wuphf

Multi-agent collaboration office for AI coding assistants

TLDR

Launch the WUPHF office and open the web UI

$ npx wuphf

Run with a specific agent provider

$ npx wuphf --provider [codex]

Use the tmux-based terminal UI instead of the web interface

$ npx wuphf --tui

Pick a predefined agent team pack

$ npx wuphf --pack [coding-team]

Choose a memory backend

$ npx wuphf --memory-backend [markdown]

Start a one-on-one session with a specific agent

$ npx wuphf --1o1 [ceo]

Run without opening the browser automatically

$ npx wuphf --no-open

Use a custom web UI port

$ npx wuphf --web-port [7891]

wuphf is a "Slack for AI employees" — a multi-agent collaboration layer that spawns a small office of role-based agents (CEO, PM, engineers, designer, CMO, CRO) sharing a common memory store and a chat-style channel. Instead of running each AI assistant in isolation, wuphf lets them see each other's messages, claim tasks, and ship work together while preserving context across sessions.Under the hood wuphf delegates each agent to an existing CLI such as claude-code, codex, or openclaw, and exposes a web UI by default at http://localhost:7891 with the team roster, a shared channel, and a message composer. A --tui mode renders the same office inside a tmux session for terminal-only environments.Memory is pluggable via --memory-backend: the markdown backend keeps a human-readable Markdown brain on disk, while nex and gbrain integrate with hosted services. Agent rosters are described by the --pack option, and custom packs can be defined in configuration.

PARAMETERS

--memory-backend name

Select the shared-memory backend: nex, gbrain, markdown (default), or none.

--pack name

Choose a preconfigured agent team: starter, founding-team, coding-team, lead-gen-agency, or revops.

--provider name

LLM CLI to drive each agent: claude-code (default), codex, or openclaw.

--tui

Use the tmux-based terminal UI instead of the web UI.

--no-nex

Disable the Nex backend while keeping local integrations.

--no-open

Do not open the default browser when the web UI starts.

--opus-ceo

Upgrade the CEO agent from Sonnet to Opus.

--collab

Enable collaborative mode (default).

--unsafe

Bypass agent permission checks; intended only for local experimentation.

--web-port N

Web UI port (default 7891).

--1o1 [agent]

Start a one-on-one chat with the CEO, or with the named agent if given.

init

Run initial setup and write a configuration file.

shred

Terminate the active office session and clean up local state.

CONFIGURATION

After wuphf init, configuration is stored locally and read on each launch. Required runtime dependencies are node / npx, python3, curl, tmux (for --tui mode), and at least one supported LLM CLI on PATH (claude, codex, or openclaw). API keys are taken from the underlying CLI's own configuration — wuphf does not store them itself.

CAVEATS

wuphf is an early-stage orchestration layer; flag names and pack contents change between releases. The --unsafe flag disables per-agent permission checks and should never be used outside isolated, throw-away environments. Because agents call external LLM CLIs, costs and rate limits depend on the provider account behind each CLI.

HISTORY

wuphf is developed by Nex and published at github.com/nex-crm/wuphf. The name is a nod to the fictional notification service from the TV show *Parks and Recreation*. It is distributed via npm and runs without a global install through npx wuphf.