$249 · Commercial license

Developer — Message

One AI bot, eight messaging platforms, two markets. Streaming chat with per-conversation memory, in-chat media generation, rate limiting — and a single environment variable to switch between the Western and Chinese channel sets.

Requirements

Node 22 (the codebase targets Node 20.19+; 22 is what the repo pins)
An API key from at least one AI provider
Bot credentials for whichever channels you want to run
Redis — optional. Without it, session memory falls back to in-process storage, which is lost on restart.

Quickstart

npm install
cp .env.example .env.local
npm run dev

It boots with nothing configured. With no keys and no bot tokens, every channel logs that it is not connecting and the bot runs in dry-run mode. Nothing crashes. Add credentials one channel at a time.

Markets — `PROFILE=us` or `PROFILE=cn`

This is the core design decision of the edition. One codebase serves both markets, and a single variable picks which channel set boots:

`PROFILE`	Channels that start
`us`	Discord, Telegram, Slack, WhatsApp
`cn`	WeCom (企业微信), Feishu (飞书), DingTalk (钉钉), WeChat Official Accounts

Channels outside the active profile are never initialised, so you do not need their credentials. Set PROFILE in .env.local.

Configuration

Core

Variable	Purpose
`PROFILE`	`us` or `cn` — which market's channels to run
`AI_MODEL`	Which chat model to use
`SYSTEM_PROMPT`	Override the bot's system prompt
`BOT_PRESET`	Load a preset persona (see `examples/presets.ts`)
`REDIS_URL`	Session memory and rate limiting. Optional; falls back to in-process.

AI providers

Variable	Enables
`OPENAI_API_KEY`	GPT models
`ANTHROPIC_API_KEY`	Claude models
`GOOGLE_GENERATIVE_AI_API_KEY`	Gemini models
`ZAI_API_KEY`	GLM models

Media

Variable	Purpose
`FAL_KEY`	fal.ai — image, video, speech
`REPLICATE_API_TOKEN`	Replicate — the alternate provider
`MEDIA_PROVIDER`	`fal` (default) or `replicate`

Channel setup

Each channel needs its own credentials from that platform's developer console. Set only the ones you intend to run. Per-channel walkthroughs ship in the repo under docs/channels/.

Western channels (`PROFILE=us`)

Channel	Variables	Where to get them
Discord	`DISCORD_BOT_TOKEN` `DISCORD_APP_ID`	Discord Developer Portal → your application
Telegram	`TELEGRAM_BOT_TOKEN`	Talk to @BotFather
Slack	`SLACK_BOT_TOKEN` `SLACK_APP_TOKEN`	Slack API → your app (Socket Mode)
WhatsApp	`WHATSAPP_ACCESS_TOKEN` `WHATSAPP_PHONE_NUMBER_ID` `WHATSAPP_VERIFY_TOKEN` `WHATSAPP_APP_SECRET`	Meta for Developers → WhatsApp Business API

Chinese channels (`PROFILE=cn`)

Channel	Variables	Where to get them
WeCom 企业微信	`WECOM_CORP_ID` `WECOM_CORP_SECRET` `WECOM_AGENT_ID` `WECOM_TOKEN` `WECOM_ENCODING_AES_KEY`	WeCom admin console → your app
Feishu / Lark 飞书	`FEISHU_APP_ID` `FEISHU_APP_SECRET` `FEISHU_VERIFICATION_TOKEN`	Feishu Open Platform
DingTalk 钉钉	`DINGTALK_CLIENT_ID` `DINGTALK_CLIENT_SECRET`	DingTalk Open Platform
WeChat OA 微信公众号	Configured per the WeChat Official Account console	WeChat MP platform

WeCom uses encrypted callbacks. The WECOM_TOKEN and WECOM_ENCODING_AES_KEY pair drives message signature verification and AES decryption — both are required, and the implementation is in src/adapters/wecom-crypto.ts. This is the fiddliest of the eight; the walkthrough in docs/channels/wecom.md covers the callback URL setup step by step.

Webhook vs. socket channels

Discord, Telegram, and Slack connect outbound (long-poll or socket) and need no public URL. WhatsApp, WeCom, Feishu, DingTalk, and WeChat OA are webhook-based — the platform calls you, so they need a publicly reachable HTTPS endpoint. In local development, use a tunnel (ngrok, Cloudflare Tunnel) and register that URL in the platform's console.

Media commands

Media generation works in-chat, on every channel:

Command	Produces
`/imagine <prompt>`	An image
`/video <prompt>`	A video
`/speak <text>`	Speech audio

Results are returned as native attachments where the platform supports it, and as links where it does not.

Sessions and rate limiting

Session memory is per-conversation: the bot remembers context within a thread or DM. Backed by Redis when REDIS_URL is set; otherwise held in process and lost on restart.

Rate limiting is applied per user to stop one person exhausting your API budget. Configure it in src/core/rate-limit.ts.

Examples

Runnable examples ship in examples/: a support bot, a daily digest bot, a media studio bot, and a community Q&A bot. Each is a working starting point rather than a snippet.

Deploying

Deploy configs are included for Fly (fly.toml) and Railway (railway.json), plus a Dockerfile for anywhere else.

npm run build
npm start

In production you will want:

A managed Redis → REDIS_URL, so sessions survive restarts
A public HTTPS URL, if you run any webhook channel
The callback URL registered in each platform's console

Full detail is in docs/deploy.md in the repo.

What this edition does not include

No billing. Unmetered by design. To run a paid bot business — credits, payment rails, multi-tenancy — see Business.