Reference
Settings reference
Every setting the YuviRed console exposes under Admin, Configuration, grouped the way the console groups them, with defaults, allowed values, who may change them and what to watch for.
Covers YuviRed 1.0 for Windows (app 1.0.0, agent 1.8.2, extension 0.5.5). Defaults come from the server's settings table (SETTINGS_SCHEMA); labels and choices come from the console. Where the two differ, the console's wording is used for the label and the server's value is the default.
How settings work
- Where: Admin, then Configuration, panel Platform Configuration. Nine sections (below), then a collapsed Advanced settings: uncommon, granular controls box and a Diagnostics box (read only). Chips at the top jump to a section. Rows marked "Advanced" in this page sit inside the collapsed advanced box of their section.
- Saving: change one or more rows, then click Save Changes. Each changed setting is saved on its own and every save is written to the audit log. A green toast says
Saved N setting(s). If the server refuses one value you seeSave failed - <key>: <reason>; the other values still saved. Discard Changes reloads the saved values. Leaving the page with unsaved edits asks first. - Confirmation boxes: for a change that reduces detection, privacy or retention, or changes fleet behavior, the console asks you to confirm before saving. These are listed as "Warning" below.
- Who may change settings: reading a setting needs the
config.readpermission. Changing one needsconfig.write. In the built-in roles,config.writebelongs to the two Admin roles only (the first account, and any account created with the Admin role). Analysts and View Only accounts can read most settings through the server but the Configuration page is Admin only in the console menu. Settings underartifact.also needartifact.manage(Admin) to read or write. - When a change takes effect: the server keeps no copy of a setting between requests. Each feature reads its setting from the database when it next needs it, so no setting in this page asks for a restart of the YuviRed service. The practical delay depends on the feature: a sign-in rule applies at the next sign-in; a notification rule at the next finding; the retention job at its next run; a background analysis at its next cycle (the "frequency" settings below); an agent setting at the agent's next configuration pull. Where a delay is longer than "next use" it is stated in the row. Treat "no restart needed" as a code observation: it was not tested for each row individually.
- Secrets: password, webhook and key fields show as masked. They are overwritten only if you type a replacement. Accounts with
config.readbut notconfig.writesee only "configured" or "not set". - Server-only settings: some values cannot be changed from the console. They are listed at the end under "Server-only and hidden settings".
- Units: "UTC" is Coordinated Universal Time. "MB" is megabytes.
1. AI & Agent Monitoring
"How YuviRed observes AI agents and learns their normal behavior." Mostly AMS (Agent Monitoring System) behavior analysis. Findings are report only; nothing is blocked.
| Label | Default | Allowed | What it does | Notes |
|---|---|---|---|---|
| Behavioral Detection | On | On, Off | Builds a behavior profile per AI agent and reports when an agent departs from its own pattern or from similar agents. | Warning on Off: "YuviRed will stop evaluating AI-agent behavior for anomalies." Existing findings are kept. Key ams.behavioral.shadow_enabled. |
| Detection Sensitivity | Balanced (Recommended) | Low: fewer findings; Balanced (Recommended); High: more findings | How strong and persistent a deviation must be before it becomes a finding. Sets two keys together (modified_z_threshold 5, 3.5 or 2.5 and hysteresis_k 3, 2 or 1). | Shows "Custom: tuned values in effect" if the two keys were set separately. |
| Learning Period | Standard (Recommended) | Fast: learn quickly; Standard (Recommended); Conservative: learn longer | How many observations are needed before an agent is judged against its own history (4, 8 or 16) and how much history a profile keeps (25, 50 or 100). | While an agent is still learning it is compared to similar agents instead. |
| Finding History | Standard (Recommended): last 500 | Compact: last 100; Standard: last 500; Extended: last 2000 | How many recent behavioral findings are kept in a rolling list. | Older findings age out of the list. |
| Agent Activity Tracking | On | On, Off | Links endpoint telemetry to the AI agents that produced it. | Warning on Off: new telemetry is no longer linked to agents and Behavioral Detection has nothing new to judge. Also decides which AI agents count against the AMS limit (see editions-and-limits.md). Key ams.correlation_enabled. |
| Rogue Agent Activity Detection | Off | On, Off | Scores each governed AI agent against its approved boundary (unapproved destinations, tool expansion, self-configuration, credential reach, restricted-but-still-active, blocklisted destinations) and raises a Critical alert when one steps outside it. | Warning on On: "can raise Critical alerts within about 2 minutes." Key agent_boundary.enabled. |
| Browser AI Scan Frequency | Standard: every 5 minutes (300 s) | Frequent: every 2 minutes; Standard: every 5 minutes; Reduced: every 15 minutes | How often endpoint agents re-scan browser history for visits to AI services. | The server accepts 30 to 3600 seconds. Applied by an agent at its next configuration pull. More frequent costs slightly more on the computer. Key agent.history_scan_interval_seconds. |
| AI Catalog Review Reminder | Quarterly (Recommended): 90 days | Monthly (30); Quarterly (90); Twice a year (180) | AI tool definitions not reviewed for longer than this are marked overdue on the catalog freshness report. | Server accepts 7 to 730 days. Key catalog.review_cadence_days. |
Advanced rows in this section:
| Label | Default | Allowed | What it does |
|---|---|---|---|
| Analysis frequency | Every 5 minutes (300 s) | Every 2 minutes; every 5 minutes; every 15 minutes; every hour | How often the behavioral engine runs a full pass. |
| Profile adaptation speed | Standard (0.3) | Slow: long memory (0.1); Standard (0.3); Fast: favors recent behavior (0.6) | How quickly an agent's learned normal follows its recent behavior. Fast can normalize a slow drift. |
| Similar agents needed for group comparison | 3 | whole number, 1 to 1000 | Group comparison is skipped when fewer agents of the same type exist. |
| First-time behavior reporting | Report all new behaviors (Recommended) (0) | Report all (0); Ignore minor new behaviors (0.25); Significant new behaviors only (0.5) | Whether a first-time connection or capability is reported on its own. |
| Recent-activity window | 24 hours (Recommended) | 6 hours; 24 hours; 3 days; 7 days | Look-back window for counting recent activity. |
| Detection mode | Report only (Recommended) | Report only; "Active response: not available in this version" (disabled) | This version always reports without acting. Active response does not exist yet; choosing it would change nothing, and the console disables it. |
| Activity tracking frequency | Every 2 minutes (Recommended) | Every minute; every 2 minutes; every 5 minutes; every 15 minutes | How often new telemetry is linked to AI agents. |
2. Prompt Protection
"What the browser extension observes, warns about or blocks before a prompt or attachment leaves for an AI service, using a preset or your own policy." This section is not a list of simple rows. It holds one versioned policy document with its own draft, impact summary, Save and apply steps. Saving the policy needs config.write; viewing it needs alerts.read. The full procedure is in sensitive-data.md.
- Presets (card per preset, plus a comparison table): Monitor Only, Light Warnings, Warnings Only, Balanced, Strong, Strict, Maximum, and Custom (a preset plus your own changes). A new installation starts in Monitor Only ("No popups, banners or blocks. Findings are still recorded in Prompt Monitoring."). A preset marked "Recommended default: review and apply" means no administrator has reviewed it yet.
- Advanced policy (collapsed): content classes, categories, rule overrides, Minimum confidence for an intervention (Low, Medium, High; Medium for every preset except Maximum, which uses Low), Contact details: act on bulk lists only, Payment cards: require card context, handling of content that cannot be inspected (Allow and record, Warn, Block), floors (administrator minimums that may only raise an action), scopes (per destination, endpoint or user), exceptions (may only lower an action; each has an owner, reason and expiry), Confidential terms (up to 200, 2 to 64 characters each), Internal domains (up to 200), and organization formats and namespaces.
- Actions per class: Observe, Warn or Block.
- Warning: "Block" stops a prompt or attachment from leaving the browser. The check runs inside the supported AI sites' pages, so those sites' own code can read your confidential terms and internal domains. Do not enter anything that must stay hidden from them. While the server is locked (not activated after 7 days, or a license that ended) the extension receives Monitor Only and nothing is blocked; blocks applied earlier by a response action stay until rolled back.
- The older setting "Prompt Blocking" (
enforcement.prompt_blocking.default_mode) and its rules are history once a Prompt Protection revision exists; the server refuses changes to them with a 409 and points you to the policy. Extensions older than 0.5.0 receive a mode derived from the policy.
3. Detection & Response
"What gets inspected, what gets blocked, and how quickly alerts should be handled."
| Label | Default | Allowed | What it does | Notes |
|---|---|---|---|---|
| Prompt Content Capture | Redacted excerpts (Recommended) | Off: no capture at all; Activity only: no content; Redacted excerpts (Recommended); Full content | The most content YuviRed may keep about an AI prompt or response event. Endpoints can capture less than this, never more. | Warning on Full content: complete prompt and output text is stored for matched events. Key privacy.capture_level. |
| Inspection Depth | Activity metadata only (Recommended) | Activity metadata only; Full content inspection | Metadata only: records that AI was used, where and by whom, not the prompt text. Full: prompt content becomes eligible for capture, encrypted at rest. | Warning on Full. The server accepts only metadata and full. Key privacy.inspection_mode. |
| Keep Original Prompt Samples | Off | On, Off | Stores encrypted copies of matched prompts for investigation. Needs Inspection Depth = Full. | Warning on On: copies are stored until the retention window ends. Viewing a decrypted sample needs dlp.view_raw_content (Admin roles) and is audited. Key privacy.raw_capture_enabled. |
| Critical alerts | 4 hours | 1 hour; 4 hours; 8 hours; 24 hours; 3 days; 1 week; No target (0) | Target response time for new alerts of that severity. | A tracking goal only, not an automated action. Key sla.critical_hours. |
| High alerts | 24 hours | same list | same | Key sla.high_hours. |
| Medium alerts | 1 week (168 hours) | same list | same | Key sla.medium_hours. |
| Low alerts | No target (0) | same list | same | Key sla.low_hours. |
Advanced rows:
| Label | Default | Allowed | What it does |
|---|---|---|---|
| Prompt samples: endpoint default | Capture samples by default | Capture samples by default (on); Skip samples by default (off) | Whether endpoints keep original prompt samples unless a tag rule says otherwise. |
| Always keep samples from endpoints tagged | empty | text (tags) | Endpoints with any of these tags always keep samples. |
| Never keep samples from endpoints tagged | empty | text (tags) | Endpoints with any of these tags never keep samples. If both lists match, "never" wins. |
| Allow sample capture without an encrypted connection (lab only) | Off | On, Off | Lets prompt-sample capture run over plain HTTP. Warning on On: prompt content could cross the network in cleartext. Leave off anywhere real. |
4. Artifact Analysis
"Whether analysts may have uploaded artifacts examined by external providers, which providers are used, and how much of an artifact is permitted to leave this appliance." Off by default. Reading or changing any row needs artifact.manage (Admin) as well as config.read or config.write; without it the server answers HTTP 403 "Artifact Analysis configuration requires the artifact.manage permission." YuviRed ships no provider credential: the account, entitlements and terms are yours. The section opens with a numbered note, "Turn this on in order", and the master switch is the last row on screen, under the heading Turn Artifact Analysis on.
| Label | Default | Allowed | What it does |
|---|---|---|---|
| Use MetaDefender Cloud | Off | On, Off | File and hash reputation by OPSWAT. Leave off until the key and the commercial-use attestation are saved. A Test Connection button under it sends the key and nothing else, and only works once Artifact Analysis is enabled. |
| MetaDefender Cloud API key | not set | secret | Your own key. Stored encrypted, never shown again. |
| Use Hive | Off | On, Off | AI-generated image detection. It has no hash lookup, so it is reached only when the policy permits an image to be submitted. Hive offers no connection test. |
| Hive image API key | not set | secret | Your own key; stored encrypted, never shown again. |
| Enable Artifact Analysis (last row) | Off | On, Off | Master switch. While off, no artifact route accepts work and no outbound provider call is made. Saving On shows a confirmation first that states exactly what the configuration then permits to leave the server. |
Advanced Provider Settings (collapsed):
| Label | Default | Allowed | What it does |
|---|---|---|---|
| Submission Policy | Hash only (Recommended) | Hash only (Recommended); Private upload allowed; Public upload allowed; Upload disabled: nothing leaves | The ceiling on what may be disclosed. Upload disabled: nothing leaves, not even hashes. Hash only: only supported hashes may be sent. Private upload allowed: file bytes only to a provider attested as private. Public upload allowed: file bytes to configured providers after an analyst authorizes. Warning when widening. |
| Ask the analyst to authorize each file submission | Off | On, Off | On: every analysis that would send a file waits for the analyst's confirmation. It never widens the policy ceiling. |
| Commercial-use attestation (each provider) | Off | On, Off | Your record that you may use the provider commercially. YuviRed records it and verifies nothing. |
| Private-processing attestation (each provider) | Off | On, Off | Your record that your agreement permits private processing. |
| Cached hash result freshness (MetaDefender) | 24 hours | 1 to 8760 | How long a stored hash result is reused. |
| Cached file result freshness (MetaDefender) | 168 hours | 1 to 8760 | How long a stored file result is reused. |
| Cached image result freshness (Hive) | 720 hours | 1 to 8760 | How long a stored image result is reused. |
| Largest accepted upload | 20 MB | 1 to 50 | Larger uploads are refused. The 50 MB ceiling is fixed in the web server and cannot be raised here. |
| Unauthorized artifact lifetime | 30 minutes | 5 to 120 | How long an uploaded, unauthorized artifact stays in encrypted staging. |
| Authorized artifact lifetime | 2 hours | 1 to 24 | How long an authorized artifact may stay in staging while providers work. |
| Staging area capacity | 200 MB | 50 to 4096 | New uploads are refused once staging holds more. |
| Analysis history retention | 365 days | 0 to 3650 | How long analysis history is kept; 0 keeps it forever. Shorter values delete history. |
5. Notifications
"Which events reach your team, and where they are delivered." A notification fires only for a NEW finding at or above the chosen severity, subject to the quiet window and the hourly cap. Use the panel Notifications: Test Send (save first) to check each channel.
| Label | Default | Allowed | What it does |
|---|---|---|---|
| Notify me about | High and above (Recommended) | Critical only; High and above (Recommended); Medium and above; Everything | Minimum severity of a new finding that triggers a notification. |
| Email alerts | On | On, Off | Send email for findings. Needs a mail server and recipients. |
| Send alerts to | empty | text (addresses) | Recipients. |
| Mail server | empty | text | SMTP host name. |
| Mail server port | 587 | 1 to 65535 | SMTP port. |
| Mail username | empty | text | |
| Mail password | not set | secret | |
| Encrypt mail connection (TLS) | On | On, Off | Turning it off sends mail unencrypted. |
| Send from address | noreply@yuvired.local | text | Change it to an address your mail server accepts. |
| Slack / Microsoft Teams alerts | On | On, Off | Chat channel switch. |
| Slack or Teams webhook URL | not set | secret | An incoming-webhook URL from Slack or Microsoft Teams. |
| Webhook feed | On | On, Off | Sends each notification as a JSON event to the URL below, for a SIEM (security information and event management system), SOAR or automation pipeline. |
| Webhook URL | not set | secret | |
| Daily summary email | Off | On, Off | Once-a-day summary of medium and low findings. High and critical findings always notify at once. |
| Send after | 07:00 UTC (Recommended) | 00:00 to 23:00 UTC | Hour for the daily summary, in UTC. |
| Don't repeat the same alert for | 6 hours (Recommended) | 1 hour; 6 hours; 12 hours; 24 hours | A repeat of the same finding on the same endpoint inside this window is not sent again. |
| Max notifications per hour | 20 (Recommended) | 10; 20; 50; 100 | Hourly cap. |
| System-health alert repeat interval (Advanced) | 24 hours (Recommended) | 6; 12; 24; 48 hours | How long a repeat page about the same degraded background job is held back. Recovery notices always send. |
Channel switches are independent of whether an address or URL is set, so you can pause a channel without clearing its credentials. The default of On for chat and webhook does nothing until a URL is saved.
6. Data & Retention
"How long telemetry, findings, and audit records are kept." The daily cleanup deletes data permanently.
| Label | Default | Allowed | What it does |
|---|---|---|---|
| Automatic Data Cleanup | On | On, Off | Runs the daily cleanup that enforces the windows below. Manual runs stay available. Warning on Off: old data accumulates without limit. |
| Raw telemetry | 30 days | whole days, 1 or more | Raw telemetry older than this is removed, but only for days already rolled up into daily summaries. |
| Daily summaries | 13 months | whole months, 1 or more | Daily summary data older than this is removed. |
| Closed alerts | 365 days | days, 0 = keep forever | Removes only closed alerts (resolved, false positive, suppressed) that are not part of an incident. |
| Audit log | 0 (keep forever) | days, 0 = keep forever | The record of who changed what. Setting a window permanently deletes older audit records. |
| AI agent signals | 120 days | days, 0 = keep forever | Raw AI-agent detection evidence rows. Only rows already processed are removed. |
| AI prompt/output events | 90 days | days, 0 = keep forever | Prompt and output events from the browser extension. |
| Decided adoption requests | 30 days | days, 0 = keep forever | Pending requests are never removed. |
| Original prompt samples | 24 hours (Recommended) | 4 hours; 24 hours; 3 days | How long an encrypted prompt sample is kept before its content is erased. |
Warning: lowering any window asks "Shorten ... retention? Records older than N ... will be permanently deleted at the next cleanup run". Moving from "keep forever" (0) to a number deletes everything older at the next run. There is no undo and Lite ships no backup command.
Other retention rows with no plain-language control (shown as raw rows under Advanced, Other settings): retention.report_file_days (30, generated report files are deleted after this many days; history entries stay; 1 to 3650), retention.report_storage_mb (2048; oldest stored files go first when exceeded; 100 to 51200), retention.ams_message_days (30), retention.ams_event_days (180), retention.ams_instance_days (90) (0 = keep forever), ams.activity_retention_days (365), ams.risk_snapshot_retention_days (180), and agent_boundary.retention_days (0 = keep forever). These raw rows delete data at their next run like the rows above, but the console shows no confirmation box when you shorten them.
Panel Telemetry Retention on the same page: Run now (dry run) shows what would be deleted; Run now deletes it. Alerts, evidence and the audit log are never purged by the telemetry job; they follow their own windows above.
7. Security & Access
"Connection security and sign-in requirements."
| Label | Default | Allowed | What it does | Notes |
|---|---|---|---|---|
| Agent Connection Security | Allow unencrypted connections | Allow unencrypted connections (off); Warn: log unencrypted connections (warn); Require encrypted connections (enforce) | How strictly agents and browser extensions must use TLS (Transport Layer Security) when reporting in. | Warning on Require: agents and extensions connecting without TLS are refused (HTTP 426) until reconfigured. Key security.tls_enforcement. Read live on every request. |
| Require multi-factor sign-in for | Admin | check boxes admin and analyst | Users with these roles must enroll an authenticator app at sign-in. View Only accounts can never be required. | Key mfa.required_roles. A user in a required role without MFA is sent to forced enrolment at next sign-in. There is no setting for the session length (8 hours) or the sign-in lockout rules; they are fixed (see roles-and-permissions.md). |
| Prevent password reuse | Last 5 passwords (Recommended) | Off; last 3; last 5; last 10; last 24 | Refuses a new password equal to any of the last N. | The server accepts 0 to 24. |
| Encryption Key Rotation Reminder | Annually (Recommended) | Quarterly (90); Twice a year (180); Annually (365); Every 2 years (730) | When the last recorded rotation is older than this, System Status marks it overdue and an operational alert is sent. | A reminder only. Rotation is never automatic. The server accepts 30 to 730 days. |
| Require per-device gateway credentials (Advanced) | Off | On, Off | AI-gateway traffic is accepted only from devices holding their own short-lived credential; the shared deployment token is refused. | Warning on On: devices without a credential lose AI-gateway ingest until they check in and receive one. Turning it off again is the immediate rollback. |
| Gateway credential lifetime (Advanced) | 8 hours (Recommended) | 1 hour; 8 hours; 24 hours | Lifetime of a device credential. | |
| Minimum agent version (Advanced) | empty (no minimum) | a version such as 1.8.2 (the box's example text still reads "e.g. 1.7.4") | Agents older than this are refused (HTTP 426) when they report in, until updated. | Warning when set. |
8. Integrations
"Single sign-on and external systems." Single sign-on uses OpenID Connect (OIDC). It stays inactive until the provider URL, client details and this server's public URL are all set.
| Label | Default | Allowed | What it does |
|---|---|---|---|
| Single Sign-On | Off | On, Off | Adds a Sign in with SSO button. |
| Login button label | empty | text | Text on the button. |
| Identity provider URL | empty | URL | Your identity provider's address. |
| Client ID | empty | text | |
| Client secret | not set | secret | |
| This appliance's public URL | empty | URL | The address users reach this console at. Required: SSO stays disabled until set. |
| Auto-create accounts on first SSO sign-in (Advanced) | Off | On, Off | New users get a read-only (View Only) account. Has no effect until approved email domains are set. |
| Approved email domains for auto-created accounts (Advanced) | empty | text | |
| Extra trusted identity-provider hosts (Advanced) | empty | text | Only for providers that serve sign-in endpoints from other host names. |
| Requested sign-in strength (OIDC acr_values) (Advanced) | empty | text | Optional authentication-strength request. |
| Require provider MFA for roles that require MFA (Advanced) | On | On, Off | Warning on Off: SSO users bypass this platform's MFA requirements entirely. |
The SIEM or webhook feed is configured under Notifications. The firewall blocklist feed token lives on the Network page. Setting up SSO against a real identity provider has not been verified in this release's documentation work; treat it as unverified until you test it with your provider.
9. System
"Platform identity and agent update behavior."
Present but inactive on YuviRed Lite. Every row below except the first two belongs to the Docker appliance edition. On a Lite server they can be switched and saved, and the audit log records the change, but they change nothing today:
- Automatic Agent Updates acts only on agents installed with an install command. Every agent on a Lite server is installed from the Add Computer package, so the switch has no agent to act on. To update an agent on Lite, run a new Add Computer download on that computer.
- Remote Browser Extension Wiring and the three advanced extension rows act only when this server serves the signed extension or a store listing exists. The Lite service does not serve the signed extension (the Windows service removes the signing key), so the setting answers "disabled" to every agent, and the Chrome Web Store ID is empty until a listing is approved. In this release nothing installs the extension into Chrome or Edge automatically; the working route is a manual "Load unpacked" (see extensions.md). Automatic installation from the store is planned for a later update.
- The two Staged rollout rows pace Automatic Agent Updates, so on Lite they have nothing to pace.
| Label | Default | Allowed | What it does | Notes |
|---|---|---|---|---|
| Organization Name | YuviRed | text | Used in notification subjects and reports. | |
| Privacy contact | empty | an email address or an https:// URL, up to 200 characters | Shown to the people your endpoints monitor, on the notice page. | The server refuses anything else. |
| Automatic Agent Updates | Off | On, Off | Agents installed with an install command update themselves, paced by the staged rollout settings. | Inactive on Lite (above). Warning on On: "Agents installed from a package, including every computer added with Add Computer on YuviRed Lite, are not changed." |
| Remote Browser Extension Wiring | Off | On, Off | Windows agents (1.7.12 or later) keep Chrome and Edge wired for the self-hosted extension: force-install policy, managed configuration, native messaging. | Inactive on Lite (above). On the appliance edition it writes machine policy as SYSTEM on every computer and is inert unless the server serves the signed extension. |
| Extension distribution mode (Advanced) | Self-hosted: appliance serves the signed .crx (managed / air-gapped) | Self-hosted; Chrome Web Store: unlisted/private (required for unmanaged endpoints) | How Chrome and Edge force-install the extension. | Inactive on Lite (above). Store mode requires agents to check in over HTTPS. |
| Chrome Web Store extension ID (Advanced) | empty | 32 letters a to p | Used in store mode; while empty, store-mode wiring stays inert. | Empty in this release. |
| Allow extension wiring over http agents (Advanced) | Off | On, Off | Permits wiring on computers whose agent checks in over plain HTTP. | Inactive on Lite (above). A network attacker can tamper with the instruction on plain HTTP regardless of this switch. Trusted networks only. |
| Staged rollout: target agent version (Advanced) | empty | a version | Inactive unless it matches the version bundled with the server. | Inactive on Lite (above). |
| Staged rollout: current stage (Advanced) | Canary: smallest group first (Recommended) | Canary; Early: wider group; Broad: whole fleet | Which cohort is unlocked for the target version. | Advance canary, early, broad by hand. |
Settings that interrupt monitoring, block activity, delete data, change access or affect licensing
"Confirmation" says whether the console asks you to confirm before saving.
| Setting | Effect | Class | Confirmation |
|---|---|---|---|
| Agent Connection Security = Require encrypted connections | Refuses agents and extensions that connect without TLS (HTTP 426). | Interrupts monitoring | Yes |
| Minimum agent version | Refuses older agents (HTTP 426) until updated. | Interrupts monitoring | Yes |
| Require per-device gateway credentials (Advanced) | Devices without a credential lose AI-gateway ingest. | Interrupts monitoring | Yes |
| Behavioral Detection Off; Agent Activity Tracking Off | Stops behavior analysis and telemetry linking; Agent Activity Tracking Off also stops new AI agents being seated against the AMS limit. | Interrupts monitoring | Yes |
| Detection mode (Advanced, AI & Agent Monitoring) | Only "Report only" works; the other choice is disabled. | Nothing | No |
| Prompt Protection presets Balanced, Strong, Strict, Maximum; any Block action; "Block" for content that cannot be inspected | Blocks prompts or attachments in the browser. | Blocks activity | The policy editor shows an impact summary before apply |
| Rogue Agent Activity Detection On | Raises Critical alerts (report only). | Alerts | Yes |
| Automatic Data Cleanup Off | Old data accumulates without limit. | Storage | Yes |
| Raw telemetry; Daily summaries; Closed alerts; Audit log; AI agent signals; AI prompt/output events; Decided adoption requests; Original prompt samples | Shortening any window permanently deletes older records at the next cleanup run. | Deletes data | Yes, when shortened |
| Analysis history retention (Artifact Analysis, Advanced) | Shorter values delete analysis history. | Deletes data | No |
Raw rows retention.report_file_days, retention.report_storage_mb, retention.ams_message_days, retention.ams_event_days, retention.ams_instance_days, ams.activity_retention_days, ams.risk_snapshot_retention_days, agent_boundary.retention_days | Delete older records at their next run. | Deletes data | No |
| Debug / Reset page (not a setting) | Clears selected data, cannot be undone. | Deletes data | Yes (on the page) |
| Require multi-factor sign-in for; Prevent password reuse; Single Sign-On and its rows | Change who can sign in and how. | Changes access | No |
| Require provider MFA for roles that require MFA = Off | SSO users bypass the MFA requirement. | Changes access | Yes |
| Auto-create accounts on first SSO sign-in | Creates View Only accounts for unknown people from the approved domains. | Changes access | No |
| Keep Original Prompt Samples On; Prompt Content Capture = Full content; Inspection Depth = Full | Store prompt content (encrypted). | Changes privacy | Yes |
| Allow sample capture without an encrypted connection (lab only) | Prompt content could cross the network in cleartext. | Changes privacy | Yes |
| Enable Artifact Analysis; Submission Policy widened | Can send hashes or files outside your network. | Changes privacy | Yes |
| Automatic Agent Updates; Remote Browser Extension Wiring and the extension rows; Staged rollout | Fleet-wide behavior on the appliance edition. Inactive on Lite (section 9). | Fleet-wide | Automatic Agent Updates: Yes; the others: No |
| License state | Not a setting: the server refuses any write to license state (HTTP 400, recorded in the audit log). Activate through the license screen (see licensing.md). | Affects licensing | Not applicable |
| Session length; sign-in lockout | Not settings: fixed at 8 hours, and 8 failures per 5 minutes per address plus 30 per hour per account. | Access | Not applicable |
Server-only and hidden settings
These cannot be changed from the console.
- Written by the system: the Diagnostics box shows read-only values, in this order: Behavioral analysis: last completed, last started, last pass; Learned behavior profiles; Recent behavioral findings (rolling buffer); Activity tracking: last completed, last started, rows scanned last cycle, progress marker; Observed-activity progress marker; Rogue Agent Activity: last completed, last started, last cycle stats; Daily summary: last sent (UTC date); Encryption key: last rotation recorded; Prompt-blocking policy version; Prompt Blocking (legacy): stored value and Legacy blocking rules: stored value (both superseded by Prompt Protection); Prompt Protection: active policy revision; Active license ID (managed via the License panel); EULA accepted: version, time and by (recorded at first-run setup; the server refuses edits); Gateway credential revocation counter.
- License state (
license.active_license_id): the server refuses writes from every role with HTTP 400 ("<key> is license state, not a configurable value. Activate or revoke a license through the License panel...") and records the attempt. - EULA acceptance record (
legal.*): refused with HTTP 400; written once at setup. - Encryption key rotation stamp (
security.encryption_key_rotated_at): written only by the rotation script, refused otherwise. - Hidden because no feature reads them:
ai.*(provider addresses, default provider, MCP servers),proxy.*,branding.logo_url,branding.accent_color,branding.theme_default,update.channel,update.auto_check,update.last_checked_at,logging.level,logging.json_format,retention.telemetry_days,ams.behavioral.peer_prior_strength. In particular the log level is not changeable from the console, and the console has no proxy setting. - Other keys with no plain-language row (shown as raw key rows under Advanced, Other settings): the internal progress and error markers
ams.correlation_*(last error, retry, quarantine, resume, stats), the Rogue Agent tuning valuesagent_boundary.*(interval, window, thresholds, family weights, retention, agents per tick), andgateway.allow_token_ingest_for_credentialed_endpoints. Do not change them unless YuviRed support asks you to. - Environment and installer settings (network mode, ports, certificate, the service account and similar) are set on the server itself (for example the
networkcommand that the Add Computer panel shows) or by the installer, not in the console. See installation.md. - A test-only Artifact Analysis provider has five settings that exist only in an isolated test stack; they are absent on a real install.