Claude Code, running 24/7.

Agentic infrastructure for Claude Code. Email triggers, scheduled agents, local memory — self-hosted on your hardware, your data stays yours.

• Email-triggered agents — forward a task, it runs, it replies
• Cron jobs that run Claude Code — scheduled work without you
• Skill library — learned patterns persist across sessions and projects
• Local persistent memory — SQLite, no cloud account, no data leaves your machine

Recipes

Recipe	What it does	Setup
Memory persistence	Claude remembers Monday on Friday	5 min
Email-triggered agent	Forward an email — agent executes the task and replies	15 min
SR&ED diary	Auto-generates diary from git commits + terminal history	coming soon
Morning briefing	Synthesized project status email every morning	coming soon
Cron agent	Run Claude Code on a schedule, unattended	coming soon

Start with Memory persistence — standalone, 5 minutes, the foundation everything else builds on.

---

Architecture

Four Node.js processes, each with one responsibility:

mcp-server.js   (port 9003)  — MCP tools: memory, skills, email, calendar
orchestrator.js (port 9009)  — Cron scheduler + email trigger → Claude Code agent
server.js       (port 9000)  — Admin UI, terminal, SpecKit dev workflow
llm-proxy.js    (port 9001)  — OpenAI-compatible LLM proxy (Claude or Gemini)

The recipes are standalone extractions — run them independently or assemble into the full system. The full system has been running unattended across four active businesses for months.