Build & Run Agents·production

IAM Agent

Desktop, Web and CLI workspace for Chat, Cowork, Code and MultiWork.

Landing Application

Key capabilities

  • Chat and project-context workflows
  • Agent and Cowork execution with approvals
  • MultiWork orchestration and task artifacts

Step-by-step guides

Use cases

Start with the outcome: open a guide, prepare prerequisites, follow the steps and verify the success signals.

01 Complete a repository task with Chat/CoworkThe agent works in the selected project, presents a plan and asks before making changes.
Audience
Developer or analyst
Outcome
The agent works in the selected project, presents a plan and asks before making changes.

Before you start

  • Desktop installation or Web access
  • Open project
  • Selected permission profile

Steps

  1. Select a workspaceOpen the project and verify its root, branch and available tools.
  2. Describe the desired resultState the goal, constraints, verification and files that must not change.
  3. Review the planResolve ambiguous assumptions before actions begin.
  4. Accept changesReview the diff and test results, then accept or request corrections.
Verify the result

The result lists changed files, checks and remaining risks; the working tree matches expectations.

If it does not work

If context is wrong, stop and reselect the workspace; do not silently expand scope.

02 Split a large outcome into parallel MultiWork tasksIndependent tasks have owners, dependencies and verifiable outcomes.
Audience
Tech lead or product owner
Outcome
Independent tasks have owners, dependencies and verifiable outcomes.

Before you start

  • A clear overall outcome
  • Known repository boundaries
  • Available agent workers

Steps

  1. Create a boardIn MultiWork, add the outcome, acceptance criteria and shared risk.
  2. Create independent cardsGive every card a scope, owner, inputs, outputs and dependencies.
  3. Start ready tasksDo not start a card until its upstream dependency has produced its artifact.
  4. Integrate resultsCheck conflicting diffs, run shared regression and complete the acceptance checklist.
Verify the result

Every completed card has evidence and the combined result passes shared regression.

If it does not work

If tasks conflict often, reduce shared file scope and assign explicit ownership.

Application sections

Open detailed manual

Every application screen has a separate page with controls, safe example values, CLI/API alternatives and status-specific recovery steps.

Open detailed manual →

Role in the ecosystem

IAM Agent - a custom surface for tasks in which the AI ​​should not only to answer, but also to perform a sequence of actions with files, tools and external services. The interface preserves the context of the task, shows intermediate results and asks for confirmation for sensitive operations.

Operating modes

  • Chat - research, explanation and preparation of materials;
  • Cowork - joint work on a task with a visible plan and artifacts;
  • Code — work with the repository, tests and local tools;
  • MultiWork — parallel execution of independent work threads.

Main scenario

The user creates a task, adds files or links and selects an available one execution circuit. Agent receives only allowed tools and executes steps, applies the testable result and leaves the user in control changes. Continuing a task uses its own context rather than the general one memory of other users.

Integrations and security

Model routing can be done via IAM.Router, content checking is done via IAM.Secure, and discovery capability

  • via IAM.Core. Credentials are passed as scoped references and are not included in the dialogue history. Actions that change the outside world must be clearly visible and, where required, confirmed by the user.

Operational boundary

Agent is a production client and not a perpetual background execution environment. Published and scheduled agents are hosted via IAM Marketplace and IAM.Hosting.