GEPA Prompt Optimizer

Multi-objective LLM optimization

Setup Required
Task Configuration
Define your optimization task and objectives

Describe what you want the AI to accomplish

Configure your API key to use this model

Training Examples
Provide input-output pairs for optimization
1
Quick Tips

• Click Settings to configure cloud APIs or local Ollama

• Provide at least 2-3 diverse training examples

• Be specific in your task description

• GEPA optimizes for multiple objectives simultaneously

• Optimization typically takes 30-60 seconds

Local Models with Ollama
Run models locally on your machine — free and private
1

Install Ollama

Download and install Ollama from ollama.com/download

Or install via command line:

curl -fsSL https://ollama.com/install.sh | sh
2

Download a Model

Open a terminal and pull a model. We recommend starting with Llama 3.2:

ollama pull llama3.2

Other popular models: mistral, gemma2, llama3.1, deepseek-r1

3

Start Ollama Server

Run the Ollama server (it may already be running after installation):

ollama serve

The server runs on http://localhost:11434 by default.

4

Enable in Settings

Click the API Keys button above and enable Ollama. Use the Test Connection button to verify it's working.

Remote Access (Deployed Sites)

If you're using this site from a deployed URL (not localhost), you need to expose your local Ollama server using a tunnel. No account required!

Quick Setup (One Command)

Install and run Cloudflare Tunnel — works instantly, no signup needed:

brew install cloudflared && cloudflared tunnel --url http://localhost:11434

Copy the generated URL (e.g., https://random-words.trycloudflare.com) and paste it in Settings → Local (Ollama) → Server URL

Alternative: Linux/Windows

Download cloudflared from Cloudflare Downloads, then run:

cloudflared tunnel --url http://localhost:11434

System Requirements

  • • macOS: macOS 11 Big Sur or later
  • • Linux: Most distributions with glibc 2.31+
  • • Windows: Windows 10 or later (via WSL2 or native)
  • • RAM: 8GB minimum, 16GB+ recommended for larger models
  • • GPU: Optional but recommended for faster inference

No optimization yet

Configure your task and provide training examples, then run the GEPA optimizer to generate an optimized prompt.