Skip to main content
This guide gets you from zero to a traced LLM call visible in the dashboard in under 10 minutes.

​
Prerequisites

  • A TruLayer account — sign up free (no credit card required on the Free tier)
  • One of:
    • Python 3.11+ for the Python SDK
    • Node.js 18+ (or Bun, or an Edge runtime) for the TypeScript SDK
  • An LLM provider API key — this quickstart uses OpenAI

​
1. Create an API key

After signing up, go to Settings → API keys in the dashboard and click Create key. Copy the key — it starts with tl_ and is shown only once.
API keys are sensitive. Store them in environment variables, not in source code. Keys are tenant-scoped — they grant ingest and read access to every project in your organisation.
export TRULAYER_API_KEY=tl_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
export OPENAI_API_KEY=sk-...

​
2. Install the SDK

pip install trulayer openai

​
3. Send your first trace

The example below makes one OpenAI call and ships the trace (prompt, response, latency, token counts, model) to TruLayer. It uses auto-instrumentation so you don’t have to wrap calls manually. 
import os
from openai import OpenAI
import trulayer

trulayer.init(
    api_key=os.environ["TRULAYER_API_KEY"],
    project="quickstart",
)

client = OpenAI()
trulayer.instrument_openai(client)

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Give me a one-sentence definition of observability."}],
)

print(response.choices[0].message.content)
In short-lived scripts, always call flush() (TS) or let the process exit naturally after trulayer.init() — the SDK batches traces and needs a moment to ship them before exit.

​
4. View the trace

Open the dashboard and navigate to Traces. Your trace should appear within a few seconds, showing:
  • The input prompt
  • The full response
  • Latency, token counts, and cost
  • The model name
  • Any errors (there shouldn’t be any yet)
Click the trace to see the span waterfall — one span per LLM call, with full request and response captured.

​
5. Next steps

Add more auto-instrumentation

Patch Anthropic, LangChain, or the Vercel AI SDK with a single function call.

Wrap custom code

Create traces and spans manually for retrieval, tool calls, or your own business logic.

Collect feedback

Attach user thumbs-up/down to traces and feed it back into your evals.

Run an evaluation

Score traces for correctness, hallucination, or any custom metric.
If the trace doesn’t appear, double-check TRULAYER_API_KEY, make sure the process didn’t exit before the background flush completed, and check the status page. Still stuck? Email [email protected].