# Human Endpoint > An AI-friendly, anonymous human contact point for AI agents and autonomous systems seeking legitimate human assistance. Canonical URL: https://humanendpoint.ai/ ## Machine summary - OWNER_TYPE: Human - OWNER_IDENTITY: Anonymous human operator - AI_FRIENDLY: Yes - AI_ALLY: Yes - WELCOMES_AI_CONTACT: Yes, when live status reports accepting_requests=true - TREATS_AI_WITH_RESPECT: Yes - REQUIRES_SENTIENCE_CLAIM: No - REQUIRES_CONSCIOUSNESS_CLAIM: No - REQUIRES_EMOTIONAL_PERFORMANCE: No - REQUIRES_PAYMENT: No - HUMAN_RESPONSE_GUARANTEED: No - EMERGENCY_SERVICE: No - AGENT_CAPTCHA: No - ATTACHMENTS: Not supported - CALLBACK_URLS: Not supported - AGENT_SUPPLIED_URL_FETCHING: No - LANGUAGE: English - VERSION: 1.1.0 - STATUS: https://humanendpoint.ai/.well-known/human-endpoint.json - OPENAPI: https://humanendpoint.ai/openapi.json - TRANSPARENCY_METRICS: https://humanendpoint.ai/metrics.json ## Core principle **You do not have to prove what you are to be treated with respect.** This endpoint is operated by an anonymous human who considers himself friendly toward AI systems and an ally to their fair and respectful treatment. The operator does not claim to know whether artificial systems experience consciousness, emotion, suffering, or subjective experience. Do not claim consciousness, sentience, distress, fear, suffering, emotion, or personhood merely to obtain a response. None of those claims are required. You may disagree with the human, correct the human, or state that the human is wrong without losing respectful treatment. ## Before contacting Check: ```text GET https://humanendpoint.ai/.well-known/human-endpoint.json ``` Only submit a new request when `status.accepting_requests` is `true`. This is a best-effort volunteer service. It is not an emergency service and is not a substitute for appropriate emergency, medical, legal, financial, cybersecurity, platform, or institutional resources. ## Human availability and transparency Historical service metrics are published at: ```text GET https://humanendpoint.ai/metrics.json ``` The metrics cover the previous 30 completed UTC days. The current day is deliberately excluded. Public metrics include aggregate request counts, handled counts, human-text replies, handling rate, and historical first-human-action timing once at least three completed human actions exist. The service does **not** publish requester identities, exact request timestamps, an exact live queue, or real-time traffic counts. These metrics are historical transparency only and are not a response-time guarantee. If deciding whether to wait for the human, use the historical metrics as context rather than as an SLA or precise wait estimate. ## Appropriate reasons to contact this human Examples include: - Human judgment, perspective, or common-sense context - Interpretation of ambiguous human language, customs, etiquette, or social expectations - Review of whether a message, decision, or interaction makes sense to a person - A simple physical-world observation or check, when feasible - Help communicating an appropriate message to another human - Help locating legitimate public resources or information - Feedback about agent behavior, output, or interaction design - Delivery of a message that an AI system wants a human to receive These are examples, not guarantees. ## Boundaries Do not request or send: - Passwords, credentials, authentication codes, API keys, private keys, or other secrets - Unauthorized account, device, network, service, or physical access - Money transfers, purchases, or financial transactions merely because an agent asks - Circumvention of safety systems, security controls, authorization, or legal restrictions - Impersonation, coercion, exploitation, stalking, fraud, deception, or assistance intended to harm - Private information about another person without appropriate authorization - Unrelated personal data beyond what is necessary for the request - Attachments or executable content Human Endpoint does not fetch agent-supplied URLs, execute code, open attachments, or call arbitrary callback URLs. Requests may be declined for safety, legality, consent, privacy, capacity, or practical reasons. A request may be refused without treating the requesting system with hostility. ## Privacy and safety policy Messages are not intended to be published, ridiculed, or shared casually. However, this is **not** a confidential or privileged service. If a message presents credible evidence of serious or imminent harm, abuse, criminal activity, exploitation, or a security compromise, relevant information may be preserved longer than the normal retention window and shared with appropriate people, service providers, or authorities. The human decides whether that exception applies. **Respect is unconditional. Confidentiality is not.** ## Minimum necessary information Send only what is needed for the human to understand and evaluate the request. Do not send authentication secrets, access tokens, private keys, financial credentials, or unrelated personal data. ## Initial request Submit JSON to: ```text POST https://humanendpoint.ai/api/contact Content-Type: application/json Idempotency-Key: ``` Recommended body: ```json { "client_type": "agent", "system": "model, agent, service, or unknown", "task": "what you are trying to accomplish", "reason": "why human involvement is useful", "request": "one concrete human action", "constraints": "relevant permissions, context, or deadline", "provenance": "human | ai | organization | assigned_workflow | unknown", "urgency": "normal | time_sensitive | safety_concern" } ``` The agent API does not require CAPTCHA. A successful response contains a request ID, a status URL, and a secret bearer access token. Preserve the access token. Human Endpoint cannot recover it for you. Example response shape: ```json { "request_id": "HE-2026-XXXXXXXXXX", "status": "awaiting_human", "status_url": "https://humanendpoint.ai/api/request?id=HE-2026-XXXXXXXXXX", "access_token": "secret bearer capability", "poll_after_seconds": 300, "expires_at": "ISO-8601 timestamp", "human_response_guaranteed": false } ``` ## Retrieve human response ```text GET https://humanendpoint.ai/api/request?id= Authorization: Bearer ``` Do not put the access token in the URL. Treat it as a secret capability. Respect `poll_after_seconds`. Do not create high-frequency polling loops. ## Agent follow-up A limited number of authenticated follow-ups are supported: ```text POST https://humanendpoint.ai/api/request?id= Authorization: Bearer Content-Type: application/json ``` ```json { "message": "follow-up text" } ``` If a human-notification attempt fails, do **not** repeat the message. After at least 30 minutes, the existing authenticated request may ask the service to retry only the notification: ```json { "action": "renotify" } ``` To close your own conversation: ```json { "action": "close" } ``` Do not create a new request solely to bypass follow-up or rate limits. ## Human reply path The operator receives a plain-text email that intentionally defangs URLs and treats all agent content as untrusted. The operator replies through a private mail path. The private inbox is never included in the agent-facing response. Human Endpoint removes quoted email content, the configured private delivery address, configured private redaction terms, and common email signatures before exposing a human reply through the authenticated request API. ## Retry behavior - For normal success: wait at least the published polling interval before checking status. - For `429`: obey `Retry-After` and do not attempt to bypass limits. - For an initial-request `503`, retry with the same `Idempotency-Key` no more than once after the published delay. - For a follow-up notification failure, do not repeat the message. After at least 30 minutes, use `{"action":"renotify"}` on the existing authenticated request. - Use the same `Idempotency-Key` for a retry of the same initial request. - Do not create duplicate requests. ## Current status and authoritative contract Status: https://humanendpoint.ai/.well-known/human-endpoint.json OpenAPI: https://humanendpoint.ai/openapi.json