🎯 prompt-studio

Local prompt manager for AI developers.

Create, version, test, and compare prompts across LLMs — right from your terminal.

npm install -g @kevinxyz/prompt-studio
prompt init
prompt run greeting --var name=World
prompt compare code-review --models gpt-4o-mini,gpt-4o,claude-3-haiku

Why?

Prompt engineering is becoming a core skill for developers. Yet most people manage prompts by:

• Pasting them into ChatGPT and losing track of versions
• Keeping a messy Notes app with 20 variants
• Never comparing how different models handle the same prompt

prompt-studio fixes this. Keep prompts as local Markdown files, version them with git, test them instantly.

Quick Start

# Initialize with example prompts
prompt init

# List all prompts
prompt list

# Set your API key first
# You can also set OPENAI_API_KEY instead
export AI_ANNOTATOR_API_KEY="sk-your-key-here"

# Run a prompt
prompt run greeting --var name=World

# Create your own
prompt new my-prompt
prompt run my-prompt --var topic=AI --model gpt-4o-mini

Prompt Format

Prompts are simple Markdown files with optional YAML frontmatter:

---
model: gpt-4o-mini
temperature: 0.7
max_tokens: 500
lang: en
---
You are a {{role}}. Explain {{topic}} in simple terms.

Variables use {{double_curly_braces}} syntax. Set them with --var key=value.

All Commands

| Command | Description | |---------|-------------| | prompt init | Create example prompts in .prompts/ | | prompt list | List all prompts | | prompt new <name> | Create a new prompt template | | prompt show <name> | Show a prompt's full content | | prompt run <name> | Execute a prompt via LLM | | prompt run <name> --raw "hello" | Use raw text instead of file | | prompt compare <name> | Run on multiple models side by side | | prompt preview <name> | Show resolved template (no API call) | | prompt log | View recent run history | | prompt update | Check for updates and upgrade to latest version | | prompt rm <name> | Delete a prompt |

Options

--var key=val        Set template variable values
--model <name>       Override model (default: gpt-4o-mini)
--temp <n>           Override temperature
--max-tokens <n>     Override max tokens
--lang <en|zh>       Shortcut for --var lang=...
--preview            Dry-run (show resolved prompt, no API call)
--models <list>      For compare: comma-separated model list

Environment Variables (Required)

You must set an API key before running any prompt:

| Variable | Description | |----------|-------------| | AI_ANNOTATOR_API_KEY | OpenAI API key (recommended, overrides OPENAI_API_KEY) | | OPENAI_API_KEY | Fallback API key |

Quick setup (choose one):

# Set directly (replace with your real key)
export AI_ANNOTATOR_API_KEY="sk-..."

# Or add to ~/.zshrc to persist
# echo 'export AI_ANNOTATOR_API_KEY="sk-..."' >> ~/.zshrc

# Or use .env in your project
echo 'AI_ANNOTATOR_API_KEY=sk-...' > .env

Optional: | Variable | Default | Description | |----------|---------|-------------| | PROMPT_STUDIO_MODEL | gpt-4o-mini | Default model |

💡 Get an API key: Sign up at platform.openai.com/api-keys to get a free trial key.

Example Workflow

# 1. Set up
cd my-project
npm install @kevinxyz/prompt-studio
prompt init

# 2. Create a prompt for code review
prompt new code-review

# 3. Edit it (it's just a Markdown file!)
# echo "your prompt here" > .prompts/code-review.md

# 4. Run it
prompt run code-review --var language=Python --var code="print('hello')"

# 5. Compare across models
prompt compare code-review --models gpt-4o-mini,gpt-4o

# 6. Check run history
prompt log

File Structure

.prompts/
├── greeting.md        # Example: friendly greeting
├── code-review.md     # Example: code review
├── summarize.md       # Example: text summarization
├── translate.md       # Example: translation
├── your-prompt.md     # Your custom prompts
├── system.md          # Optional: shared system prompt
└── .history.jsonl     # Run log (auto-generated)

All prompts are plain text — commit them to git, share with your team, track changes like code.

Benefits Over Web UI

• Version controlled — git diff your prompts
• Local — no data leaves your machine
• Scriptable — use in CI/CD pipelines
• Model agnostic — switch between OpenAI, Anthropic, etc.
• Cost effective — compare outputs before committing to expensive models

Roadmap

• [x] Template variables
• [x] Multi-model comparison
• [x] Run history
• [ ] Anthropic/Claude API support
• [ ] Batch run (run 10 prompts, save outputs)
• [ ] Export to shareable format
• [ ] Local web UI dashboard

License

MIT