For the complete documentation index, see llms.txt. This page is also available as Markdown.

Channels

Channels connect an Agent to external messaging platforms. Currently Feishu, Telegram, QQ, WeChat, Discord, and Slack are configurable; the bot credentials and session identifiers required by each platform differ.

频道设置中的飞书、Telegram、QQ、微信、Discord 和 Slack 平台入口
First select the platform to connect, then click [Add]; different platforms require different accounts and credentials to be prepared.

Manual configuration path

[Settings] → [Channels] → select platform → [Add].

1

1. Prepare a platform account

Create a bot or app according to the platform rules. If Feishu and WeChat support QR-code-related flows, you can complete them by following the on-screen prompts after enabling the channel; for other platforms, fill in the corresponding Token or app credentials.

2
飞书频道配置中的 Agent、应用凭据、允许聊天范围和权限模式
First bind a tested Agent and restrict the allowed chat scope; only expose the parts of the credentials and permissions needed to complete the task.

2. Bind the Agent and workspace

Select an Agent that has already been verified to work, and specify a workspace for channel messages. External messages will run in this context; do not choose a directory that contains unrelated sensitive files.

3

3. Restrict message sources

Fill in the allowed Chat ID, channel ID, or user ID. Leaving it blank may mean allowing all; refer to the current platform field description for details. You can first send the bot /whoami to obtain the correct identifier format.

4

4. Choose permission mode and enable

By default, [Inherit Agent Settings] is used. Public groups, multi-user groups, or untrusted sources should use a stricter mode. After enabling, first send a test message with no side effects.

Configuration item
Product default
Suggested starting point
Effect
Applicable scenario
Notes

Bind Agent

Selection required

Prepare a separate Agent for the channel

Decides who handles the messages

Group chats, bot direct messages

Do not mix with high-privilege development Agents

Workspace

Selection required

Use a dedicated directory without private data

Limit the files that can be processed

Collect in the group, deliver daily reports

Do not select the user's home directory

Allowed message sources

Depends on the platform fields

First allow only test accounts or test groups

Restrict who can trigger the Agent

Internal testing, team groups

Read the field description before leaving it blank

Permission Modes

[Inherit Agent Settings]

Use stricter approval for external entry points

Control tool operations

Public groups, multi-user groups

Not recommended to use [Full Access]

Use case: team on-call group

Create an Agent that only handles the on-call handbook, bind a knowledge base that has passed recall testing and a dedicated workspace, and allow only the on-call group to trigger it. First test the two kinds of messages “check a policy” and “couldn't find the answer”, then test generating a handoff summary that does not contain sensitive information.

Completion criteria

Non-allowed sources cannot trigger tasks; replies can be sent back to the specified group; when the knowledge base has no content, it should state that clearly; file writes will still request approval.

Platform configuration focus

Platform
Primary credentials
Session scope focus

Feishu / Lark

App ID, App Secret, or QR-code flow

Chat ID and domestic/international domain names

Telegram

Bot Token

Chat ID

QQ

App ID, Client Secret

Private chat, group, or channel identifier format

WeChat

QR-code login or credential path

Allowed user IDs

Discord

Bot Token

Channel or private chat ID

Slack

Bot Token, App Token

Socket Mode and channel ID

What should I do if I can't select a channel as the destination for a scheduled task?

First send a message to the bot proactively on the corresponding platform so Cherry Studio records the available Chat ID, then go back to the scheduled task and choose the notification target.

What should I do if the channel is connected but doesn't reply?

Check the allowed session ID, bound Agent, workspace, and permission mode, then check the channel logs. If the platform received the message but the Agent did not execute, continue by checking the API gateway and the Agent task status.

Last updated

Was this helpful?