AI Insights

Buildnote has two AI features that work with your collected data. Failure summaries explain a broken test or step in a sentence or two, and they appear on the event page automatically. AI Insights is a chat: you ask a question in plain English, the assistant queries your data, and it answers with live charts, complete dashboards, collected files or a build's pipeline timeline, right in the conversation.

Both features read the same events as the rest of the app. Nothing is sent to a model until you enable the feature and connect a model provider.

Turning AI on

You need three things in place.

First, your team needs a plan that includes the AI features, which means Pro or Enterprise. Second, go to Settings → General → Event Collection and switch on AI Insights ("Use AI to gain better insights (e.g. summarise failures or detect flaky tests)"). Admins and Billing Managers can change this switch. Other members can see whether it is on.

Third, connect a model provider under Settings → Integrations. Buildnote supports Anthropic, Google Gemini and OpenAI, and each has its own step-by-step guide. When you create the integration, you enter a name, the provider API key and the model to use. The dropdown lists the current models with the recommended one first. The key is stored encrypted and is used only for your team's requests. To switch models later, edit the integration. Nothing else in the app needs to change.

AI requests are billed on top of your plan, and every reply shows exactly what it cost. See the pricing page for the details of each plan.

Failure summaries

When AI is enabled, failed test and build events are summarised as they arrive. When you open a failed event, the status panel shows the summary first instead of the raw output. The summary tells you what failed, the likely reason and the category of failure. The original message and stack trace are one click away.

Summaries are meant for triage, not full diagnosis. They help you quickly tell whether a failed build is an infrastructure blip, a real regression or an unstable test, so you can decide whether to stop what you are doing. See Exploring Builds for where the panel appears on an event.

Asking a question

Open AI Insights from the left navigation and type your question in the box at the bottom. For example:

Which test suites failed most often this week?
Why was the last main-branch build so slow?
Show me the coverage report from the most recent release build.
Build me a dashboard of pipeline duration and failure rate by project.

Send it with the button or ⌘↵ (Ctrl+↵ on Windows and Linux). While the assistant works, you can follow its progress step by step. If you asked the wrong thing, Stop ends the run straight away.

The assistant doesn't guess what your data looks like. It loads the query language reference, writes a query, validates it, runs it against your events and fixes the query if the result is wrong. This is why a reply takes a few rounds instead of arriving instantly.

Include the period you care about in your question, such as "last month" or "since Monday". The chat has no time-range control, and unless you say otherwise, the assistant looks at the past week.

What comes back

A reply mixes text with rendered content, in the order the assistant produced them, so each chart appears under the sentence that introduces it.

Charts, tables, metrics, gauges and markdown panels are live. Each one keeps the query that produced it and runs that query against your data when the message is displayed. When you reopen an old conversation, you see current numbers, not an outdated snapshot.

Dashboards come back as a complete board when you ask for one. The widgets are already laid out on the grid and shown read-only, exactly as they will look once saved.

Copy as dashboard saves that board to your team's dashboards and takes you to it, where you can edit it like any other board. In the chat, each query keeps its own time bounds because it runs live there. When you copy the board, those bounds are removed, so the saved board follows its own time-range filter instead.

Files collected during builds are shown in the reply. Ask for a report, a log or a screenshot, and the assistant finds it in your data and previews it with the same viewer the build pages use.

Build breakdowns show the pipeline waterfall for one build: its pipelines, jobs and steps drawn against the build's total duration, with statuses and timings. When you ask where a build spent its time, you get this timeline instead of a table of rows, along with a link to the build page.

If the assistant can't find something, it tells you instead of making up an answer. For example, if you ask about a build that reported no pipeline stages, it says so explicitly.

Conversations

Conversations are saved per user. The sidebar lists yours, newest first, with a search box. New insight starts a new conversation. Each conversation has its own URL, so you can bookmark an investigation or send it to a colleague on your team.

You can rename a conversation from its menu when the automatic title no longer fits, and delete conversations you no longer need. When you reopen a conversation, its full history is shown again, including text, widgets, dashboards, files and breakdowns.

Follow-up questions include the conversation's history, which is why a request like "same query but group by suite" works. Start a new conversation when you change topic, so an unrelated question doesn't pick up the wrong context.

Seeing what the assistant did

Every reply has a footer that shows how long it took and how many steps it needed. Click the step count to open a popover that shows each step the assistant took: what it looked up, each query it validated and ran, what it chose to render, and the tokens and time each step used. While a reply is in progress, the steps appear live.

This is useful when an answer looks wrong, because you can see the exact query the assistant ran. Each reply also ends with the key query in a code block. Hover over the block to see two buttons: one copies the query so you can paste it into the Query editor or a dashboard widget, and the other downloads the full result set as CSV.

The bottom of the popover shows totals for the reply: input and output tokens, how many input tokens came from the provider's cache, the number of model round trips and the total time.

Getting better answers

Be specific about the time window, the branch and the unit you want. "Failure rate by module on main over the past 30 days" gives the assistant everything it needs. "How are we doing?" does not.

Ask for one thing at a time. A focused question gets you one well-chosen chart, while a broad question gets you several weaker ones. If a reply is close but not quite right, say what to change, such as "same query but group by suite", instead of repeating the whole question. The conversation remembers what came before.

If you want a dashboard, ask for it in its own question. The assistant then builds a laid-out board you can save, which is a better starting point than copying widgets across one by one.