How omnara-ai/omnara works
Project type and purpose
• Type: Full-stack, multi-platform orchestration system for AI “agents” with CLI, mobile, web, integrations, and backend services.
• Purpose: Provide a unified dashboard (web & mobile) and CLI wrapper to monitor, interact with, and route messages between AI agents (Claude Code, Codex CLI, n8n workflows, GitHub Actions, etc.).
File structure
• Top-level dirs:
• apps/mobile (React Native)
• apps/web (React/Vite)
• apps/packages/shared (design tokens, UI theme)
• src/backend (Python API service)
• src/relay_server (WebSocket gateway)
• src/servers/api & src/servers/mcp (auxiliary microprotocol servers)
• src/omnara (CLI entrypoint, SDK)
• src/integrations (n8n nodes, GitHub Action, headless wrappers)
• src/shared (database schemas, Alembic migrations, auth & config)
• infrastructure/docker & scripts
• Architectural hints:
• MVC/API service pattern in src/backend
• WebSocket relay for real-time sync
• Separate Notification service (push/email/SMS via Twilio & Supabase)
• CI/CD workflows and Docker Compose for local dev
README insights
• “Your AI Agents” → API → DB → Notification → Mobile/Web → back to API → Agents.
• Mentions Supabase for auth & real-time database sync in frontends.
• CLI (“omnara”) uses three modes (standard, headless, serve) that hook into the same backend.
Detail
• Drill into the WebSocket relay path, the Supabase real-time sync pipeline, and the notification triggers.
• Where shared modules (src/shared) are imported by multiple services.
• Build/deployment artifacts (Docker Compose) as a footnote or separate “Infrastructure” layer.