# I built a Chrome extension that explains PMM charts with AI — looking for feedback

**URL:** <https://forums.percona.com/t/i-built-a-chrome-extension-that-explains-pmm-charts-with-ai-looking-for-feedback/40648>\
**Category:** Percona Monitoring and Management (PMM)\
**Created:** [April 17, 2026, 8:05am UTC](https://forums.percona.com/t/i-built-a-chrome-extension-that-explains-pmm-charts-with-ai-looking-for-feedback/40648 "2026-04-17T08:05:43Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![daniil.bazhenov](https://sea1.discourse-cdn.com/flex019/user_avatar/forums.percona.com/daniil.bazhenov/32/62_2.png) [@daniil.bazhenov](https://forums.percona.com/u/daniil.bazhenov)\
**Post date:** [April 17, 2026, 8:05am UTC](https://forums.percona.com/t/i-built-a-chrome-extension-that-explains-pmm-charts-with-ai-looking-for-feedback/40648/1 "2026-04-17T08:05:43Z")

</div>

Hey everyone 👋

I want to share a small side project I’ve been working on and get some feedback  
from the community.

## The problem

I spend a lot of time in PMM dashboards, and honestly — I don’t always know what  
every chart is telling me. Some panels are self-explanatory, but others have complex  
PromQL queries, multiple series, and metric names that require context to interpret.  
My usual workflow was: take a screenshot, copy the query, open ChatGPT, paste  
everything manually, ask “what does this mean?” — and repeat for every panel.

That got old fast.

## What I built

**PMM Chart AI Explainer** — a Chrome extension that adds a small **AI** button  
to every panel on your PMM dashboard. Click it, and within a few seconds you get  
an AI-generated explanation of what the metric measures, whether anything looks  
abnormal, and what to do about it.

 ![screen-1](https://us1.discourse-cdn.com/flex019/uploads/percona1/original/3X/1/2/121e6fe0fee964d9e1872d86a04c8dbcc192294a.png)

Under the hood it:

- Intercepts the real executed PromQL (with substituted variables, not templates)
- Takes a cropped screenshot of just that panel
- Optionally reads the actual data values from Inspect → Data
- Sends everything to OpenAI GPT-4o Vision with a structured prompt
- Automatically identifies the database type from metric prefixes  
(mysql\ __, pg\__ , mongodb\ __, node\__ , etc.)

The response always includes what the metric measures and related PMM panels  
to check — and only flags anomalies if something actually looks wrong,  
so it won’t invent problems on a healthy system.

 ![screen-example](https://us1.discourse-cdn.com/flex019/uploads/percona1/original/3X/5/3/53e2420361e62b707be138998c9b3d2d1f633510.png)

There’s also a **Debug Mode** if you want to see exactly what prompt and screenshot  
are sent before making any API call:

 ![screen-debug](https://us1.discourse-cdn.com/flex019/uploads/percona1/original/3X/a/9/a97486153075905c1fccde50a971fd8426edd7b1.png)  
 ![screen-debug-send-button](https://us1.discourse-cdn.com/flex019/uploads/percona1/original/3X/2/a/2ab29f8c65c6db44132c50d12d61ef106cfe9a40.png)

## Privacy & cost

The extension uses **your own OpenAI API key** — nothing goes through any  
intermediate server. The key lives in your local Chrome profile and is sent  
only to `api.openai.com`. A typical analysis costs around **$0.01** , so $1  
is more than enough to go through an entire PMM instance.

## How to try it

1. Clone the repo: [GitHub - dbazhenov/pmm-chart-ai-explainer: Chrome extension that adds AI-powered chart analysis to Percona Monitoring and Management (PMM) dashboards using OpenAI GPT-4o Vision · GitHub](https://github.com/dbazhenov/pmm-chart-ai-explainer)
2. Open `chrome://extensions`, enable Developer mode
3. Click **Load unpacked** → select the `extension/` folder
4. Add your OpenAI API key in the popup
5. Open any PMM dashboard and click **AI** on any panel

## Looking for feedback

This is an early version and I’d love to hear from people who actually use PMM  
day to day:

- Does the AI explanation match what you’d expect from that panel?
- Are there panel types or metric families where it gets confused?
- What would make this more useful in your workflow?

Any feedback — bug reports, ideas, or just “this helped me understand X” —  
is very welcome. The repo has a CONTRIBUTING guide if anyone wants to jump in.

Thanks for reading 🙏
