Aegis Orchestrator
Zaru

The Architect

Create, manage, and monitor agents and workflows from the Architect dashboard.

The Architect

The Architect is the management dashboard for agents and workflows, available at /dashboard in the Zaru client. From here you can browse, create, edit, and monitor everything running on the AEGIS platform within your tenant.


Agent Management

Browsing Agents

The agent list shows all agents visible to your tenant. Use the search bar to filter by name or description. Scope filters let you narrow the list:

  • My Agents — agents you created
  • Shared — agents shared within your organization (Business and Enterprise tiers)
  • Built-in — platform-provided agents deployed automatically

Each agent card shows the agent name, status (active, inactive, draft), and last execution time.

Agent Details

Click an agent to open its detail view:

  • Manifest — the full agent definition including runtime, security policy, resource limits, and role assignments.
  • Versions — a history of manifest versions. Each deployment creates a new version.
  • Execution History — a chronological list of all executions for this agent with status, duration, and links to execution logs.

Creating Agents

Click New Agent to open the creation interface. Three authoring methods are available:

  • Form Editor — a guided form for configuring the agent step by step. Fields include name, description, runtime, instructions, security policy, and resource limits.
  • YAML Editor — a text editor for writing the AgentManifest directly. Syntax validation runs as you type.
  • AI Generation — describe what you want the agent to do in plain English. Zaru generates the complete manifest, which you can review and edit before deploying.

Editing and Redeploying

Open an existing agent and click Edit to modify its manifest. Changes create a new version on deploy. The previous version remains in the version history.


Workflow Management

Browsing Workflows

The workflow list shows all workflows visible to your tenant. Use the search bar to filter by name or description.

Each workflow card shows the workflow name, state count, status, and last execution time.

Workflow Details

Click a workflow to open its detail view:

  • Manifest — the full workflow definition including states, transitions, and agent assignments.
  • FSM Visualization — a state machine diagram showing states, transitions, and terminal conditions.
  • Versions — a history of workflow manifest versions.
  • Execution History — a chronological list of executions with status, current state, and duration.

Creating Workflows

Click New Workflow to open the creation interface. The same three authoring methods are available as for agents — form editor, YAML editor, and AI generation.

Workflow manifests define a finite state machine: an initial state, one or more intermediate states with agent assignments, transition conditions, and terminal states.

Executing Workflows

From a workflow's detail view, click Execute. If the workflow defines an input schema, a parameter form appears for you to fill in before starting. Execution progress is streamed to the Glass Laboratory panel in real time.


Execution Logs

Every agent and workflow execution produces a log accessible from the execution history in the detail view. Logs include:

  • Timestamped entries for each tool invocation and its output
  • Iteration progress and validation feedback
  • State transitions (for workflows)
  • Final output or error details

Click any execution in the history list to open its full log.

The Execution Logs page (/dashboard/executions) has two tabs. Agents lists your agent executions from GET /v1/executions. Workflows lists your workflow runs from GET /v1/workflows/executions: each row shows the run's workflow name, or the first eight characters of its workflow id, and its duration ends at the run's last transition once the run has completed, failed or been cancelled. Click a row to open that run's page.

Each tab loads your twenty newest at first. Load 20 more under the tab's table appends the next twenty, and is offered while the last page came back full. When a later page fails, the rows already shown stay and the tab says "Zaru could not load more executions. Try again in a moment."

A run's page opens its narrative with the repository line of each repository the run prepared, "Prepared repository <label> at /workspace/<label>: branch aegis/<id> created from <sha>", as the orchestrator wrote it, before the run's steps; an agent run's, before its first try.

A workflow run's page shows the gate the run waits at, if any, as the gate card, between the run's status line and its tabs, so you can answer the gate there when you open the run by its link (Human Gates of a Workflow Run).

On this page