Skip to content

Admin Dashboard: Agents

The Agents page is your central hub for creating, configuring, and managing AI agents. Each agent is a self-contained entity with its own model, personality (system prompt), tools, and memory scope.

Agent list view

The main view shows a table of all configured agents with key information at a glance:

Name — unique identifier (lowercase, alphanumeric, hyphens only).
Tools count — total number of tools available to the agent.
Knowledge — whether a knowledge base is configured (RAG).
Capabilities — active capabilities (Memory, Knowledge).

Click any agent row to open a side panel with the full configuration. From there you can edit settings, view tools by security zone, or delete the agent.

Creating an agent

Click “Create Agent” to open the agent form. Here is a walkthrough of each field:

Field	Default	Description
`name` *	—	Unique identifier. Lowercase, alphanumeric + hyphens. Used in API endpoints and spawn references.
`model` *	—	Dropdown populated from configured models. Determines the LLM backend.
`system` *	—	System prompt that defines agent behavior. This is the most important field.
`lifecycle`	`persistent`	`persistent` = accumulates context across sessions. `spawn` = fresh context each time.
`tool_execution`	`sequential`	`sequential` = one tool at a time. `parallel` = concurrent tool execution.
`max_steps`	`50`	Maximum reasoning iterations (1-500). Higher = more complex tasks, more tokens.
`max_context_size`	`16000`	Context window in tokens (1,000-200,000). Older messages are compressed when exceeded.
`tools`	`[]`	Select from available built-in tools (show_structured_output, manage_tasks, etc.) and custom HTTP tools.
`mcp_servers`	`[]`	Multi-select from configured MCP servers.
`can_spawn`	`[]`	Which other agents this one can create at runtime.
`confirm_before`	`[]`	Tools that require user confirmation before execution.

YAML equivalent

Everything configured through the form can also be expressed in YAML:

agents:
  my-agent:
    model: glm-5                    # Model dropdown
    system: |                       # System prompt
      You are a sales consultant for Acme Corp.
      Always be professional and helpful.
    lifecycle: persistent           # persistent | spawn
    tool_execution: parallel        # sequential | parallel
    max_steps: 100                  # 1-500
    max_context_size: 32000         # 1000-200000
    tools:
      - knowledge_search
      - show_structured_output
      - manage_tasks
    mcp_servers:
      - github-server
    can_spawn:
      - researcher
    confirm_before:
      - create_order

What’s next