Two lines to see what your LLM is actually doing

Most AI bugs are not logic bugs. They are "why did this call cost twelve cents and take nine seconds" bugs, and you cannot fix what you cannot see. So on any AI engagement, the first thing we do is make the LLM calls observable, before we touch a single prompt.

Here is the whole setup with peekr.

Install and instrument

pip install peekr

Then, at the very top of your app:

import peekr
peekr.init(project="my-app")

That is the whole integration. peekr auto-instruments OpenAI, Anthropic, and Bedrock at the class level, so every call is traced without you wrapping anything or threading a client through your code.

What you get

Every call now records its prompt, response, token counts, cost, and latency. When a request is slow or expensive, you can see exactly which call caused it instead of guessing from an aggregate bill.

Video placeholder · Tracing a call with peekr

Why this comes first

Teams reach for prompt tweaks and model swaps before they have a single trace. That is optimizing blind. Give me a codebase with observability and I can find the cost blowup in an afternoon. Give me one without, and we are all just guessing together on a call.

Instrument first. Then optimize.