Skip to main content

What is the web console?

The Agent Sentinel Web Console is a modern web application for monitoring and managing your AI agent fleet. It provides:
  • Real-time visibility into agent behavior
  • Human-in-the-loop approval workflows
  • Policy enforcement monitoring
  • Cost tracking and analytics
  • Compliance audit trails
  • Agent discovery and management

Key features

Real-time monitoring

  • Live updates: WebSocket-powered real-time data streaming across every list view
  • Notifications: Popup alerts for critical events
  • Connection indicator: Visual status of real-time connection (REALTIME / POLLING)

Intervention tracking

  • Policy enforcement: See what Agent Sentinel blocks in real-time
  • Cost prevention: Track USD saved by interventions
  • Risk analysis: Monitor critical/high-risk blocks
  • Blast radius: Understand impact of prevented actions

Approval workflows

  • Inbox: Review pending approval requests
  • Priority triage: Critical approvals highlighted with countdown
  • One-click decisions: Approve/reject with optional notes
  • Info requests: Ask agents for additional context
  • Decision history: Full audit trail of all decisions

Run telemetry

  • Execution history: Complete log of agent runs
  • Action sequences: Step-by-step view of actions within runs
  • Replay mode: Zero-cost simulation of past runs
  • Determinism analysis: Identify non-deterministic behavior

Activity ledger

  • Complete audit trail: Every action logged across all agents
  • Advanced filtering: Search by outcome, action, agent, cost, date
  • Export capabilities: JSONL, JSON, CSV formats
  • Compliance metadata: Approval status, decision rationale

Agent discovery

  • Automatic registration: Agents self-register on first use
  • Performance metrics: Success rates, cost per run, intervention rates
  • Activity tracking: 24h, 7d, 30d breakdowns
  • Recent runs: Link to agent’s execution history

Cost analytics

  • 30-day consumption: Bar chart of daily costs
  • Cost breakdown: By action type, agent, run
  • Average metrics: Unit cost, execution time
  • Budget tracking: Compare against policy budgets

Settings & configuration

  • Policy management: Create / edit / enable policies, with prose-to-IR compilation, version diff, and one-click rollback
  • Evidence graph viewer: ReactFlow DAG of evidence requirements + grounding rules
  • API key generation: Secure token management
  • Notification channels: Slack, Email, Webhooks

Design philosophy

The console features a technical, industrial aesthetic:
  • Typography: Space Grotesk (headings), DM Sans (body), JetBrains Mono (code)
  • Colors: Dark mode default with orange/red accents (FF3300)
  • Layout: Information-dense with progressive disclosure
  • Components: Border-emphasized, uppercase labels, heavy fonts
Signed-in users land on Interventions (/interventions); there is no standalone “dashboard” route. Each top-level page is a self-contained view in the sidebar:
  1. Interventions (/interventions) — Policy enforcement monitoring
  2. Approvals (/approvals) — Human-in-the-loop workflow
  3. Kill Switch (/kill-switch) — Emergency halt for agents / runs / missions
  4. Runs (/runs) — Execution history (with built-in determinism + simulation tabs per run)
  5. Replay (/replay) — Org-wide ledger of local-replay submissions
  6. Activity Ledger (/ledger) — Complete audit trail
  7. Agents (/agents) — Agent discovery and management
  8. Evals (/evals) — Benchmark scorecards + adversarial scenario generator
  9. Analytics (/analytics) — Cost and performance dashboards
  10. Compliance (/compliance) — EU AI Act oversight summary
  11. MCP (/mcp) — Model Context Protocol explorer
  12. Settings (/settings) — Policies (with version diff + rollback), API keys, notifications

Quick Start wizard

New users see a 4-step onboarding wizard:
  1. Welcome: Feature overview and introduction
  2. API Key: Generate API key for SDK integration
  3. SDK Setup: Python/JavaScript code snippets
  4. Complete: Confirmation and next steps
The wizard auto-dismisses after first view but can be manually reopened.

Multi-tenancy

The console supports organization-level multi-tenancy:
  • Organizations: Team-level data isolation
  • Projects: Logical grouping within organizations (future)
  • Environments: Dev/staging/prod separation (future)
Use the sidebar selector to switch between organizations/projects/environments (disabled in current production build).

Authentication

  • Single sign-on: Secure user authentication
  • JWT tokens: Bearer token authentication for API calls
  • API keys: Long-lived tokens for SDK integration
  • Session management: Automatic token refresh

Real-time vs polling

The console uses a hybrid approach:
  • Primary: WebSocket real-time updates
  • Fallback: Polling every 30 seconds if WebSocket disconnects
  • Status indicator: Shows “REALTIME” or “POLLING” mode

Best practices

Enable real-time updates: WebSocket connections provide instant updates - ensure they’re not blocked by firewalls.
Use filters effectively: All major views support filtering - use them to focus on relevant data.
Review approvals regularly: Pending approvals expire - check the Approvals page frequently or enable notifications.
Protect API keys: Generated API keys are shown only once - copy and store them securely before closing the modal.

Getting started

  1. Sign up at console.agentsentinel.dev
  2. Complete the Quick Start wizard
  3. Generate an API key
  4. Install SDK: pip install agent-sentinel
  5. Configure SDK with your API key
  6. Start instrumenting your agents
  7. View telemetry in real-time

See also