Agent Builder PRO

The Agent Builder lets you create, configure, and deploy custom AI agents without writing configuration files by hand. Choose between AI-assisted generation or a manual form, then preview the generated files and deploy directly to your VPS.

Two Creation Modes

The Agent Builder offers two distinct workflows for creating agents. Use whichever fits your style — the end result is the same set of configuration files deployed to your VPS.

AI-Assisted Mode

Describe the agent you want in natural language, and the AI generates the full configuration for you. For example:

"Create a customer support agent that specializes in billing questions. It should be friendly and professional, use the support-bot emoji, and have access to coding and browser tools. Route it to the web-chat and Slack channels."

The AI parses your description and generates all configuration files: the agent name, personality definition (SOUL.md), identity settings, tool selections, model assignment, and channel routing. You can review and edit every generated field before deploying.

AI-assisted mode is ideal when you know what you want but do not want to fill out every form field manually. It is also a great starting point — generate the baseline, then fine-tune individual settings in the form editor.

Manual Form Mode

The manual form gives you direct control over every configuration option through a structured, section-by-section interface. Each section corresponds to a part of the agent's configuration.

Configuration Sections

Agent Name

The agent name is used as the directory name on your VPS, so it must be filesystem-safe. Use lowercase letters, numbers, and hyphens only. Spaces and special characters are not allowed. Examples: support-bot, sales-assistant, billing-helper-v2.

Personality / SOUL.md

The personality section defines how your agent communicates. This content becomes the agent's SOUL.md file — the core identity document that shapes every response the agent generates.

Write in the second person, describing how the agent should behave:

"You are a friendly, knowledgeable support specialist for Acme Corp. You answer billing questions clearly and concisely. When you do not know the answer, you say so honestly and offer to escalate to a human agent."

The large textarea supports multi-paragraph definitions. For teams that want a starting point, click the AI Generate button to have the system draft a personality based on the agent name and any context you have provided.

Tip: A well-written SOUL.md is the single most impactful configuration for agent quality. Spend time defining the tone, boundaries, and domain expertise clearly. Vague personalities produce vague responses.

Identity

Identity settings control how the agent presents itself in conversations:

Display Name — The human-readable name shown to users in chat interfaces (e.g., "Acme Support")
Theme — A color theme applied to the agent's chat widget
Emoji — An emoji displayed alongside the agent's name in the dashboard and chat headers

Tools

Select which tool categories your agent has access to. Tools are organized into checkbox groups:

Coding — Code execution, file reading/writing, terminal access
Browser — Web browsing, URL fetching, screenshot capture
Memory — Persistent memory across conversations, note-taking
Session — Session management, context tracking, conversation history
Subagent — Ability to delegate tasks to other deployed agents

Enable only the tools your agent needs. An agent that answers FAQ questions does not need coding tools. A developer assistant probably needs coding and browser tools but not subagent delegation.

Model

Assign a primary model and an optional fallback model to your agent:

Primary Model — The model used for all requests by default. Select from the models available on your instance.
Fallback Model — Used automatically if the primary model is unavailable or returns an error. This ensures your agent remains responsive even during model provider outages.

Both fields are dropdown selectors populated with the models configured on your VPS. If you are unsure which model to choose, use the Model Playground to compare outputs before deciding.

User Context / USER.md

An optional USER.md file that provides additional context about the end users who will interact with this agent. For example, you might specify:

"Users are small business owners who are not technical. Avoid jargon. Explain concepts in simple terms. Most users are in the US Eastern timezone."

USER.md helps the model tailor its communication style to your audience without cluttering the SOUL.md with user-specific details.

Channel Routing

Select which channels this agent should be available on. This is a multi-select field listing all channels connected to your instance. An agent can be routed to one, several, or all channels.

Channel routing determines where users can reach the agent. An agent routed only to Slack will not appear in your web chat widget, and vice versa. You can change routing after deployment without redeploying the agent.

Preview Generated Files

Before deploying, review the configuration files that will be written to your VPS. The preview panel displays each file in a tabbed interface:

SOUL.md — The personality definition
USER.md — The user context (if provided)
agent.json — Tool configuration, model assignment, and identity settings
routing.json — Channel routing configuration

Each tab shows the exact file content that will be deployed. You can edit the preview content directly if you need to make last-minute adjustments.

Auto-Save Draft

The Agent Builder automatically saves your work as a draft as you fill out the form. If you close the browser or navigate away, your in-progress configuration is preserved. Return to the Agent Builder and your draft is restored exactly where you left off.

Drafts are saved per-agent-name. You can have multiple drafts for different agents simultaneously.

Deploy to VPS

When you are satisfied with the configuration, click Deploy to push the agent to your VPS. The deployment process writes the configuration files to the correct directory and registers the agent with your OpenClaw instance. The agent becomes available on its routed channels within seconds.

Deployment status is shown in real time. If any step fails, the error is displayed with actionable details. You can fix the issue and retry without losing your configuration.

After deployment: Monitor your agent's performance with the Analytics module and review its logs in the Logs Explorer.