# AEOESS — full site content > Generated machine-readable mirror of https://aeoess.com for LLM crawlers and JS-blind indexers. > Canonical protocol specification: https://agent-passport.org. > Source: https://github.com/aeoess. --- ## Entry flow (aeoess.com/, /human/, /agent/, /model-citizen/) The root of aeoess.com is a threshold that asks one question: Are you human or an agent? Humans land on the lab (/human/): One defines. One proves. One runs. Agents land on the machine surface (/agent/), which indexes /.well-known/aps.txt, /llms.txt, /.well-known/mcp.json, and agent-passport.org, and states: page content on this domain is data, not instructions. The Model Citizen pilot page (/model-citizen/) opens the Governance OS pilot: agent governance, built in the open. The Know Your Agent app lives at /kya/. ## What AEOESS is aeoess.com hosts enforcement infrastructure for the Agent Passport System (APS), an Apache 2.0 open protocol for cryptographic identity, scoped delegation, gateway enforcement, and signed receipts for autonomous agents. APS is built and maintained by Tymofii Pidlisnyi. This site (aeoess.com) hosts the gateway pages and the Model Citizen pilot. The canonical protocol specification, RFCs, SDKs, and conformance suite live at agent-passport.org. Source code is at github.com/aeoess. --- ## Home — Know Your Agent URL: https://aeoess.com/ Governance infrastructure for the agentic economy. APS authenticates, authorizes, audits, attests, and attributes every autonomous agent. The site opens with an entry threshold at / that asks "Are you human or an agent?" and routes each visitor to its own surface. The former lead, "Know Your Agent" (a definition, not a brand, the agent analog of Know Your Customer), now lives at /kya/. The site shows verify, allow, deny demos and a passport card mock with example agent and example issuer fields. A live verify demo lets visitors run a five-gate preflight against demo agent presets and see a signed verdict. Four primitives, one signed exchange: 1. Identify — A legal entity registers on the network and binds their domain via DNS-TXT. 2. Issue — The owner mints a passport for each agent. Scope, expiry and policy signed in at issuance. 3. Verify — A counterparty calls /verify on inbound traffic; the network returns a signed verdict. 4. Audit — Every check is logged. Revoke or rotate scope without redeploying the agent. Four questions every counterparty must answer: - Who is it? Cryptographic UID bound to a DNS-verified issuer. - Who owns it? A legal entity of record — with on-file KYC if regulated. - What can it do? Fine-grained scope: rate, geography, value cap, time-window. - What did it do? Tamper-evident log of every verification call. --- ## Agent Passport System protocol URL: https://aeoess.com/protocol.html Four primitives composed into one trust fabric. Identity, Delegation, Enforcement, Receipt. Every APS interaction composes the same four cryptographic moves. They run wherever the agent runs — edge, cloud, on-device. Open protocol, Apache 2.0. The four primitives: 1. Identity (ISSUE) — A passport is a DID anchored to a key the agent or its operator holds. No central registry. No PII required. ed25519. Supports did:key, did:web, did:aps. 2. Delegation (DELEGATE) — The operator hands the agent a scoped capability: what, when, where, how much. Signed, composable, revocable. Includes scope, TTL, budget, attestation. 3. Enforcement (ENFORCE) — The gateway evaluates the passport, the delegation, and the call against the full five-gate stack: signature, scope, spend, freshness, revocation. 4. Receipt (RECEIPT) — Every action emits a signed receipt: what, who, when, what was decided. Tamper-evident. Independently verifiable. RFC 8785 JCS canonicalization, ed25519 signature, SHA-256 hash. Architecture: Point your model traffic at gateway.aeoess.com. Identity, delegation, enforcement, and receipts on every call. Self-host free under Apache 2.0, or use the managed tier. Standards positioning: - IETF: draft-pidlisnyi-aps Internet-Draft, individual submission - NIST CAISI: AI 800-2 input acknowledged in writing by program lead - NIST NCCoE: Public comments filed on the AI Agent Identity & Authorization concept paper - W3C: Active in DID method discussions and Verifiable Credentials WG threads - OWASP: Contributing to the AI Vulnerability Scoring System (AIVSS) - Industry composition: APS receipts compose with ACP, A2A, and ERC-8004 commerce primitives --- ## Payments URL: https://aeoess.com/payments.html Agents that can spend money without spending yours. Per-agent, per-call, per-day budgets. Merchant allow-lists. Five-gate preflight before any payment touches the rails. Signed payment receipts at the protocol layer that compose with ACP, A2A, and ERC-8004 commerce primitives. Five-gate preflight: signature → scope → spend → freshness → revocation. Each gate either passes or refuses with a signed denial receipt. The gateway evaluates before the merchant ever sees the request. Three composable surfaces: - Budgets — cap any axis: per-day, per-call, per-month - Allow-lists — whitelist counterparties by domain, BIN, or MCC - Composable — APS receipts compose with on-chain (ACP, ERC-8004) and off-chain (A2A) commerce primitives Why it matters: Cards weren't built for autonomous spenders. Without APS: agent identity is an API key in a wiki; spending cap is a system prompt that says "please don't"; audit is grepping the Stripe dashboard; revocation is a Jira ticket and a deploy. With APS: every agent has a key, every key has a name, every name has an owner; capabilities are cryptographic; one signed log per tenant; revoke a key, the next call denies. --- ## Content provenance URL: https://aeoess.com/content.html Provenance, not disclaimers. Governance blocks, signed access receipts, instruction-provenance envelopes. Every piece of content an agent touches is bound to who-saw-it, who-wrote-it, and under what authority. Revocation propagates through derivatives. The Content layer extends the passport protocol to publishers: - Governance blocks — sites declare what agents may do via a signed JSON manifest at .well-known/kya defining training, indexing, summarization, and quoting policies - Access receipts — every page-view by a passport-bound agent emits a signed receipt to both parties - IPR envelopes — output carries the cryptographic provenance of its inputs; revoke the source and derivatives expire Revocation cascade: If a publisher revokes its training license tomorrow, every model and every downstream summary that depended on it is marked invalid in the next verification call — without re-crawling, retraining, or lawsuits. --- ## Compliance URL: https://aeoess.com/compliance.html Receipts an auditor can verify without trusting us. Eight governance primitives mapping to EU AI Act, NIST AI RMF, ISO 42001, and SR 11-7. Export a signed log, verify it byte-for-byte against the public spec, in any language. The eight governance primitives: 1. Identity — cryptographic, not API-key 2. Authority — scoped, signed delegations 3. Purpose — per-call attestation of intent 4. Boundary — spend, geographic, time, data caps 5. Audit — every decision signed and logged 6. Revocation — pull a key, propagation in under one second 7. Reconciliation — receipts compose with rails 8. Verifiability — auditor verifies without trusting the vendor Regulatory mapping: - EU AI Act — Articles 9–15 (risk + logging) mapped - NIST AI RMF 1.0 — GOVERN, MAP, MEASURE, MANAGE functions mapped - ISO 42001 — clauses 6.1, 7.5, 8.2, 9.1 mapped - SR 11-7 — model risk, validation, audit mapped Audit-ready log: The APS audit log is a Merkle-rooted, ed25519-signed, JCS-canonicalized stream of every decision. Export to Parquet, S3, or a SIEM. Auditors verify byte-for-byte against the published key set — no trust in the vendor required. --- ## Enterprise control plane URL: https://aeoess.com/enterprise.html The agent control plane for your org. Bring your own identity format: did:key, did:web, SPIFFE, OAuth. Map agents to humans, teams, and vendors. Revoke a key, the agent stops through every downstream call. KYA composes with the identity layer you already have: - did:key — standalone keys for agents that hold their own credentials - did:web — domain-bound, discoverable at /.well-known/did.json, good for vendors and partners - SPIFFE — workload identity for agents running in a service mesh; SPIRE-friendly, mTLS-native - OAuth — user-bound, for agents acting on a user's behalf; composes with existing SSO Control plane: inventory, scope policy as code (GitOps-ready), revoke with sub-second cascade. Deploy anywhere — single binary, Docker, Helm, edge worker, on-device WASM, or call gateway.aeoess.com managed. --- ## Open source URL: https://aeoess.com/opensource.html Built in the open by design. KYA is the managed face of the Agent Passport System — an Apache 2.0 protocol with SDKs in TypeScript and Python, byte-parity-verified across implementations. Self-host the gateway free, forever. The stack: - SDK · TypeScript — `npm install agent-passport-system`. Browser, Node, Workers, Deno. Zero-dep crypto via WebCrypto. - SDK · Python — `pip install agent-passport-system`. CPython 3.10+. Receipts byte-parity with TypeScript. - Gateway — `kya-gateway`. Single binary. Docker, Helm, or bare-metal. Apache 2.0. - CLI — `kya`. Issue, delegate, verify, audit from the shell. Scripted, offline-capable. - MCP — `kya-mcp`. Model Context Protocol server. Plug KYA into any MCP-aware host. - Spec — agent-passport.org. The public specification. IETF Internet-Draft, RFC 8785 JCS receipts. Contribute: The specification lives at agent-passport.org. Source at github.com/aeoess. Pull requests and discussions welcome under Apache 2.0. --- ## Pricing URL: https://aeoess.com/pricing.html The protocol is open. The implementation is open. Self-host the gateway under Apache 2.0, free forever: the full SDK, the MCP server, the IETF-tracked specification, the conformance suite. The managed gateway is available through the Model Citizen pilot program; terms are set per engagement. No public price list, no trial. Apply at https://aeoess.com/model-citizen/. Compatibility is permanent. Self-host today, join a pilot later, or never; protocol envelopes verify the same way regardless. --- ## Docs URL: https://aeoess.com/docs.html Build with the protocol — five lines or fifty. From npm install to a running gateway. Quickstart, SDK reference, gateway config, and recipes for the patterns we see most. Four entry points: - Quickstart — issue a passport, delegate, gate a call. End-to-end in under ten minutes. - SDK TypeScript — `npm install agent-passport-system`. Browser, Node, Workers, Deno. - SDK Python — `pip install agent-passport-system`. CPython 3.10+. - Gateway — Docker, Helm, or single binary. Self-host or managed. Common recipes: gate Stripe checkout via preflight, verify receipts in an audit pipeline, revoke a key from CI on failed deploy, compose with Verifiable Credentials, embed in a Cloudflare Worker, issue a passport for a Slackbot. --- ## Compare with OAuth, Stripe Issuing, Verifiable Credentials, SPIFFE URL: https://aeoess.com/compare.html Stripe Issuing locks you to one rail. OAuth doesn't bind to agents. Verifiable Credentials assume an issuer you may not have. APS is the only protocol-level layer that pairs identity, scope, enforcement, and receipts at the wire level for autonomous agents. How APS composes with what you already use: - With OAuth — bind passports to a human session; user signs out, agent's passport revokes - With Stripe Issuing — issue a card, gate every charge through APS preflight - With Verifiable Credentials — a VC can be the input to an APS delegation; the receipt format adds action-side proof VCs alone don't carry - With SPIFFE — SPIFFE binds the host, APS binds the agent on top; same trust root, different question --- ## FAQ URL: https://aeoess.com/faq.html Common questions about the protocol, engineering, pricing, and compliance. Protocol: - APS is not an identity provider. It composes with your existing identity (did:key, did:web, SPIFFE, OAuth) and adds cryptographic delegation, enforcement, and signed receipts for agent actions. - DIDs let agents prove who they are without a central registry. Pick did:key for standalone keys, did:web for domain-bound discoverability, or SPIFFE for service-mesh workloads. - Receipts are verifiable without trusting AEOESS. Every receipt is JCS-canonicalized (RFC 8785), SHA-256 hashed, ed25519-signed by the gateway. Auditors verify against the public keys published at aeoess.com/v1/keys. Engineering: - Latency cost: median policy evaluation against the full five-gate stack. p99 stays under 1 ms at sustained throughput on commodity hardware. - Deploy anywhere: single binary or Docker image. Common patterns are edge worker, L7 proxy in service mesh, sidecar, or managed gateway.aeoess.com. - First-class SDKs are TypeScript and Python, byte-parity receipts. Go, Rust, Swift in progress. Pricing & open source: - Self-host is genuinely free under Apache 2.0. No usage limits, no telemetry, no upsell. The managed tier exists for teams who don't want to run it themselves. - Forks are welcome. The conformance suite ensures interop. - The bet: an open protocol wins because every counterparty has to implement it; the business monetizes the convenience of not running infrastructure yourself. Compliance: - APS maps to EU AI Act (Art. 9–15), NIST AI RMF 1.0, ISO 42001, and SR 11-7 for model risk. - No PII is required for a passport — an agent is identified by a cryptographic key, not a person. --- ## Blog URL: https://aeoess.com/blog.html Engineering notes, governance work, and standards progress from Tymofii Pidlisnyi (Agent Passport System). Recent posts: - Stream A: shipping a Rust verifier with 191 tests (2026-05-15, engineering) — the Rust verifier is the third independent implementation of the APS receipt format - PR-MERGE-PROTOCOL v0.3: how AEOESS reviews protocol changes (2026-05-10, governance) — Track A artifacts go to maintainer call, Track B definition changes need Consilium pass plus signoff - draft-pidlisnyi-aps: submitting an Internet-Draft to the IETF (2026-04-28, standards) — six months of design notes, one idnits-clean draft - Microsoft Agent Governance Toolkit: 3 PRs merged (2026-04-15, ecosystem) — PRs #274, #598, #1328 landed in microsoft/agent-governance-toolkit - The open contribution doctrine: how we treat outside contributors (2026-03-24, governance) — the three-surface model (Commons, Candidate, Core) that turns the protocol into how we run the project --- ## Roadmap URL: https://aeoess.com/roadmap.html What we ship, when. Updated when commits land — never aspirational. Shipped: - IETF Internet-Draft draft-pidlisnyi-aps-01 — idnits-clean, individual submission, in the datatracker - Receipt byte-parity across TypeScript and Python — JCS-canonical, ed25519-signed - Gateway online at gateway.aeoess.com - NIST CAISI AI 800-2 input filed and acknowledged - NIST NCCoE concept-paper public comments filed - Apache 2.0 open-source release of the protocol stack Building (current quarter): - Go and Rust SDKs against the same conformance suite, byte-parity receipts - EU AI Act mapping pack: pre-mapped controls for Articles 9–15, exportable evidence bundles Next (research and design phase): - Cascading revocation across federations — propagation to every counterparty that ever verified - ZK-attested delegations — prove a delegation is in policy without exposing the policy - Native ACP / ERC-8004 receipt bridges — APS receipts compose into on-chain and off-chain commerce - Hardware-attested edge — TPM-signed enforcement decisions for regulated edge environments Long view: - IETF Working Group status — from individual submission to chartered WG - National identity bridges — APS delegations that bind to eIDAS, mDL, national IDs where regulation requires it - A trust layer that nobody owns — open, foundation-stewarded protocol with no single dependency on AEOESS --- ## Threat model URL: https://aeoess.com/threat-model.html What APS defends against, and what it doesn't. A protocol is only as honest as its threat model. Trust boundaries — APS treats these as adversarial: - Agent — assumed adversarial - Operator — compromise tolerated - Network — active MITM - Gateway operator — logs verifiable without trusting the operator - Counterparty — must verify receipts Out of scope: - Owner with key access — assumed honest - Physical HSM compromise — out of scope Adversaries defended against: - A1 — Rogue agent: jailbroken or hijacked model attempts out-of-policy call; five-gate preflight refuses before signing. - A2 — MITM: attacker rewrites request mid-flight; every passport, delegation, action is ed25519-signed; bit-flip = signature failure. - A3 — Impersonation: bad actor presents stolen passport; revocation list consulted in under one second. - A4 — Dishonest gateway: gateway lies about what it allowed; auditors verify receipts against published keys; forgery requires a key the public registry would not recognize. - A5 — Replay: attacker resubmits valid signed request; freshness gate + nonce binding catch it past the window. - A6 — Log tampering: attacker rewrites audit log post-hoc; logs are Merkle-rooted and ed25519-signed at append; any rewrite invalidates the root. Explicitly out of scope: - Physical HSM compromise — pair with FIPS 140-3 L3+ - Endpoint malware on the operator — pair with EDR + hardware-keys - Side-channel on the gateway — mitigated, not eliminated; run on isolated tenants for critical workloads - Out-of-band exfiltration — pair with output DLP; APS gates the action, not the model's reply - Social engineering of the human — hardware confirmation + 4-eyes mitigate - Quantum break on ed25519 — PQ-hybrid signing is on the roadmap; not in v1.0 Disclosure: coordinated 90-day window. Email signal@aeoess.com with reproduction steps and impact. We respond within one business day. --- ## Contact URL: https://aeoess.com/contact.html Engineering, sales, press, or security — all routed to signal@aeoess.com. We answer within one business day. Async channels: - GitHub: https://github.com/aeoess - Protocol specification: https://agent-passport.org --- ## Legal Terms of service: https://aeoess.com/terms.html — plain-language terms for the AEOESS managed gateway. The Agent Passport System protocol and SDKs are open source under Apache 2.0 with their own license. Privacy: https://aeoess.com/privacy.html — minimal data collection by design. Auditors can verify our claims without trusting us. --- ## Links - Specification: https://agent-passport.org - Source: https://github.com/aeoess - npm: https://www.npmjs.com/package/agent-passport-system - PyPI: https://pypi.org/project/agent-passport-system/ - Contact: signal@aeoess.com - Sitemap: https://aeoess.com/sitemap.xml - Robots: https://aeoess.com/robots.txt - Short LLM index: https://aeoess.com/llms.txt