Quickstart

Send your first trace to Sightline and add an evaluator. Takes about five minutes.

1Install the SDK

The SDK supports Python 3.9+ and Node.js 18+.

pip install sightline

npm install @sightline/sdk

2Set your API key

Create a key in Settings → API keys. Keys are scoped to one project and one environment.

SIGHTLINE_API_KEY=sl_live_4f9a…c21e
SIGHTLINE_ENV=production

3Trace your agent

Call init() once at startup. Calls to Anthropic, OpenAI and the other supported clients are captured automatically. Wrap your entry point with @sightline.agent to group them into one trace.

import sightline, anthropic

sightline.init(project="acme-support")
client = anthropic.Anthropic()

@sightline.agent(name="kb-answerer")
def answer(question: str) -> str:
    docs = search_kb(question)          # traced as a retrieval span
    msg = client.messages.create(
        model="claude-haiku-4-5",
        max_tokens=1024,
        messages=[{"role": "user", "content": prompt(question, docs)}],
    )
    return msg.content[0].text

Run your agent once, then open Traces. The trace should appear within a second:

4Add an evaluator

Evaluators score traces after they finish, so they never add latency. Start with groundedness on 20% of traffic:

sightline.evaluate("groundedness", agent="kb-answerer", sample=0.2)

Common options:

Next steps