Skip to main content

Admin Dashboard

Overview​

The admin dashboard provides system administrators with tools for monitoring, prompt management, and user administration. Access requires the admin role.

Location: /admin

Admin Roles​

Only users with admin privileges can access admin features:

RoleAdmin Access
Owner✅ Full
Admin✅ Full
Member❌ No
Viewer❌ No

Metrics Dashboard​

Location: /admin/metrics

User Growth​

Track user and organization growth:

  • Total users over time
  • New registrations by period
  • Organization creation trends
  • Active users (daily/weekly/monthly)

AI Usage​

Monitor AI generation consumption:

  • Total generations by type
  • Analysis vs. chat vs. report usage
  • Usage trends over time
  • Quota utilization across organizations

Feature Adoption​

Understand how features are used:

  • Evidence uploads per investigation
  • Analysis generation frequency
  • Report export counts
  • Template usage rates

Prompt Management​

Location: /admin/prompts

What Are Prompts?​

AI prompts are the instructions that guide Claude's analysis and report generation. Inqura stores prompts in the database for easy iteration without code changes.

Prompt Categories​

CategoryPurpose
Default FrameworkEvidence evaluation guide
Analysis PromptsQuestion, topic, gap, and summary analysis
Report Section PromptsExecutive summary, findings, methodology, etc.

Prompt Types​

TypeDescription
default_evaluation_guide617-line evidence evaluation framework
analysis_questionSingle question analysis
analysis_summaryOverall investigation summary
analysis_gapEvidence gap identification
report_executive_summaryReport executive summary section
report_backgroundReport background section
report_methodologyReport methodology section
report_findingsReport findings section
report_conclusionsReport conclusions section
report_recommendationsReport recommendations section

Template Variables​

Prompts use placeholders replaced at runtime:

VariableReplaced With
{{investigation_title}}Project title
{{focus_statement}}Investigation focus/scope statement
{{question_text}}Specific question being analyzed
{{evidence_list}}Retrieved evidence chunks
{{evidence_summary}}Summary of all evidence
{{questions_list}}All investigation questions
{{topics_list}}All topics with descriptions
{{investigator_direction}}User-provided guidance for this analysis
{{evaluation_framework}}The evidence evaluation framework
{{investigation_context}}Combined context about the investigation

Editing Prompts​

  1. Navigate to Admin → Prompts
  2. Find the prompt to edit
  3. Modify system prompt and/or user template
  4. Click Save
  5. Changes take effect immediately

Prompt Versioning​

Each prompt tracks:

  • Version number (auto-incremented on save)
  • Active/inactive status
  • Created and updated timestamps

Only active prompts are used; deactivate to disable without deleting.

Best Practices​

  1. Test changes - Try analysis after prompt changes
  2. Iterate incrementally - Small changes, then evaluate
  3. Document changes - Note what changed and why
  4. Keep backups - Copy old version before major changes

User Management​

Location: /admin/users

User List​

View all users in the workspace:

  • Name and email
  • Role assignment
  • Organization membership
  • Account status

Role Management​

Change user roles:

  1. Find user in list
  2. Select new role from dropdown
  3. Changes apply immediately

User Actions​

Available admin actions:

  • View user details
  • Change role
  • View activity (audit log)
  • Remove from organization

Usage Monitoring​

Location: /admin/usage

Organization Usage​

Track resource consumption:

  • Storage used vs. limit
  • AI generations used vs. limit
  • Team member count

Billing Status​

View subscription details:

  • Current plan
  • Billing period
  • Payment status

Overage Tracking​

Monitor soft cap overages:

  • AI generations over limit
  • Storage approaching limits
  • Alerts for action needed

Audit Health​

Location: /admin/audit-health

Audit Log Status​

Monitor audit logging system:

  • Recent audit entries
  • Any logging failures
  • Storage/retention status

Compliance Verification​

Verify compliance requirements:

  • All actions being logged
  • Log integrity checks
  • Retention policy adherence

Security Considerations​

Admin Access Logging​

All admin actions are logged:

  • Who accessed admin features
  • What changes were made
  • When actions occurred

Prompt Change Auditing​

Prompt modifications are tracked:

  • Previous version preserved
  • Who made the change
  • Timestamp of change

Principle of Least Privilege​

  • Limit admin roles to necessary personnel
  • Review admin access periodically
  • Remove admin access when role changes

AI Provider — AWS Bedrock (only)​

AI provider toggle removed 2026-05-18 (NQU-783 / ADR-0002). AWS Bedrock is now the sole inference path; there is no runtime provider switch, and /admin/settings no longer exposes a provider card.

All AI calls route through AWS Bedrock using the ECS task-role credentials. The former toggle let admins switch to an Anthropic-direct API path; it was a Bedrock-throttling escape valve (NQU-728). It was removed because, for licensed/sovereign-deployment customers, a second credential surface + second data egress path + second compliance review is a negative feature. See ADR-0002 for the decision.

Bedrock resilience (replaces the Anthropic-direct toggle)​

Resilience is now automatic — no operator action — via two mechanisms inside lib/ai/client.ts:

  1. Multi-region fallback. Primary region us-east-1; fallback us-west-2 (set via BEDROCK_FALLBACK_REGION env). On a retryable failure from the primary region (throttle, 5xx, connection), the SDK call retries on the fallback region inside the same limiter slot. Customer-greenfield installs leave the env unset to opt out (single-region IAM allowlist).
  2. Throttle shedder. A 60-second sliding-window count of RateLimitError + ThrottlingException events shared across all Bedrock limiters. When >3 throttles register in the window, quality-check calls reject pre-flight (QualityShedError) so primary analysis traffic keeps the available quota. Primary calls never shed. Emergency kill switch: BEDROCK_SHEDDER_DISABLED=true env upsert.

Live state visible via GET /api/admin/queue-stats — the shedder object reports hot, recentThrottles, fallbackRegion, disabled. Runbook for sustained-hot / persistent fallback is in docs/admin/ops/incident-response.md §3.