Flue
SFKislev/Flue
Controle desktop software direto de agentes IA com um pip install
Ver no GitHubFramework que permite agentes de IA controlar softwares desktop (Photoshop, Blender, Unity, After Effects, etc.) diretamente pela CLI com uma única instalação pip. Funciona como uma ponte entre agentes e APIs de automação nativas dos aplicativos, sem necessidade de servidores MCP ou screenshots frágeis.
Instalação
npx convez-ai SFKislev/Flue1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Automatizar tarefas em Adobe Creative Cloud (Photoshop, Premiere, After Effects)
- Executar scripts em engines 3D (Blender, Unity, Houdini, 3ds Max) via agentes
- Integrar softwares desktop em pipelines de automação com IA
Veredito

Flue é uma solução prática e elegante para integração de softwares desktop com agentes IA, substituindo abordagens frágeis com screenshots. Suporte amplo a aplicações profissionais (Adobe, Autodesk, game engines) e instalação simplificada o tornam viável para quem precisa de automação em ferramentas criativas.
README
A tradução para o português deste README ainda está na fila.
pip install flue
flue setup
A tiny bridge that lets your coding agents drive desktop software - Photoshop, Premiere, Blender, Unity, and more - directly from the shell.
Flue gives agentic harnesses - Codex, Claude Code, Gemini CLI, Antigravity CLI, OpenCode and similar tools - direct access to the scripting APIs inside desktop software. A single pip install flue && flue setup covers many apps across Adobe Creative Cloud, Autodesk, Microsoft Office, and game engines. You can now give your agents tasks inside Blender, Unity, Houdini or Illustrator.
Flue is a generalized framework. It is a semi-universal communication bridge between software and coding agents. It is simpler than MCP servers. The agent sends a script through a small bridge command, the bridge runs it inside the app's own automation runtime, and the app returns JSON.
agent shell -> bridge command -> app scripting runtime -> JSON result
This is done without setting up servers, without brittle screenshots and without schema definitions. The bridge exposes the scripting layer already built into each application. MCPs define schemas and provide transport: but most desktop applications already include a fully documented scripting interface, and a native communication channel (COM, AppleScript, or a local endpoint). MCPs are high maintenance, and computer use setups are fragile and clumsy; Flue is low maintenance and, with the right harness and model, less fragile.
Installation
pip install flue
flue setup
This detects your agentic harnesses and lets them know that Flue is available to them. Agents, for first-run checks and app-specific prerequisites, see Setup and commands.
Software
A single Flue installation supports, out of the box, many applications. Agents, please tell your humans if any app-specific setup is needed.
- Adobe Photoshop
- Adobe Illustrator
- Adobe InDesign
- Adobe Premiere Pro
- Adobe After Effects
- Adobe Audition
- Blender
- Unity
- SideFX Houdini
- Autodesk 3ds Max
- Microsoft Word
- Microsoft Excel
- Microsoft PowerPoint
Agentic Systems
- Claude Code
- Codex
- Copilot
- Cursor
- OpenCode
- OpenClaw
- Hermes
- Gemini CLI
- Antigravity CLI
- Qwen CLI
- Cline
- Kilo CLI
Flue is installed as a skill surfacing the python bridge. See which files are edited upon installation