Quickstart

Install the SDK, initialize it once, and see your first trace in the dashboard.

Two SDKs, same concepts: Python (syntropylabs-evalkit, imported as evalkit) and TypeScript (syntropylabs-evalkit). One principle drives both: `init()` does everything. After init, your LLM calls, HTTP, DB, and logs are traced automatically — including tool/function calls made by the LLM. You do not create spans by hand.

1. Install

pip install syntropylabs-evalkit      # import name stays: import evalkit

You need a subscription key (tk_live_…) from the dashboard: Tracing → create a trace project.

2. Initialize (once, at startup)

import evalkit

evalkit.init(
    subscription_key="tk_live_...",     # required
    service_name="my-api",              # shows up in the dashboard
    environment="production",           # development | staging | production
)
# ... your app runs ...
evalkit.flush()                          # call before the process exits

3. Make a traced call — nothing else to wire up

from openai import OpenAI

client = OpenAI()                       # auto-traced after evalkit.init()
resp = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.choices[0].message.content)

4. View it in the dashboard

Open Dashboard → Tracing, select your project, and click the request you just made — you’ll see the full waterfall: the LLM call, its tokens and latency, and any tool calls the model made.

Next steps

• Core Concepts — traces, spans, projects, evaluators, simulations
• Python SDK / TypeScript SDK — manual spans, function tracing (APM), framework middleware
• LLM Providers — the full auto-instrumentation support matrix
• Scenario Simulation — test your agent against synthetic users before real ones hit it