Agent Block

The Agent block is the most powerful block in Zelaxy. It connects to any major LLM provider, supports tool calling, structured JSON outputs, conversation memory, and intelligent fallback systems. Use it whenever you need AI reasoning, text generation, classification, summarization, or any language task.

Overview

When to Use

• Generate text, summaries, or analyses from input data
• Classify, categorize, or extract information from content
• Call external tools (search, APIs, databases) as part of reasoning
• Build chatbots with persistent memory
• Get structured JSON responses for downstream processing

Configuration

System Prompt

Define the agent's behavior, personality, and reasoning framework. The built-in AI Wand can generate sophisticated system prompts for you — just describe what you want the agent to do.

User Message / Context

The input the agent processes. Typically references another block's output: {{starter.input}} or {{previous_block.content}}.

Model Selection

Searchable dropdown with all supported providers. Each model shows its provider icon and performance characteristics.

Advanced Settings

Response Format (Structured Output)

Define a JSON schema to get typed, predictable responses. The AI Wand can generate schemas from natural language descriptions.

Tool Integration

Connect tool blocks (Slack, Gmail, Search, etc.) to the Agent. The agent automatically discovers connected tools and calls them during reasoning.

Outputs

Example: Research Assistant

Goal: Build an agent that researches a topic and returns a structured summary.

Workflow:

[Starter] → [Agent] → [Response]

Configuration:

• System Prompt:

  You are a research assistant. Given a topic, provide a comprehensive summary
  with key facts, recent developments, and sources. Always be factual and cite
  your reasoning.

• User Message: {{starter.input}}
• Model: gpt-4o
• Temperature: 0.3 (factual, low creativity)
• Response Format:

  {
    "type": "object",
    "properties": {
      "summary": { "type": "string" },
      "keyFacts": { "type": "array", "items": { "type": "string" } },
      "confidence": { "type": "number" }
    }
  }

Result: The agent returns a structured JSON object that downstream blocks can parse reliably.

Example: Agent with Tools

Goal: Agent that answers questions using web search and then sends results via Slack.

Workflow:

[Starter] → [Agent] → [Slack]
              ↑
        [Google Search]  (connected as tool)

How it works:

1. Connect a Google Search block to the Agent (draw line to the tools input)
2. The Agent automatically discovers the search tool
3. When the user asks a question, the Agent decides whether to search the web
4. Search results feed back into the Agent's reasoning
5. Final answer goes to {{agent.content}} → Slack message

Tips

• Use structured output for pipelines — downstream blocks can reliably parse JSON fields
• Set low temperature (0.1–0.3) for factual/classification tasks
• Set high temperature (0.7–1.0) for creative writing
• Enable fallback model for production workflows — prevents failures if primary model is down
• Connect memory blocks for multi-turn chatbot experiences