Orcheo

Orcheo is a workflow orchestration platform designed for vibe coding — AI coding agents like Claude Code can start services, build workflows, and deploy them for you automatically. Read the full documentation for guides, API reference, and examples.

Note: This project is currently in Beta. Expect breaking changes as we iterate rapidly towards 1.0.

SIGIR Reviewers: See the Conversational Search Examples for step-by-step demos from basic RAG to production-ready search.

Why Orcheo?

Vibe-coding-first: Already using Claude Code, Codex CLI, or Cursor? You don't need to learn Orcheo. Install the agent skill and let your AI agent handle setup, workflow creation, and deployment.
Python-native: Workflows are Python code powered by LangGraph — no proprietary DSL to learn.
Backend-first: Run headless in production; the UI is optional.

Quick Start

Use the installation path that matches your setup:

Prerequisite: Docker Desktop/Engine must be installed to run the stack (orcheo install --start-stack).

macOS / Linux

bash <(curl -fsSL https://ai-colleagues.com/install.sh)

Windows PowerShell

irm https://ai-colleagues.com/install.ps1 | iex

SDK

uv tool install -U orcheo-sdk
orcheo install

macOS/Linux (non-interactive)

curl -fsSL https://ai-colleagues.com/install.sh | bash -s -- --yes --start-stack

Upgrade

orcheo install upgrade --yes

orcheo install syncs Docker stack assets into ~/.orcheo/stack (or
ORCHEO_STACK_DIR), updates .env with setup-selected values, and can start the
stack with Docker Compose. Setup will prompt for
VITE_ORCHEO_CHATKIT_DOMAIN_KEY; you can skip it and continue, but ChatKit UI
features will stay disabled until you set a valid key.

The fastest way to get started with workflow building is still the Agent Skill approach.

Note: Most AI coding agents (Claude Code, Codex CLI, Cursor) require a paid subscription. Free alternatives may exist but have not been tested with Orcheo.

1. Install the Orcheo Agent Skill

Add the Orcheo agent skill to your AI coding agent (Claude Code, Cursor, etc.) by following the installation instructions in the repo.

2. Let Your Agent Do the Work

Once installed, simply ask your agent to:

Set up Orcheo: "Set up Orcheo for local development"
Create workflows: "Create a workflow that monitors RSS feeds and sends Slack notifications"
Deploy workflows: "Deploy and schedule my workflow to run every hour"

Your AI agent will automatically:

Install dependencies
Start the backend server
Create and configure workflows
Handle authentication and deployment

That's it! Your agent handles the complexity while you focus on describing what you want your workflows to do.

Guides

Manual Setup Guide — Installation, CLI reference, authentication, and Canvas setup
Conversational Search Examples — Step-by-step demos from basic RAG to production-ready search

# Quick start: Run Demo 1 (no external services required)
uv sync --group examples
orcheo credential create openai_api_key --secret sk-your-key
python examples/conversational_search/demo_2_basic_rag/demo_2.py

Provider-aware chat models use provider-specific credential names in the active
vault by default. For example, openai:gpt-4.1 looks for openai_api_key,
while deepseek:deepseek-chat looks for deepseek_api_key.

Reference

SDK Reference — Python SDK for programmatic workflow execution
Environment Variables — Complete configuration reference

For Developers

Developer Guide — Repository layout, development environment, and custom nodes
Deployment Guide — Docker Compose and PostgreSQL deployment recipes
Custom Nodes and Tools — Extend Orcheo with your own integrations

Contributing

We welcome contributions from the community:

Share your extensions: Custom nodes, agent tools, and workflows that extend Orcheo's capabilities. See the Custom Nodes and Tools guide for how to create and load custom extensions.
How to contribute: Open an issue, submit a pull request, or start a discussion. You can also publish and share your extensions independently for others to install.

Citation

If you use Orcheo in your research, please cite it as:

@article{jiang2026orcheo,
  author       = {Jiang, Shaojie and Vakulenko, Svitlana and de Rijke, Maarten},
  title        = {Orcheo: A Modular Full-Stack Platform for Conversational Search},
  journal      = {arXiv preprint arXiv:2602.14710},
  year         = {2026}
}