In this video, co-founder Alex MacLeod walks through how to connect an OpenSearch instance to an Aisle project. Once the connector is wired up, you can query your server logs in plain language — no query DSL required.
What's covered
- Adding OpenSearch as a connector in your Aisle project
- Writing a task that queries an index and surfaces relevant log entries
- Asking questions about your logs in plain language from the Aisle chat launcher
If you have logs sitting in OpenSearch, this is the fastest way to make them useful to the people who need them.
Video transcript
Today, I'm going to show you how to connect an OpenSearch instance to an Aisle project. This is a really nice way to search your server logs and correlate with other data. So, the first thing I'm going to do is go to connect and I'm going to create an OpenSearch credential. So, let's just call it OpenSearch. And my instance is hosted on Digital Ocean.
So, I'm going to paste in the correct information. 25060. The user is DO admin. And the password I have off-screen right here. Okay.
So, my credential is registered. Let's go create a project and hook this up. I'm going to go to projects, new project, and we'll call this server log explorer. First thing I want to do is manage tools and look for that OpenSearch connector. There we go.
It's shown right here. I'm also going to add some custom instructions. Now, you don't have to do this, but it will make the agent a little smarter. And I have some instructions that I have compiled over time. Mostly, what this does is it tells the agent what the indexes are in my OpenSearch instance.
And this will be different for your applications. So, it's good context to provide the agent so that it can make intelligent queries. All right. Now, we're ready to start asking questions. Uh so, one thing that I like to look for are overloaded errors from Anthropic.
This happens when Anthropic uh has too much traffic and they can't serve our requests, and they send back an error that just says, "Overloaded. We can't help you." So, we like to know when this is so that we can show an appropriate banner or contact people and let them know that, "Hey, this is an upstream issue." Uh I'm asking it to bucket the count by calendar date and generate a bar chart so that I can visualize when these errors are happening. Okay, so it's going to go ahead and search through my OpenSearch instance. And here's the result. We have a nice inline bar graph that shows us that a number of these errors occurred on June 23rd.
And above there's a table with some more specific information. So, what I would do next is say, "Well, what happened on June 23rd? Uh did Anthropic have a service issue?" So, it's going to do a more thorough targeted analysis and it's also doing a web search, which I assume means it's going to check the status page. And here we go. Indeed, yes, Anthropic had a service issue on June 23rd.
It was a capacity overload incident affecting multiple Claude platforms. So, I know that this is an upstream issue. I don't really need to worry about solving it. Perhaps I need to mitigate. Um but in this one instance, we can say, "Okay, yes, this was Anthropic."