TokenLab Python SDK

Lightweight Python client for TokenLab discovery, OpenAI-compatible APIs, and native endpoint families.

Install from PyPI:

pip install tokenlab-ai

Usage

from tokenlab import TokenLabClient

with TokenLabClient(api_key="YOUR_TOKENLAB_API_KEY") as tokenlab:
    models = tokenlab.list_models(category="chat")
    response = tokenlab.create_response({
        "model": "gpt-5.5",
        "input": "Hello from TokenLab",
    })

Native Endpoints

tokenlab.create_anthropic_message({
    "model": "claude-sonnet-5",
    "max_tokens": 512,
    "messages": [{"role": "user", "content": "Hello"}],
})

tokenlab.create_gemini_content("gemini-3.5-flash", {
    "contents": [{"role": "user", "parts": [{"text": "Hello"}]}],
})

Discovery Helpers

tokenlab.get_models_json()
tokenlab.get_pricing_json()
tokenlab.get_integrations_json()

System One decisions

Discover tokenlab.list_models(category="decision") and inspect the selected model with get_model() first. Decision models declare the systemone public operation, not a chat format.

with TokenLabClient(timeout=60.0) as tokenlab:
    decision = tokenlab.evaluate_decisions({
        "model": "jev-1.13",
        "state": {"ticket": "Please refund my duplicate charge."},
        "questions": {"refund": {"type": "noul", "instructions": "Is a refund requested?"}},
    })
    print(decision["answers"], decision["usage"])

The client reads TOKENLAB_API_KEY from the environment. This synchronous /v1/systemone call preserves typed answers, probabilities, optional confidence, and usage. It is not Chat, streaming, or Batch. A decision does not authorize a refund or another side effect. Errors are returned without automatic resubmission. Available in SDK 0.1.2+.

Base URLs

• API: https://api.tokenlab.sh
• OpenAI-compatible SDK base URL: https://api.tokenlab.sh/v1
• OpenAPI: https://docs.tokenlab.sh/openapi.json