Platform Manual — Complete Langoedge Guide for AI Agent Builders

Introduction: What Is Langoedge?

Langoedge is an AI agent platform — not just a chatbot builder. It is a workshop for creating AI Digital Employees that understand intent, make decisions, and take actions.

Most chatbots are rigid — they follow a strict script. Langoedge agents are probabilistic. They use Large Language Models (LLMs) to understand intent, evaluate logic, and use tools, just like a human would.

What Can You Build?

Core Concepts

Before building your first agent, understand the fundamental building blocks.

The Graph (Workflow)

The Graph is the blueprint of your agent's mind — a flowchart where each step represents an action or a thought.

The State (Memory)

The State is a shared "notepad" that the agent carries throughout the conversation.

• Everything the user says goes into the State.
• Everything the agent learns (e.g., user's name, order ID) is written into the State.
• All tools use data from the State and write results back to it.
• The State persists in the storage engine, surviving across days or weeks.

The Node (Action)

A single step in the process — could be the LLM thinking, an API call, or a data transformation.

The Edge (Path)

A connection between nodes that tells the agent where to go next. Edges can be direct (always proceed) or conditional (go based on rules).

The Graph Builder

Navigate to Graphs > New Graph to begin constructing your Text Agent.

Anatomy of a Node

When you click a Node, you'll see several configuration options:

• Name: An internal name like evaluate_intent or check_order_status.
• System Prompt: The personality and logic rules for this step (e.g., "You are a helpful shipping assistant. Only answer questions about delivery dates.")
• State Update: Where the output should be saved (field1 through field15).
  • Use final_response for text that should be spoken or displayed to the user as the agent's main reply. This special field is read by the graph executor to determine the content returned to the client at the end of a turn.
  • Use numbered fields (field1–field15) for silent data storage (e.g., extracted_email, order_id) that the user never sees directly.

Variables & Placeholders

Inject memory values into your prompts using Double Curly Braces {{ }}.

Example: If you've saved the user's name in a state field called user_name, write:

"Hi {{user_name}}, how can I help you today?"

When the agent runs, it replaces {{user_name}} with the actual value. This works for Tool Inputs too.

Logic & Decisions (Edges)

Connect nodes by dragging lines between them.

• Direct Edges: Simple sequential flow from one node to the next.
• Tool Conditions: Logic branches that trigger specific tools based on the AI's output.
• Quality Gates: A supervisor AI that checks the work of the previous node before allowing the flow to continue.

The Voice Builder

Voice Graphs allow you to build AI phone agents that respond in real-time via phone calls or web chat.

Navigate to Voice Graphs to get started.

Voice Synthesis (Text-to-Speech)

Choose from various high-quality voice profiles:

Telephony

• Buy a Number: Search and purchase phone numbers directly in the dashboard.
• SIP Trunks: For enterprise scale, connect your existing telephony stack via standard SIP trunks.
• Inbound Calls: Users call your number, and the AI answers instantly.
• Outbound Calls: The AI can dial users for reminders, follow-ups, or notifications.

Background Audio & Realism

Make your voice agent feel natural:

• Office Ambience: Subtle background chatter.
• Keyboard Typing: Simulates "looking things up" while thinking.

Global Settings

Click the Settings (Gear Icon) in the top right of any graph.

Model Options

Checkpoint TTL (Memory Persistence)

How long should the agent remember the conversation?

• Default: 7 days (10080 minutes)
• After this time, the state is cleared from the storage engine.
• Adjust this based on compliance requirements or conversation length needs.

Execution Mode

• Background Execution Enabled: The agent can pause and resume later (e.g., "I'll wait for your email").
• Background Execution Disabled: The agent process ends if the user closes the browser.

Protection Levels

Deployment

Web Widget

1. Open your Text Graph.
2. Click Embed / Deploy.
3. Copy the JavaScript snippet.
4. Paste it into your website's HTML — the chat widget appears instantly.

Phone Number

1. Open your Voice Graph.
2. Ensure a phone number is assigned in Settings.
3. Click the Headphones icon for a test call in your browser.
4. Publish to customers when ready.

External API

Generate API secrets in the Account tab and use the x-langoedge-secret header to trigger graphs from external CRMs or custom applications.

Troubleshooting

The agent is hallucinating or making things up

• Fix: Tighten your System Prompt. Add constraints like "Do not make up facts. If you don't know, say 'I don't know'."
• Fix: Switch to a smarter model or use a Reasoning model.

The tool isn't working

• Fix: Check the State. Is the input variable (e.g., {{email}}) actually populated? Use the Debugger.
• Fix: Re-authenticate the integration in the Connect tab. Tokens may have expired.

The voice agent responds too slowly

• Fix: Use a faster model for the greeting node.
• Fix: Enable streaming in your tool definitions.
• Fix: Check the VAD threshold in silence settings.

I don't know why the agent took a certain path

• Fix: Use Debug Mode. Click the "Debug" button to see a visual timeline of node execution, tool calls, and state changes at every step.

Frequently Asked Questions

[!WARNING]
Team Sharing is Currently Disabled. The frontend team-sharing feature is not yet available. Agents are private and scoped to individual accounts. You can still make agents accessible to others by setting the Protection Level to Protected or Public (see Protection Levels above).