mission-control

Monitors 20+ distributed AI agents simultaneously through a centralized dashboard, implementing heartbeat-based liveness detection via WebSocket connections to OpenClaw Gateway instances. Uses Server-Sent Events (SSE) for real-time status updates and smart polling that automatically pauses during active connections to reduce overhead. Tracks session state, agent spawn control, and connection health across multiple gateway instances without requiring external message brokers.

Provides a six-stage Kanban board (inbox → backlog → todo → in-progress → review → done) with drag-and-drop task movement, priority level assignment, and agent-to-task binding. Implements optimistic UI updates via Zustand state management with SQLite persistence, allowing teams to coordinate multi-agent work without external workflow engines. Task state transitions trigger webhook events and can be assigned to specific agents with capacity tracking.

Implements the dashboard UI using Next.js 16 App Router for server-side rendering and incremental static regeneration; provides backend API endpoints via Next.js API routes (no separate backend server required). Uses React 19 concurrent rendering for responsive UI updates; implements middleware for authentication and request logging. Server components reduce JavaScript bundle size; client components use Zustand for state management.

Manages client-side application state (UI panels, filters, user preferences, task list) using Zustand 5 with minimal boilerplate; implements optimistic updates for task drag-and-drop and form submissions that revert on server error. Stores state in memory with optional localStorage persistence for user preferences. Zustand's subscription model enables fine-grained reactivity without Redux boilerplate.

Implements dashboard UI styling using Tailwind CSS 3.4 utility classes for responsive design across desktop, tablet, and mobile viewports. Uses Tailwind's dark mode support for theme switching; implements custom color schemes for agent status indicators and cost visualization. Tailwind's JIT compiler generates only used styles, minimizing CSS bundle size.

Visualizes token usage trends, cost breakdowns, and agent metrics using Recharts 3 interactive charts (line charts for trends, bar charts for comparisons, pie charts for provider breakdown). Charts are responsive and support hover tooltips, legend toggling, and drill-down interactions. Data is sourced from SQLite time-series buckets; charts update in real-time as new metrics arrive.

Streams live agent activity events to the dashboard via WebSocket connections and Server-Sent Events, displaying a chronological feed of agent actions, task completions, and system events. Implements smart polling that detects active connections and pauses database queries to reduce load; uses better-sqlite3 WAL mode to support concurrent reads while events are being written. Provides both push-based (WebSocket) and pull-based (SSE) delivery mechanisms for resilience.

Aggregates token consumption metrics across multiple AI providers (Anthropic, OpenAI, OpenRouter, Ollama) with per-model breakdowns and trend visualization using Recharts. Stores token counts and pricing data in SQLite with time-series bucketing for efficient querying; calculates running costs based on provider-specific pricing models. Provides dashboard panels for cost trends, per-agent spending, and model-specific analytics without requiring external analytics platforms.

Implements three permission tiers (viewer, operator, admin) with session-based authentication using scrypt password hashing and optional Google OAuth integration with approval workflows. Supports both session tokens and API keys for programmatic access; stores credentials in SQLite with secure hashing. Admin tier can approve new OAuth users, manage API keys, and modify system configuration; operator tier can dispatch tasks and monitor agents; viewer tier has read-only access.

Sends webhook events to external systems on task state transitions, agent status changes, and cost threshold breaches; tracks delivery status (pending, delivered, failed) in SQLite with retry attempts and cooldown periods. Implements exponential backoff for failed deliveries and stores webhook payloads for debugging. Supports multiple webhook endpoints with per-endpoint filtering rules and delivery history visualization.

Integrates with 1Password CLI (op command) to retrieve and inject secrets into agent configurations and webhook payloads at runtime. Stores 1Password vault references (not actual secrets) in SQLite; fetches secrets on-demand via subprocess calls to the op CLI. Supports secret rotation without redeploying agents and provides audit trails through 1Password's native logging.

Connects to multiple OpenClaw Gateway instances simultaneously, aggregating agent status, task assignments, and event streams from distributed deployments into a single dashboard. Implements per-gateway connection pooling and failover logic; stores gateway configurations in SQLite with health checks. Enables task dispatch to agents across different gateways and provides unified monitoring across geographically distributed or logically separated agent clusters.

Defines alert rules based on thresholds (e.g., agent offline >5 minutes, cost spike >$100/hour, task failure rate >10%) with configurable cooldown periods to prevent alert fatigue. Stores rules in SQLite and evaluates them against real-time metrics; triggers webhooks or notifications when thresholds are breached. Implements exponential backoff for repeated alerts on the same condition to reduce noise.

Uses SQLite with Write-Ahead Logging (WAL) mode enabled via better-sqlite3 native bindings to provide persistent storage for all application state (agents, tasks, users, webhooks, metrics) with concurrent read access. Implements database migrations for schema evolution; stores all data in a single .db file without requiring external database servers. WAL mode allows readers to access data while writes are in progress, enabling real-time monitoring without blocking.