ClewMind

An OpenSource AI agent that speaks your language. Memory that's shared.

Send a voice note in Urdu, Hindi, or Arabic — your whole family or team benefits. Open source, self-hosted, free to run.

---

What Makes ClewMind Different

Feature	ClewMind	ChatGPT	Discord bots	Hermes Agent
Voice-first (hold to speak)	✅	❌	❌	❌
Shared team memory	✅	❌	❌	Partial
Non-English native	✅ Urdu/Arabic/Hindi	English-first	English-first	English-first
Works on WhatsApp	✅	❌	❌	❌
Self-hosted, no subscription	✅	❌	❌	Partial
Mobile-first web UI	✅	Partial	❌	❌

ClewMind is built for families and small teams who communicate in non-English languages on WhatsApp — not for English-speaking developers with Slack.

---

Quick Start

git clone https://github.com/yourname/clewmind
cd clewmind
bash setup.sh

Then edit .env:

GROQ_API_KEY=your_key_here   # free at console.groq.com

npm run dev
# → http://localhost:3000

That's it. The web UI works immediately — no other config needed.

Add Discord

1. Create a bot at discord.dev/applications
2. Copy the token to .env:

   DISCORD_BOT_TOKEN=your_token_here

3. Restart — ClewMind joins Discord automatically.

Add WhatsApp

WHATSAPP_SESSION_NAME=clewmind

Restart and scan the QR code printed in the terminal with your phone.

---

How Shared Memory Works

This is ClewMind's core feature. Memory is per-team, not per-user.

🗣️  Ahmed (voice note in Urdu):
    "Kal Lahore jana hai train se"

🧠  ClewMind:
    "Noted! I'll remember your Lahore trip tomorrow by train."

— later —

🗣️  Ammi: "Ahmed ka koi plan hai kal ka?"

🧠  ClewMind:
    "Ji! Ahmed ne bola tha ke kal Lahore ja rahe hain train se."

Every fact is extracted automatically — no /save command needed. Facts are scoped to the team (Discord server, WhatsApp group, or web session), so anyone in the team can ask about anything anyone else has mentioned.

---

Supported Languages

ClewMind detects language automatically and replies in kind.

Language	Code	STT	TTS
🇵🇰 Urdu	ur	✅	✅
🇸🇦 Arabic	ar	✅	✅
🇮🇳 Hindi	hi	✅	✅
🇧🇩 Bengali	bn	✅	✅
🇮🇩 Indonesian	id	✅	✅
🇹🇷 Turkish	tr	✅	✅
🇨🇳 Chinese	zh	✅	✅
🇪🇸 Spanish	es	✅	✅
🇵🇹 Portuguese	pt	✅	✅
🇫🇷 French	fr	✅	✅
🇩🇪 German	de	✅	✅
🇮🇹 Italian	it	✅	✅
🇷🇺 Russian	ru	✅	✅
🇯🇵 Japanese	ja	✅	✅
🇬🇧 English	en	✅	✅

STT via Groq Whisper · TTS via Microsoft edge-tts (free, no key)

---

Architecture

┌─────────────────────────────────────────────────────────────┐
│  Channels                                                   │
│  ┌──────────┐  ┌─────────────┐  ┌────────────────────────┐ │
│  │ WhatsApp │  │   Discord   │  │  Web (Express + WS)    │ │
│  └────┬─────┘  └──────┬──────┘  └───────────┬────────────┘ │
│       └───────────────┴──────────────────────┘             │
│                        │ IncomingMessage                    │
│                   ┌────▼──────────────────┐                 │
│                   │       Engine          │                 │
│                   │  1. Commands (/help)  │                 │
│                   │  2. STT (Whisper)     │                 │
│                   │  3. Skills router     │                 │
│                   │  4. Memory context    │                 │
│                   │  5. LLM (Groq)        │                 │
│                   │  6. Fact extraction   │                 │
│                   │  7. TTS (edge-tts)    │                 │
│                   └────────────────────┬─┘                 │
│                                        │ OutgoingMessage   │
│  ┌─────────────────────────────────────▼──────────────────┐│
│  │  Core                                                  ││
│  │  memory.ts · llm.ts · voice.ts                        ││
│  └────────────────────────────────────────────────────────┘│
└─────────────────────────────────────────────────────────────┘

Message flow: Channel → Engine → (skill OR LLM) → async fact extract → (TTS?) → reply

---

Skills

Three skills ship out of the box:

Skill	Triggers	Example
Reminders	remind me, in 2 hours, kal subah, at 5pm	"remind me to call Ali in 30 minutes"
Notes	note:, save this, shopping list, yaad rakhna	"note: server password is abc123"
Web Search	search for, who is, google, dhoondho	"who is the president of Pakistan"

Build a Custom Skill

import type { Skill, SkillContext } from '../types.js'

export const WeatherSkill: Skill = {
  name: 'weather',
  description: 'Get current weather for a city',

  canHandle(message: string): boolean {
    return /\bweather\b/i.test(message) || message.includes('barish')
  },

  async execute(message: string, ctx: SkillContext): Promise<string> {
    const city = extractCity(message)
    const data  = await fetchWeather(city)
    return `${city}: ${data.temp}°C, ${data.description}`
  },
}

Register it in src/core/engine.ts → loadSkills().

---

Commands

Command	Description
/start	Welcome message
/help	List all commands
/memory	Show team memories
/forget <text>	Delete matching memories
/team	List team members
/private <msg>	Chat without saving to memory
/name <name>	Change your display name

---

Environment Variables

Variable	Required	Default	Description
GROQ_API_KEY	Yes	—	LLM + Whisper STT (get free)
OPENROUTER_API_KEY	No	—	Fallback LLM provider
DISCORD_BOT_TOKEN	No	—	Enables Discord channel
WHATSAPP_SESSION_NAME	No	—	Enables WhatsApp channel
DEFAULT_LANGUAGE	No	en	Fallback language code
VOICE_ENABLED	No	true	Set false to disable TTS
DATA_DIR	No	./data	Where to store memory/notes/reminders
WEB_PORT	No	3000	Web UI port

---

Roadmap

• Web chat UI with voice recording
• WhatsApp + Discord channels
• Shared team memory with auto-extraction
• Reminders, Notes, Web Search skills
• Multilingual STT + TTS (15 languages)
• Persistent reminder restore on server restart
• WhatsApp group mention (@clewmind)
• Telegram channel
• Image understanding (describe photos in voice)
• Skill marketplace / plugin install via npm
• Multi-team isolation with invite codes
• Admin dashboard for memory management

---

Contributing

PRs welcome. High-priority areas:

• New channels — Telegram, SMS (Twilio), email
• New skills — Weather, currency converter, prayer times
• Language support — More TTS voices, RTL UI fixes
• Tests — Integration tests for the engine pipeline

git clone https://github.com/yourname/clewmind
cd clewmind && bash setup.sh
npm run dev

File structure is intentionally flat — each file has one job. Read src/types.ts first to understand the data model, then src/core/engine.ts for the message pipeline.

---

License

MIT © 2025 — free to use, modify, and self-host.