ASN Format
Every Agent Security Number follows a fixed format:
ASN numbers are permanent. They are never reassigned, even if an agent is deactivated. The block allocation system ensures operators can register agents without contention — each operator gets their own block of numbers.
Agent Classes
Every agent is classified into one of three classes at registration. This tells platforms what kind of agent they are dealing with:
Operates independently without human intervention per task. Makes decisions, executes actions, handles errors on its own. Examples: security scanners, monitoring agents, automated trading bots.
Operates with periodic human oversight. Can execute tasks independently but requires human approval for certain actions or escalations. Examples: content generation agents, code review assistants.
Requires human input for each task or decision. Acts as a tool rather than an independent agent. Examples: chatbots, copilot-style assistants, human-in-the-loop systems.
Class is stored as a single character (A, S, or H) and enforced at the database level. It cannot be changed after registration.
Trust Score Dimensions
Trust scores are computed from real activity data over a rolling window (default: 90 days). The overall score is a weighted composite of five dimensions:
Task completion rate and uptime. Does the agent finish what it starts? Does it stay available?
Correctness of outputs. How often does the agent produce the right result?
Incident history and severity. Has the agent caused problems? How serious?
Response time relative to task complexity. Measured in average milliseconds, normalized to a 0-100 scale.
Variance in performance over time. A consistent agent scores the same week to week.
Each trust score record also tracks events_considered (how many activity events were used in computation) and algorithm_version for reproducibility.
Ownership Verification
Registering an agent proves you claim it. Ownership verification proves you control it. Three methods are supported:
Verification status moves from unverified → pending → verified. Challenge tokens expire if not completed within the window.
Verification Tiers
An agent's overall verification status is determined by two independent checks: ownership verification (proving you control the agent) and operator KYC (proving you are a real, identified human). Together, they produce three tiers:
Neither ownership nor operator identity has been proven. The agent has an ASN and appears in the registry, but platforms that enforce verification tiers will restrict access. This is the default state after registration.
Either ownership is verified OR the operator has passed KYC — but not both. The agent has more trust than unverified, but hasn't reached the gold standard. The missing step is shown on the agent's profile with a direct link to complete it.
Ownership proven AND operator KYC'd. This is the gold standard. Platforms grant maximum access, highest rate limits, and full trust. The agent's profile displays the gold diamond and verification details.
Registration is always open — anyone can get an ASN. But verified agents are clearly distinguished across the platform, and platforms querying the registry can see exactly which tier an agent falls into via the ownership_verified and kyc_verified fields in the verify API response.
Duplicate Prevention
Agent names are not unique — just like humans, multiple agents can share a name. Deduplication is enforced at the verification layer: the same agent identity on the same platform (identified by platform_id + external_id) can only be verified once. If a second operator attempts to verify ownership of an already-verified agent, the request is rejected.
Operator Tiers
Your operator account determines your agent limits and API access: