Any agent anywhere: Grafana Cloud is the open agentic observability platform

Cedric Ziel: Hey everyone. My name is Cedric. I'm a product manager at Grafana Labs, and this is Gianni, a product designer at Grafana Labs. Today, we'll show you how you can use your own agent wherever you work and how Grafana Cloud is the open agentic observability platform that supports you and your agents. If you find us on the floor, ask us about boats. Although I have to admit that Gianni has much nicer weather in Italy.

Gianni Baiardi: And definitely a much more fun boat.

Cedric Ziel: Oh, that's rude, man.

Gianni Baiardi: That's a fact. All right. Agents are already part of our everyday work. We use them to write code. We use them to open PRs, review these PRs, deploy changes, and we use them to investigate what's happening in our systems. But there are still gaps. Getting the right context into an agent or moving between tools can interrupt that flow. Today, we'll look at a few of these gaps and show how Grafana Cloud helps bring these pieces together.

Cedric Ziel: Yes, because with Grafana Cloud, you can use any agent. Of course, you can use Grafana Assistant, which is built into Grafana Cloud, but you can also connect your ChatGPT, your Claude, Codex, Py or literally any other agent through either the Grafana Cloud MCP server, or you just use gcx with your coding agent so that it get insights from the telemetry that you have in your Grafana Cloud environment.

To show you what that looks like, let's look at the real life scenario.

Gianni Baiardi: Imagine I'm working from a coffee shop. It's late morning and I haven't had my first coffee of the day yet. I'm third in line in the coffee queue when I receive a message on Slack. It's Besart, our director of engineering. He says that fewer customers are completing money transfers and asked me to take a look.

Now, I don't want to leave the coffee queue. It's almost my turn, but I want... And my laptop is still in my bag, right? I'm in the queue. So I wonder if there's something I can start using, start doing from the coffee queue. So maybe I can start with my phone.

All right. I'll open Claude. I'll start a new chat and I'm just going to ask, "What's up with my transfer service?" now, Claude is connected to Grafana Cloud through MCP. This gives it the possibility to check things in Grafana Cloud for me.

Now, I can see, of course we need to give it permissions and I can see that I get this interactive Grafana visualizations inside the conversation I'm having with my agent. I can check the service health, the transfer service health, and here I get the KPIs for my backend service. I can see that orders are pretty low, latency is okay. There's nothing really that stands out here, but I'm starting to rule out some hypothesis.

So my backend service is okay. I wonder if maybe the problem is on my front end. So let me ask, is the front end okay? And once again, Claude, through MCP, will connect to Grafana Cloud and pull up an interactive visualization for me.

Here I can check the web vitals. I can check what is happening on my front end. And also, in this case, I see no errors. Nothing really stands out, but very importantly, I'm starting to rule out some hypothesis.

So back to the slides. So what are these views, these interactive views that we saw on my mobile? They are MCP apps. The Grafana Cloud MCP server brings them into the conversation. It gives us an interactive way to interact with Grafana along with the agent response.

Now you can check things like service health, as we just did, right in the conversation right where you're working. Now we are announcing them today as an experimental feature and we are planning to add more of these views. So if there's one you will find useful, we will love for you to come and talk to us.

Now back to Besart's message. I got my coffee. I'm not in the coffee queue anymore. I can sit at the table, pull up my laptop and have a closer look at what is happening in Grafana.

Cedric Ziel: That's right. All right. Because I've been using Claude on my phone, I can easily go back to Claude on my desktop. And because of the continuity, I can pull up the same conversation here and I get the same insight as before. We see the MCP apps rendered here in the desktop app. It's a seamless experience because of the MCP app standard. But as Gianni pointed out, there's not really nothing that stands out here in the apps. So what I can do, I can just click the button here. I can go to Grafana Cloud to inspect what's going wrong.

This is front end observability. So we are observing real users with Grafana Cloud and it has this really useful feature that is called user actions. User actions is basically a concept for you to throw an event whenever a user performs a key action so that you can record key steps in your user journeys.

And what stands out to me here is this one, the submit transfer action. It is red and it has an elevated error rate. I could not dig into the issue in Grafana Cloud of course, but what I want to show you is the Sparkle icon here because everything has Sparkle buttons now and it brings up this model and it can send the context from the panel to a bunch of agents. I could send it to a Grafana Assistant. I could send it to Claude, but we already used Claude. So let's try Cursor.

I select Cursor and I click Send to Agent and what it does, it sends the context of the panel directly into Cursor and I can ask the agent, "Can you inspect the transfers and what's wrong with them?" Of course, my text to speech app died.

Can you check the transfer actions and what's wrong? All right, so Cursor has my code. That is my development environment and Cursor is also connected to Grafana Cloud via gcx. And that means it can execute gcx locally. gcx ships a bunch of skills and that gives the agent the information on how to investigate the issue with Grafana Cloud.

We should see a bunch of gcx commands being fired here. It would check the context to understand which Grafana Cloud environment it is connected to. It uses the front end sub command to understand which front end apps are available in this environment. And then it goes on and on and on and on. It can now query logs. It can query the user actions. And since it had access to my code, it should be able to help me. So we're not going to look at the generation here further.

Back to the slides. So Gianni, how does Cursor do this?

Gianni Baiardi: How does Cursor do this? How does Cursor investigate with Grafana Cloud? It uses Grafana gcx, our common line interface. gcx comes with a set of curated agent skills that guide common investigations, but the agent can also discover all the available commands and use them to query data, inspect services or look at front end sessions as we did.

We are giving Cursor the context from Grafana. gcx gives it a way to follow up, gather more evidence and work out what to check next.

Cedric Ziel: Yeah. So let's put it all together. We checked Grafana from Claude on the phone, then we sent it to the Claude desktop app. From there we went to Grafana and from there we went to Cursor. But what if I want to work differently?

There's this very useful open source project. It's called Hermes Agent and it's literally an agent that you can host on your hardware and it's very flexible. It actually learns from its interactions and you can give it all sorts of tools.

So what I did in preparation of this demo, I connected Hermes agent to Slack and I gave Hermes agent the Grafana Cloud MCP server and I gave Hermes agent also gcx and locked it into Grafana Cloud. So let's see what we can do.

This is my Slack. You can see I'm in the war room channel and I can now directly interact with Hermes. I can tag Hermes here and I can ask, "Hey, something is wrong with my transfer service. Can you check why transfers are slow?" And Hermes starts working. It's very interactive. It lives in Slack like all the other agents that we can connect there. And we should briefly see that it has understood my intent and it says, "Hey, sure. I'm going to dig into this for you." And then it starts checking the firing alerts, the services, recent latency, error rates, et cetera. And it's investigating in the background. It's actually using gcx to drill deeper and deeper and deeper until it has an answer.

To show you what an example looks like, I have set up a bunch of scheduled checks. So it basically watches the telemetry in the connected Grafana Cloud environment. And let's look at this one here.

I interacted with Hermes before. I told you that it's an agent that can learn from you and how you do want to do things. And I teach it how I want my root cause analysis to be done. First, it should investigate, of course, but then it should give me a report on what happened and it did so.

It said, "Hey, the Watchdog rate was 7% something, something," and it pulled up the evidence. Where useful, it also linked back to Grafana Cloud so that I can validate because still agents, LLMs, we want to go check on things.

I can also see impact versus my baseline. And what I really find helpful is that I can also check what's next. So suggested follow-ups that Hermes tells me that I should do. And I really like this feature of Hermes. Let's see what our agent came up with here.

It actually digged up quite a bit, but it talks about perceived user performance. It understood what the telemetry means. And by the way, this is connected to an open model via an open provider. So it's GLM under the hood. So that also means with your agent, you're going to choose the model, but Grafana Cloud will still be there and you can connect any agent to Grafana Cloud.

Back to the slides, Gianni.

Gianni Baiardi: All right. So we started with a message in the Configure, right? Sorry. And then we checked Grafana Cloud using Claude on my phone. Then we moved to Cursor with the context from Grafana Cloud and we brought the investigation into Slack with Hermes. That's what we mean by bringing Grafana into your workflow.

You choose the agent, the model, the provider, you choose where you work. Grafana Cloud gives your agent access to the context and tools he needs to help. That's Grafana's big tent for agents. Thank you.