Skip to main content

Tutorial — Auto-instrument an OpenAI app

Time: 15 minutes. By the end you'll have a working OpenAI chat script that sends full traces to Eden with one line of code.

What you'll build​

A small command-line "FAQ bot" that answers a question with a few shots, and a dashboard trace showing every LLM call with prompt, response, tokens, cost, and latency.

Step 1 — Set up​

mkdir eden-faq-bot && cd eden-faq-bot
python -m venv .venv && source .venv/bin/activate
pip install eden-sdk openai
export EDEN_API_KEY="sk_eden_..."
export EDEN_ORG_ID="org_..."
export OPENAI_API_KEY="sk-..."
tip

Fastest path: use Fill in on the docs quickstart to create a key and inject it into a snippet.

Step 2 — Write the bot (uninstrumented)​

# bot.py
import os
from openai import OpenAI

client = OpenAI()

FAQ = [
    "Refund policy: 30 days, no questions asked.",
    "Shipping: 3-5 business days, free over $50.",
    "Support: help@acme.com or chat 24/7.",
]

def answer(question: str) -> str:
    resp = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {"role": "system", "content": "You are a helpful FAQ bot.\n" + "\n".join(FAQ)},
            {"role": "user", "content": question},
        ],
    )
    return resp.choices[0].message.content

if __name__ == "__main__":
    print(answer(input("Ask: ")))

Run it to confirm it works:

python bot.py
Ask: How long does shipping take?
Shipping takes 3-5 business days, and is free over $50.

Step 3 — Add Eden (two lines)​

# bot.py
import os
import eden
from openai import OpenAI

eden.configure(
    api_key=os.environ["EDEN_API_KEY"],
    org_id=os.environ["EDEN_ORG_ID"],
)
eden.instrument()      # ← this patches openai globally

client = OpenAI()

FAQ = [...]

@eden.trace_agent(name="faq-bot")   # ← every bot invocation becomes one trace
def answer(question: str) -> str:
    resp = client.chat.completions.create(...)
    return resp.choices[0].message.content

Run it again. Same answers, but now the portal is recording every call.

Step 4 — See it in the portal​

Open the portal's Traces tab. You should see a new row called faq-bot. Click into it:

  • Timeline shows the LLM call as one span (openai.chat.completions) with start/end, model, and a snippet of the prompt.
  • Tokens + cost shows input/output token counts and the dollar cost (computed from Eden's model pricing table).
  • Latency shows the wall-clock duration of the LLM call.

Step 5 — Add attributes​

Attributes show up in the portal's filter bar and get indexed for search. Add a user id:

@eden.trace_agent(name="faq-bot", attributes={"surface": "cli", "version": "1.0.0"})
def answer(question: str, user_id: str) -> str:
    eden.set_attribute("user_id", user_id)
    resp = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[...],
    )
    eden.set_attribute("input_tokens", resp.usage.prompt_tokens)
    return resp.choices[0].message.content

if __name__ == "__main__":
    import sys
    print(answer(input("Ask: "), user_id=sys.argv[1] if len(sys.argv) > 1 else "anon"))

Now user_id is searchable and you can scope the Traces tab to a specific user.

Step 6 — Tag and group runs​

@eden.trace_agent(name="faq-bot", tags=["prod", "support"], version="1.0.0")
def answer(...): ...

Tags are free-form strings — use them for environment (prod, staging), feature area (support, checkout), or anything else.

What you learned​

  • eden.configure(...) reads env vars and sets up the SDK.
  • eden.instrument() patches the OpenAI client globally — no per-call code.
  • @eden.trace_agent(...) wraps a function as a top-level span; LLM calls inside it become child spans automatically.
  • eden.set_attribute(...) adds metadata that the portal indexes.
  • tags=[...] are searchable.

Where next​