Claude Daemon

claude-daemon

Persistent daemon wrapper for Claude Code. Runs a self-improving team of AI agents with individual identities, tools, memory, and autonomous initiative. Full feature parity across Telegram, Discord, HTTP API, and CLI with seamless cross-platform session handover.

Features

• Multi-Agent C-Suite - 7 named agents (Johnny, Albert, Luna, Max, Penny, Jeremy, Sophie) with individual souls, roles, and domain ownership
• MCP Server Pool (41 servers) - Tiered shared pool of MCP servers available to all agents. Zero-config servers (fetch, git, time, memory, context7, playwright, ssh, tmux, etc.) are always on. Token-required servers auto-enable when you set the env var. Manage via /mcp list, /mcp enable, /mcp disable.
• Per-Agent Model Routing - Task-aware model selection: Albert/Luna/Max run Opus for coding and review. Johnny runs Sonnet for chat, Opus for planning. Penny/Jeremy/Sophie run Sonnet. Scheduled jobs run Haiku. Configurable per agent and per task type.
• Graceful Model Degradation - If Opus is rate-limited or unavailable, automatically falls back to Sonnet, then Haiku. Configurable fallback chain — no failed requests from transient capacity issues.
• Auto-Parallel Execution - Send multiple messages to the same agent — if busy, the daemon automatically spawns a parallel session. No /spawn needed. Also available as /spawn for explicit control.
• Fuzzy Agent Matching - Type @jony and it routes to johnny. Close-enough name typos resolved automatically with no error or fallback to the wrong agent.
• Per-Agent Channels - Bind Telegram groups or Discord channels to specific agents. Dedicated channels for Albert, Luna, etc.
• Cross-Platform Sessions - Start a conversation on Telegram, continue on Discord, pick up from CLI. Sessions follow the user, not the platform.
• Shared Brain for Claude Code CLI + macOS - Single markdown digest at ~/.config/claude-daemon/shared-brain.md exposing every daemon ag