PromptAnalyzer gives any LLM-powered Python function automatic prompt versioning, inference logging, token & cost tracking, and a local dashboard — by adding a single @track decorator. Everything runs on your machine against SQLite. Nothing leaves your laptop.

from promptanalyzer import track


@track("medical-chatbot")
def ask(message):
    return client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {"role": "system", "content": "You are a doctor's assistant."},
            {"role": "user", "content": message},
        ],
    )

promptanalyzer dashboard   # → http://localhost:4001

✨ Features

🚀 Quick start

pip install "promptanalyzer[dashboard]"

1. Decorate an LLM function with @track("project-name").
2. Run your application as usual.
3. Open the dashboard:

promptanalyzer dashboard
# http://localhost:4001

That's it. PromptAnalyzer creates ~/.promptanalyzer/promptanalyzer.db on first use.

Advanced decorator

@track(
    name="medical-assistant",
    tags=["production"],
    metadata={"team": "AI"},
)
def chatbot(message): ...

Any library (generic adapter)

@track(
    name="custom-model",
    system=lambda args, kwargs: kwargs["system"],
    user=lambda args, kwargs: kwargs["prompt"],
    response=lambda result: result,
)
def my_llm(system, prompt): ...

🖥️ Dashboard

The images above are placeholders. To capture real screenshots, follow docs/screenshots.md.

📚 Documentation

Full guides live in docs/: Quickstart · Configuration · Providers & adapters · Dashboard · CLI · Prompt versioning · Database & migrations · Performance · FAQ.

🔌 Supported providers

OpenAI · Anthropic Claude · Google Gemini · Ollama · vLLM · LiteLLM · OpenRouter · Groq · Mistral · Azure OpenAI · any custom library via the generic adapter.

See examples/ for a runnable script per provider.

⚙️ Configuration

Zero config by default. Override via environment variables:

PROMPTANALYZER_DB=sqlite
PROMPTANALYZER_SQLITE_PATH=~/.promptanalyzer/promptanalyzer.db
PROMPTANALYZER_DATABASE_URL=postgresql://user:password@localhost/dbname

PROMPTANALYZER_HOST=127.0.0.1
PROMPTANALYZER_PORT=4001

PROMPTANALYZER_AUTO_START=true      # start dashboard on import
PROMPTANALYZER_OPEN_BROWSER=true

PROMPTANALYZER_LOG_TOKENS=true
PROMPTANALYZER_LOG_COST=true
PROMPTANALYZER_SAVE_RESPONSES=true

PROMPTANALYZER_PROJECT=default
PROMPTANALYZER_ENV=development

Priority: decorator arguments → environment variables → defaults. Legacy PROMPTLOG_* variables are also accepted.

🛠️ CLI

promptanalyzer init        # create ~/.promptanalyzer and the database
promptanalyzer dashboard   # launch the dashboard
promptanalyzer migrate     # create/upgrade the schema
promptanalyzer export csv  # export runs (json | csv | markdown)
promptanalyzer doctor      # diagnose your installation
promptanalyzer reset       # wipe all local data

🧱 Architecture

@track → adapter (normalizes any provider) → background writer → SQLite → FastAPI dashboard. See ARCHITECTURE.md for the full design.

🗺️ Roadmap

• OpenTelemetry export
• Prompt evaluation & A/B testing
• Dataset management & prompt playground
• Cloud sync + team collaboration
• Authentication & multi-user
• Plugin marketplace

🤝 Contributing

Contributions welcome! See CONTRIBUTING.md.

📄 License

MIT © PromptAnalyzer Contributors