Get started
Cortex
AI coding agent for secure, review-ready software delivery: plan, build, validate, and remediate code with confidence.
Cortex brings AI assistance, code analysis, connected enterprise context and reviewable execution into the tools where you work. A conversation can begin with a question, develop into a plan, and continue through a proposed change, validation and a review of the result.
Get started
Section titled “Get started”Choose a client below, sign in, and follow the quickstart. Your Pervaziv account connects the experience; available features depend on your client, plan and organization.
Where you can use Cortex
Section titled “Where you can use Cortex”What Cortex provides
Section titled “What Cortex provides”Coding and project understanding
Section titled “Coding and project understanding”Cortex can explain selected code, investigate a repository, suggest fixes, generate tests and documentation, and review changes. Developer Tools connect these tasks to files and editor selections, while browser clients bring in relevant web and repository context. Proposed edits can be inspected as diffs before they become part of your work.
Security and remediation
Section titled “Security and remediation”File and repository scans identify security findings, while software component analysis examines dependency risk. Eligible clients also expose AI Security Review, AI Threat Model and workflows that connect a finding to a proposed patch and verification. Findings and test results are separate outputs: a suggested remediation still needs review and validation.
Connected enterprise work
Section titled “Connected enterprise work”Enterprise AI brings approved service context into Cortex through MCP connectors. A task can relate a Jira ticket to source code, use Confluence documentation, or investigate operational information from a cloud service. The connected identity and service permissions determine what Cortex can retrieve or change.
Work that continues across tools
Section titled “Work that continues across tools”Cortex Connect connects supported clients around a shared objective and saved progress. Mobile and browser experiences can initiate or follow work that needs an available development workspace. Cortex Cloud provides managed execution for eligible tasks, including validation and release evidence.
Choose the right Cortex experience
Section titled “Choose the right Cortex experience”| Your work | Start here |
|---|---|
| Understand or change code in an open project | Developer Tools |
| Research and act across browser tabs | Discover |
| Work alongside an existing web application | Browser Extensions |
| Continue a conversation or follow a task away from your desk | Mobile |
| Work from a terminal, script or CI job | CLI |
| Run eligible project work in a managed environment | Cloud |
Intelligence and user control
Section titled “Intelligence and user control”Cortex combines model, search and skill routing with planning. These capabilities help match coding, research, security and validation tasks to the available intelligence and tools. Users work with the objective and the visible result; organizations can restrict model choices, connector access and execution targets.
Selected files, pages and service results provide context for the task. Privacy and prompt-protection settings, saving preferences and approval controls vary by client. Use the options exposed in your client and review changes or consequential actions when prompted.
Organize conversations without losing the original
Section titled “Organize conversations without losing the original”Collections group related chats around a project, investigation or customer issue. A collection can hold a design discussion, an implementation request and a follow-up review without treating them as the same task. In supported VS Code, browser and mobile clients, you can return to those conversations and move them as priorities change.
Fork a Chat creates an independent conversation from completed history or a selected completed response. Use a fork to compare approaches, revisit an assumption or investigate an alternative without changing the original discussion. A fork is not a repository branch and does not approve code changes; any subsequent execution still uses the normal permissions.
From a request to governed work
Section titled “From a request to governed work”For eligible engineering objectives, Cortex can coordinate investigation, implementation, testing, review and verification in a workflow. The plan identifies dependencies, progress and decisions that need human input. Independent activities can proceed together where permitted, while dependent steps wait for the results they need.
The conversation describes the objective; the workflow records what happened. Task Center and connected workflow views show work that is running, paused, awaiting approval or ready for review. An interruption need not discard the whole effort. A changed requirement, failed check or incompatible project change can put affected work back under review while useful progress remains available. Repository-event integration, where enabled, can connect the task to pull-request, CI and security progress.
Coding standards and meaningful tests
Section titled “Coding standards and meaningful tests”Personal Coding Style lets supported clients use a Style Profile for comment density, documentation format, naming, error handling, typing, function size and test style. These preferences help generated work fit the existing project; they do not replace repository requirements or reviewer judgment.
Test Design Specification helps establish what behavior must be proven before implementation. A focused unit test may be appropriate for isolated logic; a workflow crossing identity, services or a user interface may require an integration or end-to-end check. Ask Cortex for acceptance conditions and failure cases before asking for more tests. Code coverage alone is not proof that a requirement works.
Verification then examines the proposed change against the objective and the available evidence. A patch can remain incomplete when an acceptance condition is missing, even if a narrow test passes. See validation for the distinction between generated output and a supported result.
Privacy before sharing context
Section titled “Privacy before sharing context”Eligible clients provide Quick Scan, Deep AI Privacy Scan and file-sanitizing controls before content enters a broader AI workflow. Privacy checks can flag sensitive spans such as credentials, personal information and tokens; prompt protection addresses untrusted instructions in pages, files and other inputs. Use the warnings and available redact, cancel or continue choices deliberately.
Local privacy or prompt-risk checks do not mean all Cortex reasoning runs offline. Coding, connected services and managed execution have their own network and authorization requirements. Review your client’s settings and organization policy before adding confidential context.
Product background
Section titled “Product background”- Collections, forks and governed workflows
- Personal Coding Style and Test Design Specification
- On-device privacy and prompt protection