🐟 Tuna

Dead simple prompt tuner.

Test and compare LLM prompts across multiple models in seconds.

Tuna automates prompt engineering workflows. Write your prompts, define test queries, and instantly see how different models respond — all from the command line.

Why Tuna?

Iterating on system prompts is tedious: copy-paste into chat interfaces, switch between models, manually compare outputs. Tuna eliminates this friction.

• Organize prompts and test queries in a simple folder structure
• Execute the same queries across multiple models with one command
• Compare responses side-by-side to find what works best

Quick Start

# Initialize a new assistant
tuna init my-assistant

# Edit your system prompt
echo "You are a helpful assistant." > my-assistant/System\ prompt/fragment_001.md

# Add test queries
echo "Explain quantum computing in simple terms." > my-assistant/Input/query_001.md

# Create an execution plan
tuna plan my-assistant --models claude-sonnet-4-20250514,gpt-4o

# Run it
export LLM_BASE_URL="https://api.openai.com/v1"
export LLM_API_TOKEN="your-token"
tuna exec <plan-id>

Results are saved to my-assistant/Output/<plan-id>/ for easy comparison.

Project Structure

my-assistant/
├── Input/              # Your test queries
│   └── query_001.md
├── Output/             # Generated responses
│   └── <plan-id>/
│       └── <model>/
└── System prompt/      # Prompt fragments (concatenated in order)
    └── fragment_001.md

Configuration

Variable	Description
LLM_BASE_URL	Base URL for OpenAI-compatible API
LLM_API_TOKEN	API token for authentication

Installation

go install go.octolab.org/toolset/tuna@latest

License

MIT

made with ❤️ for everyone by OctoLab