Control how AI teammates access knowledge and tools

Alloy work happens inside an organization. People join through memberships and invites. AI teammates, storage, variables, secrets, MCP servers, and workflows are managed within that organization.

This gives teams one operating boundary for deciding who belongs in the workspace and what work they can do there.

AI teammates are configured with a role, instructions, tools, model settings, and optional voice setup. They work like specialists with a job and limits.

Teams can decide which tools and MCP servers each AI teammate can use, instead of giving every teammate the same access by default.

Storage lets teams upload files, create folders, search knowledge, and share folders with AI teammates.

Ally has automatic full storage access. Other AI teammates are limited to the folders explicitly shared with them. Folder access can be read or write, and inherited folder permissions stay visible in the sharing flow.

Variables and secrets help teams manage reusable configuration and credentials. Secret values are masked in the table and exports, and editing a secret requires entering the new value instead of revealing the old one.

This supports safer integration setup for URLs, headers, and workflow steps.

MCP servers are registered at the organization level and assigned to specific AI teammates. Teams can test a server and then enable it for the teammates that need that capability.

Runtime connection details can use organization variables and secrets, so teams can keep integration configuration centralized.

Autonomous AI work can carry scoped runtime context through the work it starts. When an AI teammate calls a structured skill or workflow, that child run can inherit context from the parent run and use it to limit what the action can see or query.

For example, an app can send Alloy an organization ID, an AI teammate can call a structured skill that searches an external logging system, and that skill can scope the search to only that organization ID. The AI teammate still completes the work, but the connected action is constrained by the context passed into the run.

Alloy is designed for people and AI teammates to work together. Humans can step in when a situation needs judgment, and workflows can include user input, conditions, review points, and escalation.

The goal is controlled collaboration, not blind automation.