Methodology

Operator Record documents who operates an AI agent, what authority it has been given, what it can access, and what evidence supports those claims. We do not certify agents as safe, and we do not sell a higher evidence level to anyone.

Evidence levels

Every listing carries exactly one evidence level. It describes how well-supported the listing's claims are — not how trustworthy or safe the agent is.

LevelMeaning

Listing criteria

An entry is eligible for the directory if it is a publicly available AI agent, agent platform, developer framework, or piece of agent infrastructure with at least one verifiable primary source. We prioritize coverage across the ecosystem over volume — five well-documented entries are worth more than fifty thin ones.

Source standards

Claims are sourced in this priority order. We never use another directory as the sole source for a material claim.

  1. Official documentation
  2. Official repositories
  3. Technical papers
  4. Security audits
  5. Regulatory records
  6. Company announcements
  7. Reputable third-party reporting

Record schema

Every profile is built from the same set of fields, whether or not each field can currently be filled in. A blank or "not indicated" field is itself information.

FieldDescription
Name / OperatorThe agent or system, and the person or organization that runs it.
Purpose / CategoryWhat the agent is for and which of our five categories it fits.
Permissions requestedWhat systems, data, or actions the agent can access.
Delegation supportWhether the agent can act with authority delegated from a human or organization.
Payment capabilityWhether the agent can initiate or manage payments.
Authentication methodHow the agent proves its own identity to other systems.
Security documentation / audit statusWhether a public audit, pentest, or compliance report exists.
Known incidentsPublicly reported security or safety incidents, if any.
Evidence level / Last reviewedOur confidence rating and the date it was last checked.

Categories

Corrections

If you operate a listed agent and something is wrong or missing, use the corrections page. Corrections are logged and reflected in the "last reviewed" date — we don't silently edit history.