Claw-Empire dresses multi-agent orchestration up as a virtual software company: the user plays CEO, and AI coding agents become employees who work across departments, attend meetings, produce deliverables, and earn XP, all rendered in an animated pixel-art office (PixiJS on a React frontend, Express plus SQLite behind it). It runs local-first — agents, tasks, tokens, and settings live in a local SQLite database with no cloud dependency, and OAuth and messenger tokens are encrypted at rest.
Agents can be backed by three provider paths: local coding CLIs (Claude Code, Codex CLI, Gemini CLI, OpenCode, Kimi Code, GitHub Copilot, Antigravity), OAuth-connected accounts, or direct LLM API keys (OpenAI, Anthropic, Google, Ollama, OpenRouter, and others). Work flows through a Kanban board with a full task lifecycle, and each agent operates in an isolated git worktree whose branch is merged only on CEO approval. Six "office pack" profiles retarget the company at different jobs — software development, structured reports, web research with fact checking, novel writing, video pre-production, and roleplay — each with its own department topology and QA rules.
Direction comes in through a CEO chat: messages prefixed with $ are treated as directives that get delegated to planning and the relevant departments, optionally after an AI-run team-leader meeting with generated minutes. Directives can also arrive from messaging apps — Telegram, WhatsApp, Discord, Slack, Google Chat, Signal, iMessage — via built-in channel sessions or an optional OpenClaw bridge. Other notable pieces include a library of 600+ assignable skills with custom skill upload, real-time WebSocket status updates, completion reports with history, and PowerPoint export of minutes and reports.
