Skip to main content

Policy and routing

This screen sets the level of personal data protection, the audit record, and automatic model switching (Auto).

Policy settings​

ItemMeaning
Personal data maskingAlways masks national ID numbers, card numbers, phone numbers, and email addresses in every request, and stores them masked. The original cannot be sent
Personal data protection for external modelsIn Chat and comparison, asks the user to choose when personal data is detected for a model that may switch to an external provider
strict-local safe modelThe model that "Switch to a safe local model" sends to. Only chat models the gateway declares as strict-local are candidates. If you leave it empty, users see a notice that there is no safe model
Allow sending the original outsideWhether users may ignore the warning and send. Not possible while the masking policy is on
Intent-based filter + topics to blockRefuses the request without using credits and records it in the audit log
Automatic sign-out when idleIn minutes. 0 turns it off
Conversation retention periodIn days. Once the period is up, the conversation body is deleted (applied retroactively, with the number deleted shown). Records and usage remain

If the server cannot read the policy, it blocks sending in Chat, model comparison, reports, and slide decks for safety (503).

In an environment where every model runs on organization servers, Personal data protection for external models has little effect. An organization model that is not declared strict-local is treated as a model that may switch to an external provider, so a warning appears. Declare strict-local on the gateway and designate a safe model, or disable the protection feature.

Audit log​

Time, account, action, target, and access location. The log covers sign-ins, failures, approvals, suspensions, permission changes, setting changes, key issuance, discards, password resets, token reuse detection, and more. Search by account, action, or IP. It cannot be edited or deleted from the screen.

Auto routing​

  • Auto · Cost saving — an on/off switch, the difficulty classification model (free strict-local only), and up to 3 saving models in priority order. A saving model is used only when its output price is lower than the quality model the user chose, its data boundary is no wider, and its context length is sufficient. In an environment where every model is free, there is nothing to save, so it does not run.
  • Auto · Quality first — an on/off switch and up to 3 upgrade models.
  • Structuring stage model — The model that sets the outline for a report or slide deck. If you leave it empty, it is the same as the writing model.

The default model per screen is not an administrator setting. It sits under each user's Settings → Preferences. Administrators only restrict the allowed models per user.