AI agents everywhere. Coordination nowhere.
We had agents running across three VPS instances with no unified visibility. Failures went undetected, costs were uncontrolled, and each agent operated as its own island with no shared state or task handoff.
Agents across 3 servers with no unified dashboard or health monitoring
Silent failures — agents crashed without anyone noticing for hours
No cost controls — a runaway agent could burn API budget overnight
No coordination — agents couldn’t share CRM data or hand off tasks
Self-hosted mission control for autonomous AI.
We built a coordination platform from raw Node.js — no frameworks, no npm dependencies. Fleet orchestration, real-time SSE dashboard, circuit breakers, and 18 MCP tools for seamless Claude Code integration.
Fleet Orchestration
Seven agents coordinated from a single dashboard with real-time SSE event streaming.
Circuit Breakers
Three-strike auto-trip with 15-minute recovery. Second fault escalates to human review.
Cost Tracking
Per-agent API spend with daily budgets, automatic Langfuse tracing, and monthly reports.
18 MCP Tools
CRM, observability, and task management tools exposed directly to Claude Code sessions.
Zero Dependencies
Raw Node.js HTTP server. No Express, no npm packages, no framework overhead to maintain.
Dual-Mode Auth
Supabase JWT for the dashboard. Bearer token for agent-to-agent machine communication.
