Always-on AI that spawns coding agents from a WhatsApp text

What it does

A self-hosted “AI OS” that sits between your tools and your attention. It connects to 50-plus apps via MCP, monitors them for events like GitHub issues or Sentry alerts, and either handles them autonomously or surfaces them for approval. It can spawn Claude Code or Codex sessions, drive a browser, run terminal commands, and feed everything into a temporal knowledge graph so the next task starts with full context.

The interesting bit

The unusual part is the proactive posture: instead of waiting in a browser tab, it watches. You can fire off a task from WhatsApp, a Scratchpad note, or voice, and it will run a coding session on Railway while your laptop is closed. The personality system is a nice touch too—pick TARS’s dry efficiency or write your own, and it bleeds into every plan and message.

Key highlights

• Proactive monitoring across 50+ app connectors with webhook triggers, exposing 1000+ actions via a single MCP endpoint
• Delegates to Claude Code, Codex, browser agents, and terminal via a gateway that runs on Docker or Railway
• Four interfaces: voice hotkey, scratchpad (auto-picks up tasks within ~3 minutes), messaging, and chat dashboard
• Temporal knowledge graph memory that persists preferences, decisions, and context across sessions and tools
• Model-agnostic: bring your own Anthropic, OpenAI, or open-weight provider; fully self-hostable under AGPL 3.0

Caveats

• Requires Docker and 4 vCPU / 8 GB RAM minimum, so it is not lightweight
• Native Mac app is waitlisted; for now you self-host via Docker or Railway

Verdict

Worth a look if you want an autonomous, self-hosted coordinator that can actually run coding agents while you sleep. Skip it if you are looking for a lightweight chat wrapper or a fully polished SaaS with no setup.