Chat API

Use the Chat API to integrate PolyAI agents into non-voice channels. It provides a simple set of REST endpoints for starting conversations, sending and receiving messages, and handling handoffs–ideal for webchat widgets, web SDKs, and SMS platforms.

The Chat API powers webchat integrations including in-browser widgets, web SDK implementations, and SMS.

Chatting with the API

1. Start a conversation Call POST /{version}/{account_id}/{project_id}/chat/create. The response includes a conversation_id and the agent’s initial response. You can optionally pass integration_attributes to provide custom context to the agent, and variant_id to target a specific variant.

2. Send and receive messages Call POST /{version}/{account_id}/{project_id}/chat/respond with the conversation_id. message is optional. The response includes the agent’s response, an end_conversation flag, and may include a handoff object.

3. Close a conversation Call PUT /{version}/{account_id}/{project_id}/chat/close with the conversation_id in the body. The response returns { "success": true } on success.

Endpoints

Base URLs

Endpoint format: /{version}/{account_id}/{project_id}/chat/{operation}

• version: API version (for example v1)
• account_id: Your PolyAI account ID (for example poly-scs-uk or ws-xxxxxxxx)
• project_id: Your PolyAI project ID (for example PROJECT-191bfa2a)
• operation: create, respond, or close

Finding your account_id and project_id

Your account_id and project_id are the first two path segments of your Agent Studio URL, immediately after the host:

https://studio.<region>.poly.ai/<account_id>/<project_id>/...

For example, if your Studio URL is https://studio.uk.poly.ai/acme-uk/acme-team-4/agent, then:

• <region> = uk (so calls go to api.uk-1.platform.polyai.app)
• account_id = acme-uk
• project_id = acme-team-4

Account and project identifiers are also shown in prefixed form in Agent Studio — both the slug form (visible in the URL) and the prefixed form are accepted in API paths:

The same IDs are exposed inside Agent Studio functions as conv.account_id and conv.project_id — see the Conversation object reference.

Authentication

All requests must include the following headers (case-sensitive):

Getting your credentials

Both X-API-KEY and X-TOKEN are provisioned by PolyAI — there is no self-serve endpoint to create a Chat API connector or generate a connector token.

To request access, contact your PolyAI representative with:

• Your account_id
• Your project_id
• The client environment you need (live, pre-release, or sandbox)
• Your variant_id (optional, only if targeting a specific variant)

PolyAI will provision a Chat API connector for your project and return:

• An API key to use in the X-API-KEY header
• A connector token to use in the X-TOKEN header

Example: Create chat

POST /v1/{account_id}/{project_id}/chat/create

curl -X POST \
  "https://api.<region>.platform.polyai.app/v1/ws-xxxxxxxx/PROJECT-xxx/chat/create" \
  -H "X-API-KEY: YOUR_API_KEY" \
  -H "X-TOKEN: YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "variant_id": "VARIANT-xxxxxxxx",
    "integration_attributes": {
      "user_id": "12345",
      "customer_tier": "premium"
    }
  }'

Response

{
  "conversation_id": "CONV-1234567890",
  "response": "Hi, how can I help you today?"
}

Example: Send message

POST /v1/{account_id}/{project_id}/chat/respond

curl -X POST \
  "https://api.<region>.platform.polyai.app/v1/ws-xxxxxxxx/PROJECT-xxx/chat/respond" \
  -H "X-API-KEY: YOUR_API_KEY" \
  -H "X-TOKEN: YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "conversation_id": "CONV-1234567890",
    "message": "What'\''s your return policy?"
  }'

Response

{
  "conversation_id": "CONV-1234567890",
  "response": "Our return policy is 30 days with proof of purchase.",
  "end_conversation": false
}

Response with handoff

{
  "conversation_id": "CONV-1234567890",
  "response": "Transferring you to a live agent.",
  "end_conversation": true,
  "handoff": {
    "destination": "live_agent_queue",
    "reason": "billing_question"
  }
}

Example: Close chat

PUT /v1/{account_id}/{project_id}/chat/close

curl -X PUT \
  "https://api.<region>.platform.polyai.app/v1/ws-xxxxxxxx/PROJECT-xxx/chat/close" \
  -H "X-API-KEY: YOUR_API_KEY" \
  -H "X-TOKEN: YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"conversation_id": "CONV-1234567890"}'

Response

{
  "success": true
}

Passing custom data with integration_attributes

The integration_attributes field lets you pass custom data when creating a chat. These attributes are accessible in your project functions at conv.integration_attributes.

When to use

• Pass user context (user ID, session ID, authentication status)
• Include customer information (tier, preferences, history)
• Send external system references (CRM ID, ticket number)
• Provide A/B test parameters or feature flags

Accessing in project functions

def start(conv):
    # Get integration attributes (always check for None)
    attrs = conv.integration_attributes or {}
    
    user_id = attrs.get("user_id")
    customer_tier = attrs.get("customer_tier", "standard")
    
    # Store in state for use throughout the conversation
    if user_id:
        conv.state.user_id = user_id
    
    # Customize greeting based on tier
    if customer_tier == "premium":
        return "Welcome back! As a premium member, how can I assist you today?"
    return "Hello! How can I help you?"

Targeting a variant with variant_id

Pass variant_id in the create request body to route the conversation to a specific variant. This is useful for multi-site deployments where different variants serve different brands, languages, or regions.

{
  "variant_id": "VARIANT-xxxxxxxx"
}

Notes

• create accepts an optional request body with variant_id and integration_attributes.
• respond requires conversation_id; message is optional.
• handoff may appear in the respond response with destination and reason.
• close requires a JSON body containing conversation_id.
• Header names are case-sensitive: X-API-KEY, X-TOKEN.
• Use the regional base URL closest to your deployment.