Skip to main content

Welcome to Jetty! 🚀

Jetty is an agentic evaluation platform for AI models that lets you build powerful workflows using simple JSON configurations. Create complex AI pipelines, run evaluations, and scale your AI experiments with ease.


🏁 Get Started

📖 Quick Start

New to Jetty? Get up and running in minutes.

Start Here →

🧠 Step Library

Explore the building blocks of Jetty workflows.

Learn More →

📚 Guides & Tutorials

Practical examples for common use cases.

Browse Guides →

🤖 Agents & MCP

Use Jetty from your coding agent or editor.

Get Started →


⚡ Quick Example

Here's a simple workflow that analyzes text with an AI model:

{
"init_params": {
"user_prompt": "Explain quantum computing in simple terms"
},
"step_configs": {
"analyze": {
"activity": "litellm_chat",
"model": "gpt-4",
"prompt_path": "init_params.user_prompt"
}
},
"steps": ["analyze"]
}

How it works:

  • init_params.user_prompt provides the input text
  • The analyze step uses path expressions to access that input
  • LiteLLM processes the prompt with GPT-4 and returns results

🔗 Key Features

Path Expressions - Connect data between components using simple dot notation:

  • Step outputs: step_name.outputs.result.text
  • Input parameters: init_params.prompt
  • Array access: images[0].path or results[*].score

JSON Configuration - Define complex workflows without code

Built-in Error Handling - Robust path resolution with fallbacks

Scalable Execution - Run workflows locally or in the cloud