Enterprise AI

Enterprise AI is the governed connection between a user’s objective, approved organizational context and the Cortex work surface where the task begins. It combines selected workspace or browser context, MCP connector tools, organization policy, model profiles, search and engineering skills while keeping one durable session and audit trail.

The enterprise connection layer supports ten MCP connector families: GitHub, Atlassian, Slack, Microsoft 365, Google Workspace, AWS, Microsoft Azure, Azure DevOps, Salesforce and Google Cloud. See MCP connectors for the services behind each connection and the authorization flow.

Get started

Choose the relevant MCP connectors, follow connector setup and confirm a permitted read operation. Configure governed actions before enabling operations that change enterprise data.

What Enterprise AI coordinates

• Model routing chooses an allowed model profile for the request.
• Search routing decides when current public information is needed.
• Skill routing applies reviewed engineering guidance to the task.
• Planning separates analysis from actions and identifies approval points.
• MCP connectors supply scoped tools for enterprise systems without placing provider credentials in the prompt.
• Cortex Connect preserves the objective and progress when work moves between mobile, browsers and a connected development workspace.
• Cortex Cloud can continue eligible work in a managed execution environment with validation and release evidence.

Common workflows

Enterprise context can connect a Jira requirement with a GitHub repository, summarize a Slack incident before examining AWS or Azure resources, relate a Salesforce case to engineering work, or carry research from a browser into VS Code. The exact sequence is bounded by the user’s service permissions and the organization’s enabled tools.

Objective	Relevant context	Potential outcome
Implement a requirement	Jira, Confluence and an authorized repository	An implementation plan or reviewed project work
Investigate an incident	Slack discussions and scoped cloud resources	A contextual explanation and approved next steps
Understand a customer issue	A Salesforce case and engineering records	A relationship between the customer request and product work
Continue research in development	Selected browser sources and a connected workspace	A preserved objective with supporting context

These are examples of how supported connections can be combined. Each task uses the services and operations enabled for the acting user.

Context, actions and results

Context supplies information for a decision; a connector action performs a specific operation in an external service. Cortex can use the first without automatically gaining permission for the second. Required approvals remain part of the task when an operation changes records, messages or resources.

The visible result may be an explanation, plan, service response or project artifact. A task using Cortex Cloud can also return validation or release evidence tied to its project revision. The selected tools and execution target determine the available output.

From task direction to execution planning

Routing and planning solve different problems. Routing identifies relevant intelligence, current public information or engineering guidance. Planning defines the activities required to reach the objective, including dependencies, acceptance conditions and the checks needed before completion.

For a security-sensitive change, the plan might separate investigation, remediation, regression tests, independent review and integration. For a multi-area feature, permitted independent work can proceed together while dependent steps wait. If more scope is discovered, the workflow can bring it back for review rather than hiding it in an optimistic completion summary.

Ten connections do not mean one permission boundary

The connector families provide different service contexts and tool sets: source and delivery, project knowledge, communication, productivity, cloud operations and CRM. A useful task selects the relevant services and resources instead of indiscriminately attaching everything connected to the organization.

A Jira requirement and GitHub change may be enough for implementation planning. A customer-impact investigation may also need an authorized Salesforce case. An operational incident may need a Slack thread and scoped AWS, Azure or Google Cloud resources. Each step uses its own service authorization and remains subject to permitted reads, writes and review.

Continuity, evidence and human decisions

Cortex Connect makes a supported objective and its progress available across clients. A browser or mobile request can reach an available connected VS Code workspace; Cortex Cloud supplies managed execution for eligible tasks. Neither path changes the authority granted by the user or organization.

Workflow views distinguish discussion from execution state. Collections keep related chats together, forks explore alternatives, and a governed workflow records the work performed, validation evidence and decisions still needed. Completion depends on what was checked, not only on the quality of the final explanation.

Product background

• Planning coordinated engineering work
• Connecting Salesforce with enterprise workflows
• Governed workflows and reviewable results

Administration

Admins define which model routes and profiles are available on each surface, which credentials can be used, and whether personal or deployment-managed provider accounts are permitted. They can also restrict writes, require approval for consequential operations, define retention and choose eligible execution surfaces. Users see only the effective options.

Use the least context needed for the objective. A connector does not make all of its data available to every prompt; identity, scope, resource permissions and workflow policy are evaluated for each operation.

Start with MCP connectors, follow the connector setup process, and define governed actions before enabling writes.