You have dashboards. Good ones, probably. The trouble is that a dashboard shows you what happened, not why it happened. And when you’re an analyst or a founder who needs an answer ten minutes before a meeting, a wall of PromQL panels won’t get you there fast enough.
This guide covers grafana ai integration with HotBot AI: asking your metrics questions in plain English, getting anomalies explained the way a colleague would explain them, and drafting weekly reports without hand-writing a single query. You’ll get concrete workflows, exact prompts, and a note on which HotBot model tier fits each job.
HotBot is an independent AI chat service that gives you one subscription across 800+ models from every major provider, plus its own HotBot Chat, Chat Plus, and Chat Pro engines. It works as a Grafana ChatGPT alternative with a wider model selection.
What You’ll Need
- A Grafana instance (Grafana 10.0.0 or newer works best for plugin-based approaches)
- A data source already connected — Prometheus, Loki, CloudWatch, whatever you’re running
- Read access to your dashboards and queries (Admin if you want write actions later)
- A HotBot account — the free tier, $7.95/week or $39.95/quarter all work for this
- Optional: the ability to copy query results or export panel data (CSV, JSON) for pasting into chat
You don’t need to be a PromQL expert. Not needing to be one is most of the point.
Estimated Time: 20–30 minutes for setup, then seconds per question Difficulty Level: Beginner to Intermediate
Step 1: Decide How HotBot Talks to Grafana
There are two ways to do this. Pick based on how much plumbing you want to deal with.
Option A — Copy-paste (fastest, zero setup). You grab a query, a result table, or a panel’s exported data from Grafana and paste it straight into a HotBot chat. This works today, needs no plugins, and respects whatever access controls you already have. Most “just tell me what this means” questions get solved this way.
Option B — Plugin bridge. If you want the AI to query Grafana live, there’s an open-source plugin, tamcore/grafana-llmanalysis-app, that connects any OpenAI-compatible LLM endpoint to your Grafana instance. It queries Prometheus, Loki, and Alertmanager in real time via tool calling. It requires Grafana ≥ 10.0.0 and an endpoint that implements the OpenAI POST /v1/chat/completions API.
For most analysts and founders, start with Option A. You’ll get value in ten minutes and can graduate to Option B when copy-paste starts feeling like a chore.
Step 2: Set Up the Copy-Paste Workflow (Option A)
This is the one to start with. Here’s what to do.
- Open the Grafana panel you care about. Click the panel title, then Inspect → Panel JSON or Inspect → Data to see the raw query and values.
- To grab the underlying query, use Edit and copy the query string (for example, a PromQL line like
rate(http_requests_total[5m])). - To grab the data, use Inspect → Data → Download CSV.
- Open HotBot Chat, paste, and ask your question.
A starter prompt that works well:
Here's a Grafana query and its results (CSV below).
Explain in plain English what this metric shows, what's
normal, and anything that looks off. Assume I'm a founder,
not an SRE.
Query: rate(http_requests_total[5m])
[paste CSV here]
No plugin, no tokens, no YAML. You’re already talking to your metrics.
Step 3: Ask Plain-English Questions of Your Metrics
This is the everyday workflow, and it’s where ai data analysis with grafana stops being a buzzword and starts saving you time.
To see average latency you’d normally write rate(http_request_duration_seconds_sum[5m]) / rate(http_request_duration_seconds_count[5m]). One blogger described exactly this frustration on Mustafa.net. With HotBot you just ask.
Some prompts that work:
- “Show me what average request latency looked like over the last 24 hours and whether it’s trending up.”
- “I need the PromQL for error rate as a percentage of total requests. Explain each part.”
- “Translate this dashboard’s five panels into three sentences I can say in a standup.”
Model tier for this job: Use HotBot Chat, the fast everyday tier. These questions don’t need deep reasoning, and you want the answer quickly.
When you ask HotBot to write a query, give it context: your data source (Prometheus vs. Loki), the metric names, and what “good” looks like. A reference table from a Grafana dashboard tutorial shows the patterns — error rate is sum(rate(http_requests_total{status=~"5.."}[5m])) / sum(rate(http_requests_total[5m])). Paste examples like that and HotBot adapts them to your metric names.
Step 4: Explain Anomalies Like a Human Would
Grafana Cloud has its own AI-powered anomaly detection and outlier detection that learns your historical patterns, per Grafana Labs. Detection tells you something’s weird. It doesn’t always tell you why in language your CEO understands.
That’s where HotBot comes in.
Take a concrete case: someone pointed anomaly detection at their disk I/O and it flagged sustained high write activity at 4 AM. The cause turned out to be a misconfigured backup job thrashing the array, as they described on Mustafa.net. Detection caught it; understanding it was the plain-English step.
A prompt for that situation:
This anomaly fired at 4 AM: disk write throughput jumped
from a ~10 MB/s baseline to 180 MB/s for 40 minutes, then
dropped. Here's the surrounding metric data and the list of
scheduled jobs. Walk me through the 3 most likely causes,
ranked, with what I'd check to confirm each.
Model tier for this job: Reach for HotBot Chat Plus when you want step-by-step thinking — ranking causes, weighing evidence, showing its work. When the anomaly spans multiple metrics, logs, and traces and you need deeper reasoning, step up to HotBot Chat Pro, which brings a 1M-token context window and vision. That last part matters because sometimes the fastest input is a screenshot of the spike, and Chat Pro can read it directly.
Step 5: Draft Weekly Reports Automatically
Grafana Open Source lacks native reporting — Skedler points this out, and it’s a real gap. You can schedule PDF exports with add-ons, but turning raw exports into a narrative is the hard part, and it’s the part nobody wants to do on a Friday afternoon.
So hand it off. Feed HotBot your week’s data and let it write the story.
Here's this week's key metrics vs. last week (CSV attached):
request volume, p99 latency, error rate, and two incidents.
Write a 200-word weekly reliability report for a non-technical
leadership audience. Lead with the headline, note what
improved, flag the one thing we should watch, and end with a
recommendation.
Model tier for this job: HotBot Chat Plus is the sweet spot for structured, multi-step summarization without overkill. If you’re consolidating a month of data or multiple services into one exec brief, HotBot Chat Pro and its larger context window earns its keep.
When a third-party model is the better pick: HotBot gives you 800+ models under one subscription. If your team standardizes on a specific model for report tone, or you want one known for long-context digestion, switch to it in the same chat. You’re not locked to a single model, which is the practical benefit of a real ChatGPT alternative.
Step 6: Level Up to Live Querying (Optional)
Once copy-paste feels limiting, wire up Option B. With the grafana-llmanalysis-app plugin, you go to Administration → Plugins → LLM Analysis and set the Endpoint URL, Model, API Key, and a Grafana Service Account Token (required for tool calling). After that, the model can query your infrastructure in real time instead of waiting for you to paste data.
If you’re on Grafana Cloud specifically, there’s also an official Grafana MCP server. A usecarly.com breakdown explains that it exposes 50+ tools for querying dashboards, running datasource queries, and reading alerts. The catch: it only works inside a conversation you start. Nothing watches your metrics while you’re away, so don’t assume the AI is standing guard. It isn’t.
Common Issues
HotBot’s answer is generic or wrong. You probably didn’t give it enough context. Paste the actual query, the metric names, and a sample of real data — not just “explain my latency.”
The numbers don’t add up. Check that your CSV export didn’t truncate. Grafana’s Inspect → Data sometimes only shows the visible time range. Widen the range before exporting.
Plugin won’t connect (Option B). Confirm your endpoint implements the OpenAI /v1/chat/completions API and that tool calling is supported. Verify the Grafana Service Account Token is set correctly.
You want automated, always-on monitoring. HotBot is conversational — it answers when you ask. For always-on detection, lean on Grafana’s own anomaly and outlier tooling (Grafana Labs), then bring the findings to HotBot for explanation.
Limits and Privacy
When you paste metric data into any AI chat, you’re sending that data to a model. Don’t paste secrets, customer PII, or credentials — scrub or aggregate first. For sensitive environments, the plugin route keeps you in more control of what leaves your instance.
One more thing worth repeating: HotBot explains, drafts, and reasons. It doesn’t run on a schedule watching your systems. Pair it with Grafana’s native alerting for the always-on layer.
What’s Next
Start with Step 2 today — one panel, one question, one paste. Once you trust the answers, build a saved prompt template for your weekly report so Friday takes five minutes instead of an hour.
When you’re ready, explore the full model lineup and pick tiers per task. If you’re comparing plans, the free tier, $7.95/week or $39.95/quarter options let you test everything before committing.
Your dashboards already know what happened. This gives you something that can tell you why.
FAQ
Do I need to write PromQL to use HotBot with Grafana?
No. You can ask HotBot to write queries for you, or paste results and ask what they mean. Give it your metric names and data source type for the best results.
Is HotBot affiliated with Grafana?
No. HotBot is an independent AI chat service. Third-party names like Grafana identify the tools and models you can work with — they don’t imply any partnership or endorsement.
Which HotBot tier should I use for anomaly analysis?
Use HotBot Chat Plus for ranked, step-by-step cause analysis. Step up to HotBot Chat Pro when you need deep reasoning across many signals or want to paste screenshots, thanks to its 1M-token context and vision.
Can HotBot monitor my metrics automatically?
No — it’s conversational and answers when you ask. For always-on detection, use Grafana’s built-in anomaly and outlier tools, then bring the alerts to HotBot for a plain-English explanation.
What’s the cheapest way to try this?
Start on the free tier with the copy-paste workflow. If you need more, plans are available at $7.95/week or $39.95/quarter.
Is HotBot a real ChatGPT alternative for data work?
Yes. One subscription gives you access to 800+ models plus HotBot’s own Chat, Chat Plus, and Chat Pro engines, so you can pick the right model per task instead of being locked to one.
