Skip to main content
In this tutorial you’ll build a fully working AI agent from scratch using the Splox workflow builder. No templates, no shortcuts — just three nodes wired together.

​
What You’ll Build

Completed AI agent workflow with Start, Agent, and Tool nodes connected By the end of this tutorial, you’ll have an autonomous agent that:
  • Receives a user message via the Start node
  • Thinks and decides what to do via the Agent node
  • Takes real actions using a connected Tool node
Time to complete: 5 minutes
Skill level: Beginner

​
Prerequisites

Before you begin, make sure you have:

​
Step 1: Create a New Workflow

  1. Go to app.splox.io and open the Workflows section
  2. Click the + button on the left panel
  3. Give it a name — e.g. My First Agent
  4. The workflow canvas opens with an empty graph
Creating a new workflow in Splox

​
Step 2: Add a Start Node

The Start node is the entry point of your workflow.
  1. Right-click on the canvas → Start
Adding a Start node to the workflow canvas
Every workflow needs exactly one Start node. It has no configuration — it simply marks where execution begins.

​
Step 3: Add an Agent Node

The Agent node is an LLM that reads the input, reasons about it, and decides whether to call a tool or respond directly.
  1. Right-click on the canvas → Agent
  2. In the config panel:
    • Model — pick any model (e.g. claude-sonnet-4-5 or gpt-4o)
    • System Prompt — write a short instruction, e.g.:
You are a helpful assistant. Use available tools when needed to answer the user's question.
Configuring the Agent node with a model and system prompt

​
Step 4: Add a Tool Node

The Tool node gives your agent access to the Tool Router — a smart layer that automatically finds the right tool from the MCP marketplace based on what the agent needs to do.
  1. Right-click on the canvas → Tool
That’s it. The Tool Router handles the rest: when your agent runs and needs a capability (e.g. web search, sending a message), it will find the matching tool from your connected MCP integrations. If you don’t have the required connection yet, the agent will surface a link to connect it from the MCP marketplace. Tool node on the canvas

​
Step 5: Connect Everything

Now wire the nodes together:
  1. Start → Agent: drag from the Start node’s output handle to the Agent node’s input handle
  2. Agent → Tool: drag from the Agent node’s TOOLS handle to the Tool node’s input handle
Make sure you connect from the TOOLS handle (not the PARALLEL handle) on the Agent node. This enables the tool-calling loop so results return back to the Agent.
All three nodes connected — Start to Agent to Tool

​
Step 6: Run It

  1. Open the Chat panel (bottom toolbar → chat icon)
  2. Type a message, e.g. Search the web for the latest AI news
  3. Hit Enter — your agent will call the tool, get the result, and respond
Agent responding in chat after calling a tool

​
Advanced: Configuring Your Agent

Click on any node to open its configuration panel. The Agent node has three tabs: Model tab
SettingDescription
LabelRename the node
Provider / ModelSwitch between LLM providers and models (e.g. Anthropic, OpenAI, Gemini)
System PromptInstructions that define your agent’s behavior and persona
SkillsAttach reusable skill sets to your agent
CredentialsUse your own API keys instead of platform credits
Enable Realtime (Voice)Switch to voice/realtime mode
Memory tab — configure context window limits, trim strategy, summarization, and more. Execution tab — set max iterations, tool choice, streaming, and other runtime settings. Agent node configuration panel

​
Frequently Asked Questions

When an LLM calls a tool, it needs to receive the result back to continue reasoning. The TOOLS edge creates this loop — the Agent calls the Tool, gets the result, and then decides what to do next (call another tool or respond to the user).
Yes. Add more Tool nodes and connect them all to the Agent’s TOOLS handle. The agent will automatically know about all connected tools and choose the right one based on context.
The TOOLS handle is used for tool-calling loops — results return to the Agent. The PARALLEL handle is for regular sequential flow where the Agent’s output passes to the next node without looping back.

​
Next Steps

  • Memory — the Agent node is always multi-turn by default. Open the node config to tune memory settings: limit type (messages or tokens), max messages/tokens, trim strategy, summarization, and more.
  • Add more tools — connect multiple Tool nodes to give your agent more capabilities
  • Go parallel — check out Parallel Agents to run multiple agents simultaneously
  • Build multi-agent systems — explore Multi-Agent Teams
  • Join the community — share your builds at community.splox.io or Telegram