tencent cloud

Tencent Real-Time Communication

ドキュメントTencent Real-Time Communicationライブ配信Live ManagementLive Streaming Management SystemUniversal Comment Review

Universal Comment Review

Download
フォーカスモード
フォントサイズ
最終更新日: 2026-07-29 16:30:09

Overview

LiveKit Manager supports two text moderation modes:
Mode
Description
Configuration Value
Cloud Moderation
Uses the built-in cloud moderation engine of Tencent Cloud IM to automatically block inappropriate messages.
cloud
Custom Moderation
Suitable for real-time interactive moderation in live rooms. Uses a pre-send interception mechanism to submit user messages to a online manual review workflow, allowing only approved messages to go to appear in the live room.
custom

Architecture



Quick Start

Step 1: Quick Validation with Minimal Prototype (Verification Only, Not for Production)

cd packages/custom-moderation-server

# # Install dependencies
npm install

# # Copy configuration file
cp config/example.env config/.env

# # Start (port 9001)
npm start
Warning:
This is a minimal prototype, provided in the repository, designed to let you verify"IM Callback → Intercept → Moderation List → Approve"the full chain within minutes.
When going live, please implement and deploy your own moderation server based on the Custom Moderation Server API Specification and configure its address in CUSTOM_MODERATION_BASE_URL.

Step 2: Configure Main Server for Custom Moderation

Edit packages/server/config/.env:
# # Switch to custom moderation mode
MODERATION_MODE=custom

# # Custom moderation server address (demo server defaults to http://localhost:9001)
CUSTOM_MODERATION_BASE_URL=http://localhost:9001

# # API Key (optional, demo server has none configured by default, leave empty)
# CUSTOM_MODERATION_API_KEY=your_api_key

Step 3: Configure Callbacks in the TRTC Console

1. Go to TRTC Callback Configuration.
2. Under Social Interaction Callback, click Edit next to Social Interaction Callback URL, enter the callback URL (e.g., http://your-domain:9001/im-callback), and click OK.
3. Under Social Interaction Callback, find Callback Before Sending Messages in Group, enable the toggle, and click Save.

Step 4: Test

# # Simulate IM callback (message intercepted)
curl -X POST http://localhost:9001/im-callback \\
  -H "Content-Type: application/json" \\
  -d '{
    "CallbackCommand": "Group.CallbackBeforeSendMsg",
    "GroupId": "test_live_room",
    "From_Account": "user_001",
    "MsgBody": [{"MsgType": "TIMTextElem", "MsgContent": {"Text": "Hello"}}]
  }'

# # View moderation list
curl -X POST http://localhost:9001/moderation/list \\
  -H "Content-Type: application/json" \\
  -d '{"Receiver": "test_live_room", "PageNo": 1, "PageSize": 10}'

Custom Moderation Server API Specification

All endpoints are prefixed with {CUSTOM_MODERATION_BASE_URL}, configured via env.

Authentication (Optional)

If CUSTOM_MODERATION_API_KEY is configured, LiveKit Manager includes the following in all requests:
X-Api-Key: {CUSTOM_MODERATION_API_KEY}

Endpoint 1: Full Moderation Toggle

Query Toggle Status

GET /moderation/toggle
Request parameters: None
Response:
{
  "ActionStatus": "OK",
  "ErrorCode": 0,
  "Enabled": true
}

Set Toggle Status

POST /moderation/toggle
Request body:
{
  "Enabled": true
}
Response:
{
  "ActionStatus": "OK",
  "ErrorCode": 0,
  "Enabled": true
}
Field Description：
Enabled: true indicates full moderation enabled (all messages intercepted),false indicates disabled (messages delivered normally).

Endpoint 2: Moderation Message List

POST /moderation/list
Request body:
{
  "Receiver": "live_room_id",
  "PageNo": 1,
  "PageSize": 20
}
Field Description：
Field
Type
Required
Description
Receiver
string
Yes
Live room ID (corresponds to GroupId).
PageNo
number
Yes
Page number (starting from 1).
PageSize
number
Yes
Page size (max 100).
Response (Aligned with Tencent Cloud DescribeCloudAuditRecordDetailV2.Data array format):
{
  "TotalCount": 100,
  "RequestId": "req_1234567890",
  "Data": [
    {
      "ContentId": "msg_001",
      "From_Account": "user_alice",
      "Content": "Intercepted message content",
      "Time": "2026-06-30 14:30:00.000"
    }
  ]
}
Field Description：
Field
Type
Required
Description
TotalCount
number
Yes
Total record count (for pagination calculation).
RequestId
string
No
Request ID.
Data[].ContentId
string
Yes
Unique message ID(for subsequent Delete/Approve).
Data[].From_Account
string
Yes
Sender IM account.
Data[].Content
string
Yes
Message text content.
Data[].Time
string
Yes
Message timestamp (format YYYY-MM-DD HH:mm:ss.SSS).
Note:
In custom moderation mode, the Label (moderation label) field is not needed. The corresponding column will be hidden in the frontend.

Endpoint 3: Delete Moderation Records

POST /moderation/delete
Request body:
{
  "ContentIds": ["msg_001", "msg_002"]
}
Response:
{
  "ActionStatus": "OK",
  "ErrorCode": 0,
  "DeletedCount": 2,
  "RequestId": "req_1234567890"
}
Field Description：
Field
Type
Description
DeletedCount
number
Number of records actually deleted.

Error Response Format

All endpoints return on error:
{
  "ActionStatus": "FAIL",
  "ErrorCode": 500,
  "ErrorInfo": "Error description"
}

IM Callback Format

Group.CallbackBeforeSendMsg Callback

Tencent Cloud IM callback before group message delivery; your server must handle this callback.

Request (Tencent Cloud → Your Server)

{
  "CallbackCommand": "Group.CallbackBeforeSendMsg",
  "GroupId": "live_room_123",
  "Type": "Public",
  "From_Account": "user_456",
  "Operator_Account": "",
  "Random": 123456,
  "MsgBody": [
    {
      "MsgType": "TIMTextElem",
      "MsgContent": {
        "Text": "Message content sent by user"
      }
    }
  ]
}
Key fields:
Field
Description
GroupId
Live room ID.
From_Account
Sender IM account.
MsgBody[].MsgContent.Text
Text message content.

Response (Your Server → Tencent Cloud)

Intercept Message (not delivered to the live room):
{
  "ActionStatus": "OK",
  "ErrorCode": 1,
  "ErrorInfo": "Message intercepted for moderation"
}
Approve Message (delivered normally):
{
  "ActionStatus": "OK",
  "ErrorCode": 0,
  "ErrorInfo": ""
}
Note:
ErrorCode of 0 indicates Approve，non- 0 indicates interception. Callback timeout defaults to Approve(ensuring message deliverability).

Callback Timeout Fallback

Tencent Cloud IM callback timeout is 2 seconds. If your server does not respond within 2 seconds, IM will automatically approve the message. automatically approve the message.
Recommendations:
Your server's callback handler should perform fast storage operations (e.g., SQLite INSERT), keeping processing time below milliseconds.
Do not perform time-consuming external API calls in the callback handler.

Moderation Message Lifecycle

Approve Flow (handled by LiveKit Manager)

1. Admin clicks Approve.
2. LiveKit Manager calls send_group_msg (with NoMsgCheck to skip re-moderation) to resend the message as the original user.
3. After successful resend, call your server's POST /moderation/delete delete corresponding record.
4. Each message is processed independently: each succeeded message is deleted, failed ones remain in the database.

Delete Flow

1. Admin clicks Delete.
2. LiveKit Manager call your server's POST /moderation/delete delete corresponding record.

Dual-track Moderation Mode Comparison

Feature
Cloud Mode
Custom Mode
Moderation Data Source
Tencent Cloud DescribeCloudAuditRecordDetailV2
Custom Database
Moderation List Fields
id / User ID / Content / Recognition Type / Time
id / User ID / Content / Time
Action Buttons
Approve / Delete / More（Correction Whitelist）
Approve / Delete
Full Moderation Toggle
None
Yes
Delete Implementation
IndexedDB Frontend Marker
Custom Server DELETE
Approve Implementation
send_group_msg + IndexedDB Marker
send_group_msg + Custom Server DELETE
Correction Whitelist
Supported
Not Supported

FAQ

What happens to the live room if my server goes down?

Chat callback timeout (2 seconds), Tencent Cloud IM will automatically approve the message.Messages are not lost. However, new messages will not appear in the moderation list, until your server recovers.

Why doesn't the moderation list have a 'Recognition Type' column?

In Custom moderation mode, moderation logic is controlled by you; there are no 「Porn/Abuse/Ad」predefined tags like "Porn/Abuse/Ad". Moderators see all intercepted messages, decide whether to approve each message.

After approving a message, why are other previously intercepted 「messages from the same user 」still in the list？

Each message is processed independently. Batch approve can select multiple messages to process together.

Can I switch back to Tencent Cloud moderation?

Yes. Change MODERATION_MODE back to cloud and restart the service. Switching will not lose data — your server data remains intact.

How do I implement custom moderation rules (e.g., sensitive word filtering)?

Implement in your server's IM callback handler. Example:
// callback.js
const db = require('./db');  // // Database module, provides insertMessage etc. APIs
const blockedWords = ['sensitive_word_1', 'sensitive_word_2'];
//   // Extract text content from MsgBody (assuming only the first text message is processed)
const content = req.body.MsgBody[0]?.MsgContent?.Text || '';

if (content && blockedWords.some(w => content.includes(w))) {
  //   // Intercept and store message in database
  db.insertMessage({
    content_id: `msg_${Date.now()}`,
    group_id: req.body.GroupId,
    from_account: req.body.From_Account,
    content: content,
    msg_body: JSON.stringify(req.body.MsgBody)
  });
  res.json({ ActionStatus: 'OK', ErrorCode: 1 });
} else {
  // Approve
  res.json({ ActionStatus: 'OK', ErrorCode: 0 });
}

前の記事へ: SDK Customization (React)次の記事へ: Anchor Core View (Android)

ヘルプとサポート

この記事はお役に立ちましたか？

フィードバック