Skip to main content
Agents

A separate AI agent for each part of your business

Each question is classified and sent to the matching agent. Every agent has its own documents, instructions and the conversation history in view.

Example agents

Example agents. You write the instructions and upload the documents for your own.

Sales

Jordan, sales agent

Plans and pricing

Answers pricing questions from your plan documents, compares plans for the customer, and hands buyers who want to talk to a person over to your sales team.

Specialties

Plan comparisonPricing answers from your docsAnnual billing questionsHandoff to your sales team

Reads from

Your plans and pricing pageBilling termsSales FAQ

Routes on

pricepricingcostbuypurchaseupgradequoteplan
Product

Taylor, product agent

Features and comparisons

Explains what your product does, suggests features that fit the customer’s question, and points to the case studies you have published.

Specialties

Feature questionsAdd-on suggestionsProduct comparisons

Reads from

Your feature listYour published case studiesComparison pages

Routes on

comparefeatureintegrationscalecase studyversus
Custom

Your own agent

Your instructions, your documents

Set up an agent for your own workflow: write its instructions, choose the keywords that route to it, add guardrails, and give it its own documents.

Specialties

Custom instructionsRouting keywordsIts own documents

Routes on

custommy agentcreate

How a question is routed

5 Connected Primitives
Question
Customer question arrives

From the website widget (WebSocket) or WhatsApp (Meta Cloud API)

Routing
Intent classifier

Decides if it is about a refund, technical support, billing or sales

Agent
Matching agent

The support, sales or product agent takes the conversation

Your documents
Document search

Finds matching passages and checks the confidence score

Customer reply
Streamed reply

The answer streams to the customer with its sources

Sample data

Suggested routing rule

Sample: return questions keep coming in, so order tracking and refund tools are linked to the support agent.

90%confidence
typescriptSuggested Configuration
// Sample routing rule
router.registerIntent({
  intent: "order_inquiry",
  targetAgent: "support_specialist",
  tools: ["order_tracking", "refund_calculator"],
  groundingScoreMin: 0.85
});
Retrieved Context Chunks (2)
Routing

How a question reaches the right agent

Each question is classified and sent to the matching agent. Anything the agents can't answer goes to your team.

Agent routing diagram

LANGGRAPH ROUTING DAG ENGINE

Topological State Transition & Intent Scoring

Incoming Conversation Stream
CHANNEL: WEB_WIDGETLATENCY: 1.1ms

"Where is my order #8492 and can I initiate a refund?"

Cos Similarity0.984 (Order/Refund Intent Match)
Resolved Execution Agent
Support Specialist
Active Memory Bank Loaded

State committed to LangGraph route node. Agent responds with strictly grounded vector documentation.

  1. 01

    Intent and topic match

    Each question is checked against your department classifiers: returns and orders go to support, pricing questions go to sales.

  2. 02

    Quick path for greetings

    Greetings and simple questions get a short reply without a full document search.

  3. 03

    Matching agent takes over

    The conversation goes to the agent with the right instructions and documents for that topic.

  4. 04

    Human fallback

    If the agent is not confident in an answer, or the question needs a person, the chat moves to your agent inbox.

Set up your first agent

Paste your website link or upload your docs, then copy one embed snippet.