Getting started
What do I need to get started?
What do I need to get started?
To get started, Cortex recommends configuring at least 1-2 metrics in Delivery, 1-2 metrics in Reliability, and 1-2 Scorecards in Initiatives or Vigilance. This requires deployment, version control (for pull requests), and incident data flowing into Cortex from the following supported tools:
For setup steps, see Configuring OpEx Review.
Data sources
What data feeds the report?
What data feeds the report?
The report uses all datasets available in Eng Intelligence:
- Version control
- Deployments
- Incidents
- Work items
- AI tooling
- Custom metrics
- Scorecard scores
How are production and non-production deployments distinguished?
How are production and non-production deployments distinguished?
Deployments can include optional environment information, which is used to separate production metrics from non-production metrics. To send environment information with your deployments, use the Cortex deploys API.
Does generating a report query my source systems or affect rate limits?
Does generating a report query my source systems or affect rate limits?
No. Cortex ingests and stores data from your integrations, and the report queries that stored data rather than your source systems. Generating a report adds no load or rate-limit pressure on your tools.
Will reliability, observability, and security metrics be added?
Will reliability, observability, and security metrics be added?
Additional Cortex data may become available to OpEx Review over time.
Data storage and security
What data does Cortex store, and how is it secured?
What data does Cortex store, and how is it secured?
Cortex stores the engineering metadata it ingests using the credentials you provide. That data is tenant-isolated and encrypted. Cortex maintains SOC 2 and ISO 27001 compliance.
Does Cortex store my source code?
Does Cortex store my source code?
No. Cortex stores engineering metadata and event data, such as pull request, deployment, and work item metadata, Scorecard scores, and trace events. It doesn’t store your source code or codebases.
How long is data retained?
How long is data retained?
Cortex retains out-of-the-box Eng Intelligence metrics for at least 6 months and custom metrics for at least 24 months. If you need a longer retention period, contact your Customer Success Manager.
How insights are generated
How are highlights and concerns identified?
How are highlights and concerns identified?
The agent works through a multi-step process across your engineering metadata and org structure:
- It retrieves your data through structured calls, so every figure traces back to a real query result.
- It evaluates that data within the DRIVE framework (Delivery, Reliability, Initiatives, Vigilance, and Efficiency).
- It ties each signal to the teams, services, and domains responsible, using your catalog and ownership data, and rolls results up the way your organization is structured.
- It compares trends against historical baselines, so the report surfaces meaningful changes rather than normal week-to-week noise.
How are AI-generated insights validated to prevent hallucinations?
How are AI-generated insights validated to prevent hallucinations?
Every metric in the report comes from Cortex’s query engine through structured calls. Before each report is created, a grounding step checks each AI-generated insight against the underlying data.
Which AI provider powers OpEx Review?
Which AI provider powers OpEx Review?
OpEx Review is powered by Anthropic’s Claude, using different models for different tasks.
Using Cortex AI in OpEx Review
What can I ask Cortex AI in OpEx Review?
What can I ask Cortex AI in OpEx Review?
Cortex AI is built for follow-up questions on whatever metric or insight you’re looking at, such as which teams are driving a change, how a metric compares to last quarter, or what a threshold means. Suggested prompts in the chat give you a place to start.For more examples, see What you can ask.
What should I expect from a Cortex AI response?
What should I expect from a Cortex AI response?
Responses are short (typically two to four sentences or a brief bulleted list) and are meant to be discussed in your OpEx meeting. They call out specific teams or services, suggest a next step when there is one, and link to a relevant data source so you can dig in further.Cortex checks any metric it surfaces against the query engine, but as with any AI-generated summary, it’s worth double-checking a number before it drives a decision. For more information, see What to expect from a response.