Turn Feishu into your AI super-gateway. A production-grade Feishu/Lark channel plugin for OpenClaw — the brilliant AI agent framework.

• Human-like Message Processing — Bot reads all accumulated messages before responding, just like a human catching up on a conversation
• Intelligent Batching — Groups messages by chat, flushes on trigger (like @mention) with full context
• Mention Preservation — Non-bot @mentions are preserved as @[Name](open_id) so Agent can @ users back
• Extensible Triggers — @mention is just one trigger type; architecture supports keywords, schedules, etc.
• History Messages API — Fetch chat history with pagination for context gathering
• Flexible Access Control — DM policies (open/pairing/allowlist) and group policies (open/allowlist/disabled)
• Dual Domain Support — Works with both Feishu (China) and Lark (International)
• Interactive Card Tool — AI can send rich interactive cards with buttons, layouts, and formatted content
• Multi-Account Support — Configure multiple Feishu apps in a single OpenClaw instance
• Per-Sender Tool Policy — Fine-grained tool permissions based on message sender identity

# npm
openclaw plugins install @xzq-xu/feishu

# GitHub (for testing)
openclaw plugins install github:xzq-xu/openclaw-plugin-feishu

Edit ~/.openclaw/openclaw.json:

{
  "channels": {
    "feishu": {
      "enabled": true,
      "appId": "cli_xxx",
      "appSecret": "xxx",
      "domain": "feishu",
      "dmPolicy": "pairing",
      "groupPolicy": "open"
    }
  }
}

Or use environment variables (takes precedence if config values are empty):

export FEISHU_APP_ID="cli_xxx"
export FEISHU_APP_SECRET="xxx"

The plugin downloads images, files, and audio from Feishu messages. By default, files are saved to the system temp directory (e.g., /tmp/openclaw-feishu-media/). You can customize this:

{
  "channels": {
    "feishu": {
      "mediaDir": "~/.openclaw/media/feishu"
    }
  }
}

Supported media types:

• Images: PNG, JPEG, GIF, WebP
• Files: PDF, DOC, TXT, etc.
• Audio: Opus/Ogg (Feishu voice messages)

Enable the bot to autonomously decide whether to respond in group chats, like a human observer.

How it works:

                    消息进入
                        │
           ┌────────────┴────────────┐
           │                         │
      @bot / 触发？               否
           │                         │
           ▼                         ▼
    ┌──────────────┐         累积到缓冲区
    │ 立即触发      │         新消息重置防抖
    │ 必须回复      │               │
    └──────────────┘               ▼
                           防抖到期(3s无新消息)
                                   │
                                   ▼
                         双条件满足？
                         消息数 >= N 且 时间 >= T
                                   │
                              ┌────┴────┐
                            满足      不满足
                              │         │
                              ▼         ▼
                       交给 Agent    继续等待
                       让它决定
                       是否回复

1. Trigger mode (default): Bot only responds when @mentioned - must reply
2. Auto-reply mode: Bot observes group chat and decides whether to respond

Dual conditions (both must be met):

• minMessages: Accumulated message count threshold
• minTimeMs: Time elapsed since first message in buffer

Debounce: Every new message resets the debounce timer. Only when no new messages arrive for debounceMs, the conditions are checked.

Agent decision: If conditions are met, the agent receives all buffered messages with a system hint. The agent can:

• Reply normally → message sent to group
• Output [NO_RESPONSE] → silently dropped, no message sent

{
  "channels": {
    "feishu": {
      "autoReply": {
        "enabled": true,
        "minMessages": 5,
        "minTimeMs": 60000,
        "debounceMs": 3000
      }
    }
  }
}

Note: @mentions always trigger a response regardless of auto-reply settings.

Example with streaming enabled:

{
  "channels": {
    "feishu": {
      "enabled": true,
      "appId": "cli_xxx",
      "appSecret": "xxx",
      "blockStreamingCoalesce": {
        "enabled": true,
        "minDelayMs": 500,
        "maxDelayMs": 2000
      },
      "streamingCard": {
        "enabled": true,
        "title": "正在思考..."
      }
    }
  }
}

Configure multiple Feishu apps in a single OpenClaw instance:

{
  "channels": {
    "feishu": {
      "enabled": true,
      "domain": "feishu",
      "dmPolicy": "pairing",
      "accounts": {
        "main": {
          "appId": "cli_main_xxx",
          "appSecret": "secret1",
          "botName": "Main Bot"
        },
        "support": {
          "appId": "cli_support_xxx",
          "appSecretFile": "~/.secrets/support-bot.txt",
          "botName": "Support Bot",
          "dmPolicy": "allowlist",
          "allowFrom": ["ou_admin1", "ou_admin2"]
        }
      }
    }
  }
}

Each account inherits from the base config and can override any setting. The appSecretFile option allows reading secrets from a file instead of storing them in config.

Configure tool permissions based on message sender within groups:

{
  "channels": {
    "feishu": {
      "groups": {
        "oc_your_group_id": {
          "requireMention": false,
          "toolsBySender": {
            "张三": { "deny": ["shell", "write_file"] },
            "ou_admin_id": { "allow": ["*"] },
            "*": { "allow": ["web_search", "feishu_card"] }
          }
        }
      }
    }
  }
}

Resolution priority: toolsBySender[senderId] > toolsBySender[senderName] > tools > toolsBySender["*"]

The plugin registers the following tools that AI agents can use:

Send rich interactive cards with structured content:

{
  "tool": "feishu_card",
  "args": {
    "title": "Status Report",
    "titleTemplate": "green",
    "elements": [
      { "tag": "markdown", "content": "**Task**: Completed\n**Duration**: 3.2s" },
      { "tag": "hr" },
      { "tag": "action", "actions": [
        { "tag": "button", "text": { "tag": "plain_text", "content": "View Details" }, "type": "primary", "url": "https://example.com" }
      ]}
    ]
  }
}

Supported elements:

• div — Text block with plain_text or lark_md
• hr — Horizontal divider
• markdown — Markdown content block
• note — Footnote with nested elements
• action — Button row (supports url for links, type: default/primary/danger)
• column_set — Multi-column layout
• img — Image with img_key

Header colors: blue, wathet, turquoise, green, yellow, orange, red, carmine, violet, purple, indigo, grey

Retrieve message history from a chat:

{
  "tool": "feishu_list_messages",
  "args": {
    "chatId": "oc_xxx",
    "pageSize": 20
  }
}

Unlike typical bots that respond to each message immediately, this plugin processes messages like a human would:

1. Startup Window (10s): When the bot connects, it buffers all incoming messages
2. Trigger Detection: @mention (or other triggers) signals the bot should respond
3. Context Gathering: Bot reads ALL buffered messages, not just the trigger
4. Single Response: Bot responds once with full conversation context

Example: Bot was offline, 5 messages arrive:

  User A: "Let's discuss the project timeline"
  User B: "@bot what do you think?"        ← trigger
  User A: "Budget is around $100k"
  User C: "@bot please summarize"          ← trigger
  User A: "Deadline is next Monday"

OLD behavior: Bot responds twice (to each @mention), missing context
NEW behavior: Bot sees all 5 messages, understands full context, responds ONCE

Non-bot mentions are preserved with their open_id, enabling the Agent to @ users in responses:

Inbound message:  "Hi @张三 what do you think?"
Parsed content:   "Hi @[张三](ou_xxx) what do you think?"
ParsedMessage.mentions: [{ name: "张三", openId: "ou_xxx" }]

Agent response:   "I agree with @[张三](ou_xxx)'s point..."
Sent to Feishu:   "I agree with <at user_id="ou_xxx">张三</at>'s point..."

Bot mentions (@bot) are stripped completely to reduce noise.

The @mention is just the default trigger. The architecture supports:

• Keyword triggers: Respond when specific words appear
• Scheduled triggers: Periodic check-ins
• Custom triggers: Implement the Trigger interface

// src/core/triggers/index.ts
export interface Trigger {
  name: string;
  check(ctx: TriggerContext): boolean;
}

1. Go to Feishu Open Platform
2. Create a self-built app
3. Enable permissions: im:message, im:chat, contact:user.base:readonly
4. Events → Use Long Connection mode
5. Subscribe to event: im.message.receive_v1
6. Get App ID and App Secret from Credentials page
7. Publish the app

See CHANGELOG.md for version history.

MIT

Inspired by samzong/moltbot-channel-feishu.