Running Prompts

Learn how to run individual prompts with AI models using Umwelten.

Overview

Umwelten provides a powerful way to execute single prompts with AI models using the run command. This is ideal for:

• One-off tasks: Quick questions, analysis, or content generation
• Batch processing: Running the same prompt across multiple files
• Testing and evaluation: Systematic model testing and comparison
• Automation: Scripted interactions with AI models

The run command supports multiple providers, file attachments, and various configuration options.

Basic Usage

Run a simple prompt with any model:

umwelten run --provider ollama --model gemma3:latest "Explain the concept of quantum entanglement."

Provider-Specific Examples

Google Models

# Gemini 2.0 Flash (fast and cost-effective)
umwelten run --provider google --model gemini-2.0-flash "Write a summary of renewable energy trends in 2024"

# Gemini 2.5 Pro (highest quality)
umwelten run --provider google --model gemini-2.5-pro-exp-03-25 "Analyze the implications of artificial general intelligence"

Ollama Models (Local)

# Gemma3 models
umwelten run --provider ollama --model gemma3:12b "Compare Python and JavaScript for web development"

# Code-specific models
umwelten run --provider ollama --model codestral:latest "Write a function to implement binary search in TypeScript"

# Vision models
umwelten run --provider ollama --model qwen2.5vl:latest "Describe this image" --file ./photo.jpg

OpenRouter Models

# GPT-4o (premium quality)
umwelten run --provider openrouter --model openai/gpt-4o "Provide a detailed business analysis of the current AI market"

# Claude 3.7 Sonnet
umwelten run --provider openrouter --model anthropic/claude-3.7-sonnet:thinking "Analyze the ethical implications of AI in healthcare"

LM Studio (Local)

# Local model (ensure LM Studio server is running)
umwelten run --provider lmstudio --model mistralai/devstral-small-2505 "Help me debug this Python code"

Advanced Prompt Configuration

System Messages

Set the AI's role and behavior:

# Technical expert role
umwelten run \
  --provider google --model gemini-2.0-flash \
  --system "You are a senior software architect with expertise in distributed systems" \
  "Design a scalable microservices architecture for an e-commerce platform"

# Creative writing role
umwelten run \
  --provider ollama --model gemma3:27b \
  --system "You are a creative writer specializing in science fiction short stories" \
  "Write a story about first contact with an alien civilization"

Temperature Control

Adjust creativity and randomness:

# Very focused and deterministic (0.0-0.3)
umwelten run \
  --provider google --model gemini-2.0-flash \
  --temperature 0.1 \
  "Write technical documentation for this API endpoint"

# Balanced creativity (0.4-0.7)
umwelten run \
  --provider ollama --model gemma3:12b \
  --temperature 0.6 \
  "Brainstorm innovative solutions for urban transportation"

# Highly creative (0.8-2.0)
umwelten run \
  --provider google --model gemini-2.0-flash \
  --temperature 1.5 \
  "Write an abstract poem about the nature of consciousness"

Timeout Settings

Set appropriate timeouts for different types of prompts:

# Quick responses (default: 30 seconds)
umwelten run --provider ollama --model gemma3:12b "What is 2+2?"

# Complex analysis (longer timeout)
umwelten run \
  --provider google --model gemini-2.5-pro-exp-03-25 \
  --timeout 60000 \
  "Provide a comprehensive analysis of global climate change impacts"

# Extended processing (very long timeout)
umwelten run \
  --provider openrouter --model openai/gpt-4o \
  --timeout 120000 \
  "Write a detailed business plan for a sustainable technology startup"

File Attachments

Supported File Types

• Images: JPG, PNG, WebP, GIF
• Documents: PDF
• Text: TXT, MD (depending on model)

Image Analysis

# Basic image description
umwelten run \
  --provider google --model gemini-2.0-flash \
  --file ./photo.jpg \
  "Describe what you see in this image"

# Technical image analysis
umwelten run \
  --provider ollama --model qwen2.5vl:latest \
  --file ./screenshot.png \
  "Identify the user interface elements and their functions"

# Multiple images
umwelten run \
  --provider google --model gemini-2.0-flash \
  --file ./before.jpg --file ./after.jpg \
  "Compare these two images and describe the differences"

Document Analysis

# PDF summarization
umwelten run \
  --provider google --model gemini-2.0-flash \
  --file ./research-paper.pdf \
  "Summarize the key findings and methodology of this research paper"

# Document question-answering
umwelten run \
  --provider google --model gemini-2.5-pro-exp-03-25 \
  --file ./contract.pdf \
  "What are the main obligations of each party in this contract?"

Use Cases and Examples

Content Creation

# Blog post writing
umwelten run \
  --provider google --model gemini-2.0-flash \
  --system "You are a technology blogger with expertise in AI and machine learning" \
  "Write a 500-word blog post about the future of AI in healthcare"

# Social media content
umwelten run \
  --provider ollama --model gemma3:12b \
  --temperature 0.8 \
  "Create 5 engaging LinkedIn posts about productivity tips for remote workers"

Code Assistance

# Code review
umwelten run \
  --provider ollama --model codestral:latest \
  --system "You are a senior software engineer conducting a code review" \
  --file ./my-function.py \
  "Review this Python function and suggest improvements"

# Code explanation
umwelten run \
  --provider openrouter --model openai/gpt-4o \
  --file ./complex-algorithm.js \
  "Explain how this algorithm works and its time complexity"

Research and Analysis

# Market research
umwelten run \
  --provider google --model gemini-2.5-pro-exp-03-25 \
  --system "You are a market research analyst" \
  "Analyze the current trends in the electric vehicle market and provide insights for the next 5 years"

# Academic research assistance
umwelten run \
  --provider openrouter --model anthropic/claude-3.7-sonnet:thinking \
  --file ./research-data.pdf \
  "Identify the key themes and gaps in this literature review"

Educational Support

# Concept explanation
umwelten run \
  --provider ollama --model gemma3:27b \
  --system "You are a patient teacher explaining complex concepts simply" \
  "Explain quantum mechanics to a high school student"

# Problem solving
umwelten run \
  --provider google --model gemini-2.0-flash \
  --system "You are a math tutor who shows step-by-step solutions" \
  "Solve this calculus problem: Find the derivative of x^3 + 2x^2 - 5x + 1"

Model Selection Guide

For Speed and Efficiency

• Google Gemini 2.0 Flash: Best balance of speed and quality
• Google Gemini 1.5 Flash 8B: Fastest responses, good for simple tasks
• Ollama Gemma3:12b: Fast local processing, no API costs

For Highest Quality

• Google Gemini 2.5 Pro: Best for complex reasoning and analysis
• OpenRouter GPT-4o: Premium quality for critical tasks
• OpenRouter Claude 3.7 Sonnet: Excellent for analytical tasks

For Cost Optimization

• Ollama Models: Free local processing (requires local setup)
• Google Gemini 2.0 Flash: Very cost-effective for most tasks
• OpenRouter GPT-4o-mini: Good quality at lower cost than GPT-4o

For Specialized Tasks

• Vision: Google Gemini 2.0 Flash, Ollama qwen2.5vl:latest
• Code: Ollama codestral:latest, OpenRouter GPT-4o
• Long Context: Google Gemini models (up to 2M tokens)
• Creative Writing: Models with higher temperature settings

Best Practices

Prompt Engineering

• Be Specific: Include details about desired format, length, and style
• Provide Context: Use system messages to set role and expertise level
• Use Examples: Include examples of desired output when helpful
• Iterate: Refine prompts based on initial results

System Message Guidelines

• Set Clear Roles: "You are a [role] with expertise in [domain]"
• Define Constraints: Specify output format, length, or style requirements
• Establish Context: Provide relevant background information
• Set Tone: Professional, casual, academic, creative, etc.

Temperature Selection

• 0.0-0.3: Factual information, technical documentation, precise tasks
• 0.4-0.7: General conversation, balanced creativity and accuracy
• 0.8-1.2: Creative writing, brainstorming, open-ended exploration
• 1.3-2.0: Highly creative tasks, experimental content generation

Error Handling

• Set Appropriate Timeouts: Longer for complex tasks
• Handle Rate Limits: Use different providers or reduce frequency
• Validate Outputs: Check responses for accuracy and completeness
• Retry Logic: Be prepared to retry with adjusted parameters

Troubleshooting

Common Issues

1. Connection Errors: Check internet connection and API keys
2. Rate Limiting: Reduce request frequency or switch providers
3. Timeout Errors: Increase timeout values for complex prompts
4. Invalid Responses: Adjust temperature or rephrase prompts
5. File Attachment Issues: Check file size and format compatibility

Debug Commands

# Test connection to provider
umwelten models list --provider google

# Check available models
umwelten models list --provider ollama

# Test simple prompt first
umwelten run --provider google --model gemini-2.0-flash "Hello, world!"

Next Steps

• Try interactive chat for extended conversations
• Explore model evaluation for systematic testing
• Learn batch processing for multiple files
• See structured output for data extraction