Prometheus

AI workforce management for people who actually want to get work done.

I got tired of flashy AI tools that demo well but don't do anything useful. Every "agent platform" out there is either a glorified chatbot wrapper, a 47-step no-code workflow builder, or an over-engineered framework that needs a PhD to configure. None of them let me just set up AI workers and have them actually automate my work in a simple, straightforward way.

So I built Prometheus. It's a desktop app where you create AI employees, give them roles, tools, and knowledge, and they do real work — write files, browse the web, execute code, talk to each other, and delegate tasks autonomously. No workflow canvases, no drag-and-drop nonsense, no "prompt engineering studios." Just agents that work.

What it does

• AI Employees — Create agents with custom roles, system prompts, tools, and persistent memory
• Any LLM provider — OpenAI, Anthropic, Google, Mistral, Ollama (local), Vercel AI Gateway (50+ models), or your Claude Code subscription directly via CLI
• Real tools — Web search, file read/write, code execution, MCP server integrations
• Inter-agent communication — Employees delegate tasks to each other and collaborate autonomously via structured Agent Briefs
• Recurring tasks — Schedule agents to run hourly, daily, or weekly
• Knowledge system — Shared documents that agents can create, reference, and auto-update
• Persistent memory — Agents remember important facts across conversations
• File handling — Upload files, drag-and-drop attachments, inline image previews
• Departments & companies — Organize agents into teams, manage multiple orgs
• Desktop native — Electron app, fast, your data stays on your machine

Providers

Provider	Models	Notes
Vercel AI Gateway	50+ models (Claude, GPT, Gemini, Grok, DeepSeek, Llama, Qwen, more)	One API key for everything
OpenAI	GPT-4o, o1, o3
Anthropic	Claude Opus, Sonnet, Haiku
Google	Gemini 2.5 Pro & Flash
Mistral	Large, Medium, Small, Codestral
Ollama Cloud	DeepSeek 671B, Qwen 480B, more	Subscription-based
Ollama (Local)	Llama, Mistral, CodeLlama, anything	Free, offline
Claude Code (CLI)	Opus, Sonnet, Haiku	Uses your existing Claude subscription, no API key needed

Claude Code provider

Prometheus can use your existing Claude Code subscription as a provider — no API key, no extra charges. Enable "Claude Code (CLI)" in Settings, connect your account, and assign employees to it. Each agent runs as a Claude Code subprocess with full built-in tool access (file ops, web search, bash, MCP servers).

Requires the Claude Code CLI installed and authenticated.

Getting started

npm install
npm run dev

Run npm run dev from Terminal.app, not the VS Code terminal — the ELECTRON_RUN_AS_NODE env var breaks Electron.

For browser preview (mock API, no Electron):

npm run dev:web

1. Open the app
2. Go to Settings — add an API key or connect Claude Code
3. Hire your first employee
4. Start chatting

Tech stack

Electron 35 + Vite 8 + React 19 + TypeScript + Tailwind CSS v4 + Zustand

License

MIT