Channel Overview

GolemBot's gateway connects your assistant to IM platforms. Each platform is handled by a channel adapter that translates between the IM SDK and GolemBot's assistant.chat() API.

Supported Channels

Channel	Transport	Public IP Required	SDK
Feishu (Lark)	WebSocket	No	@larksuiteoapi/node-sdk
DingTalk	Stream (WebSocket)	No	dingtalk-stream
WeCom	WebSocket	No	@wecom/aibot-node-sdk
Slack	Socket Mode (WebSocket)	No	@slack/bolt
Telegram	Long-polling	No	grammy
Discord	Gateway WebSocket	No	discord.js
WeChat (微信)	Long-Polling	No	None (built-in fetch)
Custom	Any	Depends	Your own adapter class

All channels work behind NAT

All 7 built-in channels use WebSocket, long-polling, or Socket Mode — no public IP or port forwarding required.

Architecture

IM Platform → Channel Adapter → assistant.chat() → text response → adapter.reply()

The gateway:

1. Reads channels config from golem.yaml
2. Dynamically imports the SDK for each configured channel
3. Starts all adapters in parallel alongside the HTTP service
4. Routes incoming messages through session key resolution → context building → assistant.chat()
5. Accumulates the full text response, splits it within platform limits, and sends chunk by chunk

Message Limits

Each platform has a maximum message length. GolemBot automatically splits long responses:

Channel	Max length	Split behavior
Feishu	4,000 chars	Multi-message
DingTalk	4,000 chars	Multi-message
WeCom	2,048 chars	Multi-message
Slack	4,000 chars	Multi-message
Telegram	4,096 chars	Multi-message
Discord	2,000 chars	Multi-message
WeChat	2,000 chars	Multi-message
Custom	Configurable via maxMessageLength	Multi-message

Message Format Conversion

GolemBot automatically converts standard Markdown to each platform's native format:

Channel	Output Format	Conversion
Feishu	Card v2 (interactive)	Native Markdown rendering — headings, lists, code blocks, tables
Slack	mrkdwn	**bold** → *bold*, *italic* → _italic_, links rewritten
Telegram	HTML	**bold** → <b>, code blocks → <pre><code>, blockquotes → <blockquote>
Discord	Markdown (native)	No conversion needed — Discord renders Markdown natively
DingTalk	Markdown (native)	Passed through as-is
WeCom	Plain text	Markdown stripped (WeCom text API has limited formatting)
WeChat	Plain text	Passed through as-is (WeChat text API)

The AI agent is encouraged to use standard Markdown syntax (headings, lists, bold, code blocks, etc.) — each adapter handles the platform-specific conversion automatically.

Group Chat, Mentions & Quote Reply

GolemBot supports group chat response policies (mention-only / smart / always), incoming and outgoing @mention resolution, and quote reply on supported platforms.

See Group Chat for full details.

Inbox & History Fetch

When enabled, messages are queued persistently and the bot catches up on missed messages after restart via intelligent triage.

See Inbox & History Fetch for the full guide and platform support table.

Custom Adapters

Connect any platform by writing an adapter class and referencing it with _adapter in golem.yaml:

channels:
  my-platform:
    _adapter: ./adapters/my-platform.mjs   # or npm package name
    token: ${MY_PLATFORM_TOKEN}

See Channel Adapter API for the full interface, implementation guide, and examples.

Image Support (Multimodal)

GolemBot supports image messages across all 7 built-in channels. When a user sends an image (photo, screenshot, file attachment), the adapter downloads it and passes it through the full pipeline:

User sends image → Adapter downloads to Buffer → gateway passes to assistant.chat()
→ image saved to .golem/images/ → file path appended to prompt → agent reads the file
→ cleanup after response

Channel	Image Source	Caption/Text
Feishu	image messages + inline images in post (rich text)	Post text content extracted
Slack	File attachments with image content type	Message text preserved
Telegram	message.photo (picks largest size)	message.caption used as text
Discord	message.attachments with image content type	Message text preserved
DingTalk	picture messages + images in richText	Rich text content extracted
WeCom	image messages via media API	Text set to (image)
WeChat	CDN download + AES-128-ECB decrypt	Text set to (image)

How it works: Images are saved as temporary files in .golem/images/ and referenced by absolute path in the prompt. This works universally with all engines (Cursor, Claude Code, OpenCode, Codex) since every coding CLI can read local files. For Codex specifically, GolemBot also forwards image attachments natively via --image <path>. Files are automatically cleaned up after the agent responds.

HTTP API: The POST /chat endpoint also accepts base64-encoded images — see HTTP API.

Proactive Messaging (Scheduled Tasks)

Beyond responding to incoming messages, GolemBot can proactively send messages to IM channels on a schedule. Define tasks in golem.yaml and the agent will execute prompts automatically, pushing results to your IM channels.

See Scheduled Tasks for full configuration, management commands, and use case examples.

Starting the Gateway

golembot gateway --verbose

The --verbose flag enables per-channel log lines, useful for debugging.

SDK Dependencies

Channel SDKs are optional peer dependencies. Install only what you need:

# Feishu
pnpm add @larksuiteoapi/node-sdk

# DingTalk
pnpm add dingtalk-stream

# WeCom
pnpm add @wecom/aibot-node-sdk

# Slack
pnpm add @slack/bolt

# Telegram
pnpm add grammy

# Discord
pnpm add discord.js

# WeChat (no SDK needed — uses built-in fetch)
# Just add the token to golem.yaml

If a configured channel's SDK is not installed, the gateway will print an error with installation instructions.

What's Next

• Feishu, DingTalk, WeCom, WeChat, Slack, Telegram, Discord — per-channel setup guides
• Group Chat — response policies, @mention, quote reply
• Inbox & History Fetch — crash-safe queue, offline catch-up
• Channel Adapter API — build custom adapters