Create custom AI agents in Hanzo Chat with system prompts, model selection, and MCP tool access.

Agents

Agents are custom AI personas with specific system prompts, model preferences, and tool access. They appear as selectable options in the Chat UI.

Creating Agents

Via chat.yaml

Define agents in your chat.yaml configuration:

agents:
  - name: "Code Reviewer"
    model: "zen4-coder"
    systemPrompt: |
      You are a senior code reviewer. Analyze code for:
      - Bugs and logic errors
      - Security vulnerabilities
      - Performance issues
      - Style and readability improvements
      Always explain your reasoning and suggest fixes.
    tools:
      - filesystem
      - web

  - name: "Research Assistant"
    model: "zen4-pro"
    systemPrompt: |
      You are a research assistant. Search the web, analyze
      documents, and provide comprehensive answers with citations.
      Always cite your sources.
    tools:
      - web

  - name: "Writing Editor"
    model: "zen4"
    systemPrompt: |
      You are a professional editor. Review writing for clarity,
      grammar, tone, and structure. Suggest specific improvements
      while preserving the author's voice.

  - name: "Data Analyst"
    model: "zen4-thinking"
    systemPrompt: |
      You are a data analyst. Help users understand data, create
      visualizations, write SQL queries, and interpret results.
      Think step-by-step through complex analysis.
    tools:
      - filesystem

Via UI

Open Chat settings

Go to Agents section

Click Create Agent

Configure name, model, system prompt, and tools

Save — the agent appears in the model selector

Agent Configuration

Field	Description	Required
`name`	Display name in the UI	Yes
`model`	Default model for this agent	Yes
`systemPrompt`	Instructions prepended to every message	Yes
`tools`	List of MCP tool servers the agent can use	No
`description`	Short description shown in the UI	No
`icon`	Custom icon URL	No

MCP Tools for Agents

Agents can use MCP (Model Context Protocol) tools for extended capabilities. First define the MCP servers, then reference them in agent configs:

mcpServers:
  filesystem:
    command: npx
    args: ["@hanzo/mcp-filesystem"]
    env:
      ALLOWED_DIRS: "/data"

  web:
    command: npx
    args: ["@hanzo/mcp-web"]

  github:
    command: npx
    args: ["@hanzo/mcp-github"]
    env:
      GITHUB_TOKEN: "${GITHUB_TOKEN}"

agents:
  - name: "DevOps Agent"
    model: "zen4-coder"
    systemPrompt: "You are a DevOps engineer..."
    tools:
      - filesystem
      - github

See MCP Tools for the full list of 260+ available MCP tool servers.

Agent Presets

Agents can also be saved as presets for quick switching:

Configure an agent in model settings

Adjust temperature, top_p, and other parameters

Click Save As Preset

The preset preserves both the system prompt and parameters

Built-in Agents

Hanzo Chat includes several pre-configured agents:

Agent	Model	Purpose
Default	zen4	General-purpose chat
Coder	zen4-coder	Code generation and review
Analyst	zen4-thinking	Deep analysis and reasoning
Quick	zen4-mini	Fast responses for simple tasks

Agents

On this page