Creating Agents

Create the Agent

Start by defining your agent's identity.

Basic Information

1. Go to Agents in the sidebar
2. Click Create Agent
3. Fill in the details:
   • Name - A clear, descriptive name
   • Type - Chat, Summarization, Recommendation, Analyzer, or Other
   • Description - What this agent does and when to use it

Choosing the Right Type

Select a type that matches your use case. This helps organize your agents and provides relevant defaults.

Configure Entry Point

Define what inputs your agent accepts.

Parameter Setup

Each parameter needs:

• Name - How you'll reference it (e.g., user_message)
• Type - string, number, boolean, object, or array
• Required - Whether the agent fails without this input
• Default - Optional fallback value

Best Practices

Keep parameters focused. A well-designed agent has clear, minimal inputs that are easy to understand and use.

Add Agent Steps

This is where the AI magic happens.

Creating an Agent Step

Click + to add an Agent Step and configure:

• Model - Select your LLM (GPT-4, Claude, Gemini)
• Prompt - Instructions for the model
• Temperature - Control creativity vs consistency

Using Placeholders

Reference dynamic values in your prompts:

User's question: {{input.user_message}}

Previous analysis: {{step_1.output}}

Add Integrations (Optional)

Extend your agent with external capabilities.

API Tools

Attach your configured integrations as tools on any Agent Step — any REST API you've imported becomes callable by the LLM.

Sub-Agents

Invoke another Fruxon agent as a step. Great for breaking complex logic into reusable pieces.

Configure Exit Point

Define what your agent returns.

Mapping Outputs

Connect values from your workflow to the exit point. You can return:

• Direct LLM output
• Processed/transformed data
• Structured JSON objects

Output Format

Design outputs that are easy for consumers to use. Consider who will call this agent.

Test

Validate your agent before deployment.

Interactive Testing

1. Click Test to open the test panel
2. Enter sample inputs
3. Run the agent

Review Results

Examine the execution trace to see:

• What each step produced
• Token usage and timing
• Any errors or unexpected behavior

Deploy

Make your agent available for use.

Creating a Deployment

1. Click Save to create a revision
2. Review your changes
3. Click Deploy to publish

What Happens Next

Your agent is now available via:

• REST API with your API key
• Platform connectors (Slack, Teams, etc.)
• Other agents as a sub-agent