What Is a Node?

A Node is the fundamental building block of any Langoedge graph. Each node represents a single action — an AI thought, an API call, a data transformation, or a routing decision.

Definition — Node: A discrete step in an AI agent workflow. Nodes can be LLM calls, tool executions, custom code, or routing conditions that determine the next step.

Rather than thinking of your chatbot as one giant brain trying to do everything at once, Langoedge encourages you to break complex workflows into distinct, specialized nodes.

Why Separate Nodes?

- **Focus:** LLMs perform better with one clear directive per step. - **Predictability:** Securely route users down different paths based on intent. - **Debuggability:** See exactly which step succeeds or fails in the debugger.

State & Memory Flow

While nodes focus on specific tasks, they share information via the State — a central repository that travels from node to node as a conversation progresses.

Initialization

When a chat begins, an empty State object is created for that specific Thread ID. The state is stored in our persistent database and persists across days.

Reading

Each node reads past messages, extracted variables, and tool results currently stored in the State.

Writing

Before finishing, each node appends its findings (e.g., an extracted order number or a drafted response) back to the State.

Memory Preservation

Because nodes only append to the State rather than overwriting it, every subsequent node always has full context of everything that occurred before.

Node Types

Langoedge provides four primary node types, each designed for a specific job.

1. Method Nodes (The Brain)

Method Nodes are where actual AI processing happens.

Purpose: They take the conversation history, read your custom instructions, and generate a text response.
When to use: Greetings, intent classification, summarization, email drafting, tone analysis.
Tool Binding: You can give a Method Node permission to use specific tools. If the AI decides it needs one, it routes the graph to a Tool Node.

2. Tool Nodes (The Hands)

Tool Nodes act as the bridge between the AI and the outside world.

Purpose: They execute pre-configured data tasks — hitting a REST API, querying a Vector Database, or sending an email.
Key distinction: Tool Nodes do not use AI. They fire the task and return raw data to the State.
When to use: CRM updates, database queries, email sending, file uploads, webhook triggers.

3. Custom Python Nodes

Run raw backend code in a secure, sandboxed environment.

Purpose: Use when you need 100% deterministic results — math, date formatting, complex data transformation, or string manipulation.
Security: Code is evaluated via our Python parser and validator to prevent malicious code execution.
When to use: Converting AI-generated messy text into strict JSON, date math (e.g., "tomorrow" to ISO-8601), array filtering, checksum computation.

4. Conditional Edges (Routing Nodes)

While technically "Edges" rather than standalone processing nodes, conditional edges act as traffic controllers.

Purpose: They look at the current State and decide where the graph should go next.
When to use: "If sentiment is angry, route to Escalation. If happy, route to Checkout."
Quality Gates: A special type of conditional edge that spins up a Supervisor AI to evaluate the previous node's output against rules.

Design Patterns

Pattern 1: The Secure Refund Bot

Automate refunds by connecting an internal API without exposing the "Refund" capability directly to the LLM.

graph TD User([User requests refund]) --> NodeA(Node: Evaluator) NodeA -->|Valid for refund| NodeB(Node: Refund Executor) NodeB -->|Secure API hit| API_2{{"Tool: Process Refund"}} NodeB --> NodeC(Node: Conclusion) NodeA -->|Not valid| NodeC NodeC --> End([Polite formatting])

Best Practice: Always isolate sensitive tools (like processing payments) behind an evaluation node to ensure the AI has verified all prerequisites.

Pattern 2: Legal Compliance Quality Gate

Ensure a support AI never writes an email making false legal or financial promises.

graph TD User([Angry email]) --> NodeA(Node: Draft Reply) NodeA -->|Sends draft to supervisor| Gate{"Compliance Check"} Gate -->|Approved| NodeB(Node: Send Email) Gate -->|Failed| NodeA NodeB --> End([Email Sent])

Pattern 3: Data Extraction Pipeline

Parse messy user input into clean, structured data using AI + Python collaboration.

graph TD User([Messy input]) --> NodeA(Node: AI Parser) NodeA -->|Raw JSON string| NodeB(Node: Python Validator) NodeB -->|Clean structured data| End([Output])

How to Choose the Right Node Type

Task	Recommended Node
Greeting / conversational response	Method Node
Classify user intent	Method Node
Send email or Slack message	Tool Node
Query a database	Tool Node
Format dates or run math	Custom Python Node
Convert text to JSON	Custom Python Node
Check if a response meets rules	Conditional Edge (Quality Gate)
Route based on sentiment	Conditional Edge (Tool Condition)

Debugging Nodes

Use the Langoedge Dashboard's Debug Mode to:

See exactly which nodes were triggered.
View the State at every step.
Check how long each node took to execute.
Inspect which tools the AI decided to call.

This transparency makes it easy to fine-tune agent behavior and identify bottlenecks.

Frequently Asked Questions

What is the difference between a Method Node and a Tool Node?

A Method Node uses an LLM to think and generate responses. A Tool Node executes a predefined action (API call, database query) without AI involvement. Method nodes are the "brain," Tool nodes are the "hands."

Can I chain multiple Tool Nodes?

Yes. You can chain Tool Nodes together — for example, first check order status, then process a refund, then send a confirmation email. Each tool's output can feed into the next node via the State.

Is Custom Python Node secure?

Yes. Python code is evaluated using our Python parser in an isolated execution namespace. Malicious operations like file access, network calls, or command execution are blocked.

What is a Quality Gate?

A Quality Gate is a conditional edge that uses a second AI to review the output of the first AI. It returns True or False based on your rules — perfect for compliance, brand voice, and tone checks.

Back to Text Graph

Review the full Text Graph documentation.

Node Deep Dive — Master AI Agent Building Blocks