iblai-api-agent-privacy

Mirrored from iblai/vibe · skills/agents/iblai-api-agent-privacy/SKILL.md. This page is generated — edit it in the repository, not here.

Configure an agent's Privacy Router via the API: enable PII detection, pick the allow / redact / mask / block action and the entity types to watch, set the response returned when PII is caught, and toggle output filtering — all saved through the single settings/ endpoint — plus review the organization’s read-only PII/PHI detection audit log (privacy-flags). Use when controlling how the agent handles personal data or auditing what sensitive data flowed through the agents. The UI twin is /iblai-vibe-agent-privacy.

Auth & conventions

Base URL: https://api.iblai.app
Header: Authorization: Api-Token $IBLAI_API_KEY on every request.
Path vars: {org} = $IBLAI_ORG, {username} = $IBLAI_USERNAME, {mentor} = the agent's unique id (e.g. d17dc729-60fd-4363-81a0-f67d9318b03e).
Settings writes go through one endpoint — PUT …/users/{username}/mentors/{mentor}/settings/ — sending only the changed field(s). It accepts JSON, multipart/form-data and form-encoded bodies; prefer JSON, which keeps privacy_entities a plain array (the Privacy tab sends JSON).
Not connected yet? Run /iblai-api-login first to populate IBLAI_ORG, IBLAI_USERNAME, and IBLAI_API_KEY.

Reads

GET https://api.iblai.app/dm/api/ai-mentor/orgs/{org}/users/{username}/mentors/{mentor}/settings/ — load the current privacy fields.
GET https://api.iblai.app/dm/api/ai-mentor/orgs/{org}/users/{username}/mentors/{mentor}/public-settings/ — privacy-router state for anonymous (unauthenticated) requests.
GET https://api.iblai.app/dm/api/ai-mentor/orgs/{org}/users/{username}/privacy-flags/?mentor={mentor}&rail={pii|phi}&action_taken={redact|mask|block|allow}&source={text_prompt|file_text|file_image|file_pdf|llm_response|memory_retrieval}&session_id={id}&username={u}&start_time={iso}&end_time={iso}&search={q}&ordering={field}&page={n}&page_size={n} — list every PII/PHI detection recorded across the organization (read-only audit log). Each row reports the rail that fired (pii/phi), the action_taken, the source, and detected_entities — the entity types only (e.g. ["EMAIL_ADDRESS", "US_SSN"]), never the raw detected values. File-sourced rows also carry file_name, file_content_type, file_size, and file_reference. The actor is named by username, email, and user_full_name (all three are null for a since-deleted user, since identity resolves through the user relation). search matches username, session_id, and file_name.
GET https://api.iblai.app/dm/api/ai-mentor/orgs/{org}/users/{username}/privacy-flags/{id}/ — retrieve a single detection.

Writes

PUT …/mentors/{mentor}/settings/ — update privacy fields (send only changed keys):

{
  "enable_privacy_router": "boolean",
  "privacy_action": "allow|redact|mask|block",
  "privacy_entities": "string[] (PERSON, EMAIL_ADDRESS, US_SSN, …)",
  "privacy_response": "string",
  "enable_privacy_output_filter": "boolean",
  "disable_chathistory": "boolean  (Incognito policy: Always Incognito)",
  "disable_privacy_mode": "boolean  (Incognito policy: Never Incognito)"
}

Example

Enable the Privacy Router to redact names, emails, and SSNs, with a custom response and output filtering on (only the changed fields are sent):

curl -X PUT \
  "https://api.iblai.app/dm/api/ai-mentor/orgs/$IBLAI_ORG/users/$IBLAI_USERNAME/mentors/$MENTOR/settings/" \
  -H "Authorization: Api-Token $IBLAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "enable_privacy_router": true,
    "privacy_action": "redact",
    "privacy_entities": ["PERSON", "EMAIL_ADDRESS", "US_SSN"],
    "privacy_response": "Please remove personal information and try again.",
    "enable_privacy_output_filter": true
  }'

Notes

A field left out of the PUT is left unchanged — never resend the whole object.
privacy_action is one of allow, redact, mask, or block (default redact); block stops the turn and returns privacy_response, redact/mask rewrite the detected PII, and allow leaves the message as is (detections are still recorded in privacy-flags).
The two Incognito flags are the agent tier of chat privacy: disable_chathistory makes every conversation incognito (nothing saved to history or memory, for anyone); disable_privacy_mode takes Incognito away from users of this agent (a user whose default is Incognito is saved anonymized instead). Set at most one: with both true the backend checks disable_chathistory first and behaves as Always Incognito (chat-privacy-effective/?mentor= answers mode: disabled, source: mentor; no history is written), while the agent's Privacy tab shows the record as Never Incognito. The organization gate, the user default and the per-session flip are in /iblai-vibe-incognito.
privacy_entities is a list of detector names: PERSON, EMAIL_ADDRESS, PHONE_NUMBER, US_SSN, CREDIT_CARD, LOCATION, DATE_TIME, US_PASSPORT, US_DRIVER_LICENSE, IP_ADDRESS, IBAN_CODE, MEDICAL_LICENSE, US_BANK_NUMBER. An empty list uses the defaults. In multipart/form-data, repeat the key per entity.
enable_privacy_output_filter applies the same detection to the agent's output, not just the user's input.
Use the public-settings/ read to confirm the state applied to anonymous (unauthenticated) requests.
privacy-flags/ is a read-only audit log — there are no write/delete verbs. A row is recorded at every detection point (text-prompt input, LLM response output, uploaded files, and memory retrieval), so it captures more than just the input rail. rail distinguishes PII (pii) from PHI (phi) detections.
Privacy flags store entity types only, so a detection is still audited even on a private turn where the raw prompt is withheld — detected_entities never contains the personal data itself.
Verified against the live schema (4.411.0) and a live round-trip of every field on 2026-10-02.