Skip to main content
Cortex AI is Cortex’s AI-powered chat assistant. It answers questions about your engineering data in plain language, using the same data that powers your Scorecards, dashboards, and reports. Questions that would otherwise become a research task, like who owns this service, which teams are behind on a Scorecard, or how deployments are trending over the last quarter, get answered where you are, without assembling the data yourself. Every answer is grounded in your real Cortex data.
Cortex AI is not limited to the entities you can see
  • Anyone with the Enable AI Chat permission can ask about any metric or Scorecard score for any entity in your workspace, including ones hidden from them elsewhere in Cortex.
  • Your own conversations stay private.
​Treat the Enable AI Chat permission as access to your organization’s engineering data as a whole, and grant it accordingly.
Cortex AI chat is in Research Preview
  • Responses can be slow, especially for questions spanning many teams or entities.
  • Not all Cortex data is reachable from chat yet, so it can’t answer every question about your organization.
  • Apart from building Eng Intelligence modules and dashboards you ask for, it can’t take action on your behalf. It won’t create or edit Scorecards, Initiatives, or Workflows, and it won’t change the report’s content or filters.
​We’re actively working on improvements. Share feedback with your account team or through the in-product feedback controls.

Accessing Cortex AI

​Admins and users with the Enable AI Chat permission can use Cortex AI. The same permission covers chat inside OpEx Review. It does not cover the Slack AI assistant or the Cortex MCP, which have their own access. ​There are two ways to access Cortex AI chat, and they share the same conversations:

​Ask Cortex AI

​Ask Cortex AI sits at the bottom of the main sidebar and opens chat in a panel beside whatever you’re looking at, so you can ask a question without losing your place.
  1. ​Click Ask Cortex AI in the main sidebar, or press ⌘J (Ctrl+J on Windows and Linux).
    'Ask Cortex AI', highlighted at the bottom of the main sidebar.
  2. Enter your question, then press Enter.
​The side panel follows you as you navigate. On an entity or Scorecard page, Cortex AI grounds your next question in what you’re looking at. This means you can ask “who owns this?” or “why is this failing?” without naming the thing. Move to a different page and the line updates, so you always know what your next question will be about. ​Press the shortcut again to close the panel. The conversation stays as you move around Cortex, so closing and reopening the panel picks up where you left off.

Cortex AI

Cortex AI in the main sidebar opens a dedicated page built for longer sessions, with your conversation history beside the chat.

What you can ask

Cortex AI is built for the questions that come up while you’re working, rather than for a fixed set of reports. For example:
  • Who owns payments-service, and when was it last deployed?
  • Which services are failing our production readiness Scorecard, and what’s the most common failing rule?
  • How has deployment frequency changed for my teams over the last quarter?
  • Which Workflows have been failing this week?
  • What’s the status of our Go 1.22 migration Initiative?
  • How do I set up a custom Scorecard rule?
  • Build a module showing daily PR cycle time for the last 30 days as a line chart.
Cortex AI can reach:
  • Your catalog - Entities, ownership, teams, and entity details
  • Scorecards - Definitions, scores, and how entities are performing against them
  • Initiatives - What’s in flight and how it’s progressing
  • Workflows - Recent runs and which ones are failing
  • Eng Intelligence metrics - Deployment frequency, cycle time, incident metrics, and custom metrics, broken down by team, entity, or time period
  • Cortex documentation - How a Cortex feature works, answered from the public docs
To compare what Cortex AI can reach with the Slack AI assistant and the Cortex MCP, see Data available in each channel. See the Cortex AI prompt library for a full list of prompt patterns organized by use case and role.

Managing your conversations

On the Cortex AI page, your past conversations are listed beside the chat.
The Cortex AI page with past chats highlighted.
To access the Cortex AI page, select Cortex AI from the main sidebar. From this page you can:
  • Select a chat to pick it back up.
  • While in a chat, click the overflow menu icon in the upper-right corner to rename or delete it.
  • Click New chat to start a fresh conversation.
A conversation you started from an OpEx Review report includes a link back to the report it originated from.

Who can see your chats

Your conversations are private to you. Nobody else in your organization can see them, including admins, and they aren’t visible to anyone you send the URL to; a conversation link only opens for the person who created it.

Building Eng Intelligence modules and dashboards

Describe the view you want in plain language, and Cortex AI builds an Eng Intelligence module or a full custom dashboard for you, without configuring Data Explorer yourself. For example:
  • Build a module showing PR cycle time by team for the last 30 days as a bar chart.
  • Build me a delivery health dashboard with deploy frequency, PR cycle time, and merged PRs for the last 30 days.
Before proposing anything, Cortex AI checks the configuration against your data and your permissions, so what it proposes loads correctly once you save it. It then shows a Draft card that describes what it will create:
  • Module cards show the module’s name and a summary of its metric, grouping, date range, and chart type. Click Preview to see the chart with your data. To fine-tune the module yourself, click Edit in Data Explorer from the preview, make your changes, then save the module from Data Explorer.
  • Dashboard cards show the dashboard’s name and description, the filters that apply to every module, and the list of modules it will include.
Changes you make in Data Explorer don’t update the card in chat. If you edit a module in Data Explorer and then click Save module on the card, Cortex saves the original version from chat, not your edits.
When you ask an Eng Intelligence question in chat, Cortex AI may offer to turn the answer into a module.
Unlike chat answers, modules and dashboards that Cortex AI builds are checked against your own permissions in Cortex. If a module would need data you can’t access, Cortex AI tells you why instead of proposing it.

Refining a proposal

To change a proposal, ask in the same conversation, for example “group by repo instead” or “make it last quarter.” Cortex AI shows a new card with the updated module or dashboard. Earlier cards stay in the conversation, and you can still save any of them.

Saving a module or dashboard

Users with the Edit Eng Intel Dashboards permission can save a module or dashboard from chat. Without this permission, the save button doesn’t appear on the card. On the card you want to keep, click Save module or Save dashboard. Before saving, Cortex checks the configuration again. If something changed since the card was created, such as a metric being removed or your access changing, Cortex tells you why and doesn’t save. Otherwise, Cortex creates the module or dashboard with you as its creator and marks the card Saved with a link to it. From there, you can manage it like any other module or dashboard. See Saving the view and Managing custom dashboards. Dashboards that Cortex AI builds work like this:
  • Cortex AI can include up to eight modules in a dashboard it builds.
  • Filters that cover the whole request, such as an organization or a date range, are set once on the dashboard and apply to every module. A filter you ask for on one module stays on that module. See Using dashboard-level filters.
  • If any module can’t be built, Cortex AI explains why instead of proposing the dashboard.
  • Users can select which modules to include when saving a new dashboard by selecting the checkbox next to each module.

Limitations

Line, vertical bar, and KPI charts show one metric. Tables can show several. If you ask for more metrics than a chart supports, Cortex AI suggests the closest chart type that fits.
Last modified on November 2, 2026