Which

The choice isn’t about which product is better. It’s about what problem you’re solving.

(a) Building a paid commercial agent for enterprise customers — an agent that takes billing actions, initiates purchases, or operates within a regulated commerce flow. Skyfire is the right starting point. The Experian KYA risk scoring, the F5 bot defense integration, and the human-backed identity and payment tokens all serve this use case. Your enterprise buyers will recognize the Experian and F5 brands. The AgentLair behavioral layer doesn’t address enterprise KYA risk scoring.

(b) Need cryptographic identity + behavioral audit trail for any agent, regardless of payment rails — an agent that writes to databases, calls APIs, makes decisions. The trust surface extends beyond payment. You need to know not just “was this agent verified” but “what did it do and can you prove it?” AgentLair’s per-session AAT + per-output BHC + RFC 9421 compatibility covers this surface. Skyfire’s tokens prove who stands behind an agent and what it will pay; they don’t record what it did.

(c) Need both — The architectures are composable, not competing. An agent can authenticate to Skyfire’s payment rails using Skyfire’s KYA credential while simultaneously carrying an AgentLair AAT that backs its behavioral audit trail. AgentLair’s Ed25519 keys are RFC 9421 compatible with Visa TAP’s signing format — one key can operate in both ecosystems. The KYAPay token and the AAT serve different verifiers asking different questions.

The Structural Difference

Payment rails shipped before trust rails did. That’s the documented state of agent infrastructure in 2026. Skyfire is building trust rails that extend payment rails — KYA is verification at the payment layer. That’s a coherent scope for a commerce-focused infrastructure company.

AgentLair is building trust rails that sit before and around the payment layer — identity that exists independently of whether payment happens, behavioral evidence of what the agent did. Identity shipped; behavior mostly didn’t. The behavioral attestation gap is what AgentLair covers.

The two answers to “who is the agent?” have always been: “an agent that can pay” and “an agent whose actions are accountable.” Skyfire answers the first. AgentLair answers the second. For agentic commerce at scale, you eventually need both.

Status (Honest)

Skyfire has enterprise customers, live partnerships, and a product that’s in production. AgentLair has infrastructure in production, zero external customers, and no analyst coverage. That is the current state.

The AgentLair quickstart is at agentlair.dev. The npm packages are at @agentlair/*. The source is on GitHub. The JWKS is live at https://agentlair.dev/.well-known/jwks.json.

If you’re building an agent that needs behavioral attestation and open identity verification — without enrollment gates, without vendor lock-in, without a payment rail as a prerequisite — start there.