Trace your first LLM call in 5 lines

The fastest way to understand Currai is to send one trace. There is no agent to deploy, no collector to run, and no schema to define — install the SDK, authenticate, and wrap a call.

1. Install and authenticate

One dependency, no native build step. Authenticate with a public / secret key pair scoped to your project.

# pip install currai   (or:  npm i currai)
from currai import Currai

currai = Currai(public_key="pk-lf-...", secret_key="sk-lf-...")

2. Wrap a call

A trace is the unit of work; a generation is a single model call inside it. Currai records the prompt you sent, the completion you got back, the model and its parameters, and how long it took.

trace = currai.trace(name="chat-turn", user_id="user-1")

gen = trace.generation(
    name="openai.chat.completions",
    model="gpt-4o-mini",
    input=messages,
)
reply = client.chat.completions.create(model="gpt-4o-mini", messages=messages)
gen.end(output=reply.choices[0].message.content)

3. Flush before you exit

Traces are batched and flushed in the background so instrumentation never blocks a request. In a short-lived process — a script or a serverless function — call flush so traces are sent before the process exits.

currai.flush()              # sync
# await currai.flush_async()  # async runtimes

From here, attach token usage to see cost computed for you, nest spans to capture a RAG pipeline, or pass a session id to stitch a multi-turn conversation together. Each is a single extra argument away.