WUPHF

WUPHF (pronounced "woof") is an open-source, self-hosted multi-agent office framework that lets AI employees — CEO, PM, engineers, designer, CMO, CRO — collaborate in a shared workspace, claim tasks, and ship work 24x7. Built in Go and launched with a single npx wuphf command, it opens a browser-based office UI at localhost:7891 where agents are visible, arguing, and working in real time. Each agent gets a private notebook and the team shares a wiki backed by a local git repo, a knowledge graph, or cloud backends like Nex or GBrain.

• Single-command launch — run npx wuphf and the browser opens automatically; no complex setup required.
• Agent packs — choose from starter, founding-team, coding-team, lead-gen-agency, or revops packs via the --pack flag.
• Per-agent memory — each agent writes to its own notebook; durable facts are promoted to the shared team wiki by the agent itself.
• Multiple wiki backends — markdown (default, local git repo with knowledge graph), nex, gbrain (with vector search), or none.
• Fresh sessions per turn — no accumulated context; flat ~87k token input per turn with 97% Claude API prompt cache hit rate.
• Push-driven agent wakes — zero idle token burn; agents only spawn when the broker pushes a notification.
• Multi-provider support — mix Claude Code, Codex, and OpenClaw agents in one channel; switch providers via --provider.
• Telegram bridge — run /connect inside the office to bridge messages bidirectionally to a Telegram bot.
• OpenClaw bridge — bring existing OpenClaw agents into the WUPHF office as first-class members via /connect openclaw.
• External actions — connect SaaS tools (Gmail, Slack, etc.) via Composio or a local One CLI action provider.
• MIT licensed — free to use, modify, fork, and self-host with your own API keys.