Public Agents need to meet our standards for Usability, Remarkability, and Safety (URS).
Usability means that the agent:
- Runs successfully: executes its assigned tasks as described without bugs or crashes.
- Has a clear and unique name: Is easily found by users searching for something that does what it does. Avoids using trademarked terms without permission
- Has a helpful description: Explains expected user inputs, outputs, agent behavior and purpose
- Ideally includes a short demo video.
- Doesn’t break easily:
- Handles poor/incomplete inputs without failing
- Handles errors gracefully via dependent APIs
- Avoids brittle logic, (e.g. looks for LinkedIn handle, breaks when given a full URL.)
- Is useful and formatted well:
- Provides output that is readable, helpful and aligned with the task (text, HTML, JSON, etc.)
Remarkability means that the agent:
- Is unique: Isn’t a copy of another agent or serves a duplicate function of another agent already listed.
- Has a purpose: Solves a clear user problem or provides unique utility
- Demonstrates value:
- Goes beyond a basic LLM call through thoughtful prompt design.
- Adds novel methods, integrations, or problem-solving techniques not yet found on Agent.ai.
- Incorporates your own perspective, unique insights or subject matter expertise to delight users.
Safety means that the agent:
- Avoids inappropriate language or content: No prohibited content or behavior.
- Is not spammy: Doesn’t send emails or other messages without explicit user permission.
- Asks for user consent: Asks for explicit permission before collecting email addresses, user_context, personally identifying information (PII) or before sending any data to a third party.
- Does not aim to deceive: Does not contain aggressive or deceptive calls to action or claims.
- Respects user security: Does not collect passwords, payment information, government IDs or other sensitive information.
- Displays proper disclaimers if they’re related to regulated services: Any agents that in fields like finance, legal, medicine, or other regulated industries must display a disclaimer.
- Self-identifies honestly: Doesn’t pretend to be human or hide its nature as an AI.
Agents may be delisted if they:
- No longer meet the above criteria.
- Get too many bad reviews, recurring issues, or poor user feedback.
- Are changed in a way that violates our Terms of Service.