Chat Response Commands

This document describes the server-to-client message events for chat responses in the WebSocket Gateway.

Message Event Types

message

The primary event for streaming chat responses. Server sends incremental content as the AI generates the response.

Data Fields:

Example:

{
  "event": "message",
  "data": {
    "output": "Hello! I'd be happy to help",
    "stop": false,
    "ok": true
  }
}

When stop is true, the stream has ended:

{
  "event": "message",
  "data": {
    "output": "",
    "stop": true,
    "ok": true
  }
}

conv-message

Sent when a conversation is created or updated, providing conversation identification information.

Data Fields:

Example:

{
  "event": "conv-message",
  "data": {
    "conversationId": "conv-123456",
    "messageId": "msg-789012"
  }
}

error

Sent when an error occurs during processing.

Data Fields:

Example:

{
  "event": "error",
  "data": {
    "error": "Model service temporarily unavailable"
  }
}

auth-connect-complete

Sent when the WebSocket authentication handshake is successfully completed.

Example:

{
  "event": "auth-connect-complete",
  "data": {
    "ok": true
  }
}

set-model-complete

Sent when the model selection/setting operation is completed.

Data Fields:

Example:

{
  "event": "set-model-complete",
  "data": {
    "model": "gpt-4-turbo",
    "modelShortName": "GPT-4",
    "ok": true
  }
}

pong

Heartbeat response from server to confirm connection is alive.

Example:

{
  "event": "pong",
  "data": {}
}

Message Flow Sequence

A typical chat interaction follows this message sequence:

1. Connection established → auth-connect-complete
2. Model set (if needed) → set-model-complete
3. Send chat request (Client → Server with event: "gen")
4. Multiple message events with incremental output content
5. Final message event with stop: true
6. Conversation info → conv-message (if new conversation)

Content Filtering

The client should filter special markers from the output content:

• <｜end▁of▁sentence｜> - End of sentence marker
• <｜begin▁of▁sentence｜> - Begin of sentence marker
• ### - Section marker
• Leading question marks (? or ？) on first line

Error Handling

When receiving an error event:

1. Display the error message to the user
2. Consider disconnecting and reconnecting if the error is connection-related
3. Log the error for debugging purposes