Settings, license & system

v1.4.x · updated 2026-07-10

This page covers three “system” Web UI screens: Settings, License and System.

Settings

The “Settings” screen is the editor for the instance settings catalog. Keys owned by dedicated pages (auth, models, external access) are not duplicated here. Features:

live search across settings;
categories with collapsible groups (collapsed by default, state remembered);
the control type follows the setting type: boolean → toggle, enum → a dropdown in the AiHummer design system, integer → number, else text/password;
dependent fields: a setting is shown only when it makes sense for the current value of its “driver” setting — for example the Deepgram keys appear only when the STT engine is deepgram, and the embedder parameters only in the embedding memory mode; the same rule applies inside plugin configuration forms (a field with depends_on appears only for the matching value of its controlling field — e.g. the Telegram inbound-webhook URL is hidden in polling mode);
default values are displayed filled in — you see what actually applies, not an empty box;
a source chip is shown only for overridden values (db/env); CloudLocked keys are read-only;
on save you’re told whether the value applied live or needs a restart.

The “env is bootstrap-only, everything else lives in the DB” principle is in Configuration.

License

The “License” screen shows the current plan and entitlements. The plan card: the tier name (community / starter / business / enterprise / vip), a validity status (perpetual / valid until … / expired / invalid / grace period), and the licensee. Usage bars (AI employees, tenants vs the max; 0 = unlimited), a channels note (all or core only). A feature matrix (white_label, rbac, audit, sso, rls, sla) with ✓/✗. The upgrade button goes to the public pricing page. Offline activation is available: paste a license token and click “Install”.

System

The “System” screen is version and update. The version card: the current version and commit, the latest available release, an “update available / up-to-date” badge, and vendor pin/deny notes. The “Check for updates” and “Update now” buttons (host-native self-update; hidden on managed cloud with a “vendor-managed updates” note). The “About” block has a toggle to hide AiHummer branding (white-label), available when the plan grants it.

Configuration — env vs DB.
Licensing — plans and entitlements.
Upgrade policy — how to update.

Data protection (DLP)

Off by default. Checks text before every model request, including auxiliary models. Enabled protection blocks images and audio that cannot be inspected. Detection is rule-based and may miss unusual formats; add your own rules. Audit contains no original matches.

Requires Family butler, Business or higher.

Action: Mask / Block. Custom sensitive data: Exact text, Regular expression.

DLP replaces matching text with [REDACTED] or stops the model call; a blocking rule takes precedence. Stored chat history is unchanged. Built-in identification and address rules target Russian formats; unusual or other national formats need custom rules. Text embedded in tool arguments is checked too. Signed model reasoning that cannot safely be changed is blocked when it contains a match.

Only a tenant-wide administrator with settings:write may read or change the rules. A workspace administrator does not receive this access. Rules are encrypted; simultaneous edits return a conflict instead of overwriting changes. If an enabled policy cannot be read or its audit cannot be stored, the model request is stopped. An expired entitlement does not silently disable protection; an administrator can renew it or explicitly turn protection off.

For corporate collection, authenticate to GET /v1/admin/dlp/events?after=0&limit=1000&format=ndjson with a tenant-wide policy:read grant. Continue from the returned X-Next-Cursor until a page is empty. Events contain time, a random request identifier, policy revision, action, rule category and match count; no source text or matched values. Retention uses the existing audit retention setting. The browser downloads up to 1000 records; continuous collection uses the paginated API.

Setup assistant

Connect a model and successfully test it on the Models page to open the setup assistant.

Changes proposed by the setup assistant require your confirmation. Its conversation is separate from ordinary chats; enter credentials on the secure settings pages.