Skip to content

Merwynkumar/clawblink

BranchesTags

Repository files navigation

ClawBlink – No‑Code AI Agent Builder from Telegram / WhatsApp Chat

Describe an AI agent in plain English over Telegram or WhatsApp and get results back in the same chat.
No code, no config files, no dashboards – just say what you want and ClawBlink builds and runs it.

You: "Every morning at 8am, check BBC news and send me a digest."
You: "Every 10 minutes, explain me a Python concept with a simple example."
You: "Monitor my GitHub repo for new issues and summarize them."

ClawBlink: Agent created.
  Trigger: scheduled (daily 08:00) | actions: http_request → llm_analyze → notify

✨ Highlights

  • No‑code agents – One chat message → working automation. News, prices, GitHub, reminders, tutorials.
  • Chat‑first – Create, run, and control agents from Telegram or WhatsApp. No web UI.
  • Channel‑native – Agents created in Telegram stay in Telegram; WhatsApp agents stay in WhatsApp (configs/telegram vs configs/whatsapp).
  • Local‑first – Ollama, Gemini free tier, or any OpenAI‑compatible API. Your data, your machine.
  • Smart URLs – At creation time, ClawBlink validates URLs and picks known-good sources (BBC, Python docs, prices, etc.) so agents get real data, not 404s.
  • Small & forkable – Minimal Python codebase, easy to read and extend.

🧩 Features

📰 24/7 News Digest
BBC News Digest		 💼 AI Job Radar
LinkedIn AI Job Monitor		 🧠 Python Skill Tutor
Python Concept Agent		 ⏰ Smart Routine Manager
Reminder Agents
Discovery • Insights • Trends Search • Filter • Alerts Learn • Practice • Code Schedule • Automate • Nudge

⌨️ CLI quick commands

# Telegram
clawblink gateway

# WhatsApp Web (QR login)
clawblink channels login      # link device (scan QR)
clawblink channels gateway    # Node WhatsApp gateway
clawblink whatsapp-bridge     # Python bridge

📰 News

  • 2026‑03‑16 – 🚀 ClawBlink launched. No-code AI agents from Telegram and WhatsApp – describe what you want in plain English, get running automations in under a minute.
  • 2026‑03‑15 – 🧠 Smart agent creation: URLs validated at creation time, intent-based URLs (news, Python, prices, weather), and chat-readable outputs so every agent delivers useful results.

💡 Why ClawBlink?

ClawBlink is inspired by projects like nanobot, OpenClaw, and OneClaw – but focused on one thing: let anyone spin up useful background agents from chat, in minutes, without learning a new framework.

nanobot OpenClaw / agent frameworks OneClaw-style SaaS ClawBlink
Create agents Write Python skills Design workflows / config Click UI, pay monthly Describe in plain English
Where it lives CLI + many channels Custom infra Hosted dashboard Your own Telegram bot and/or WhatsApp number
Setup Install + configure Multi-service setup Account + payment pip install + .env
Target user Developers Advanced devs Anyone (SaaS) Anyone who can chat
Codebase ~4k+ LOC Hundreds of k LOC Closed source ~1.9k LOC
Cost Free Free (but complex) $/month Free, self-hosted
  • If you want an ultra-flexible research agent platform, check out nanobot.
  • If you want a simple, forkable, no-code agent builder from chat, ClawBlink is for you.

🎯 What ClawBlink Can Do

Send a message to your Telegram bot or WhatsApp – ClawBlink turns it into a real agent.

Message What you get
"Every morning, BBC news digest" Scheduled agent, daily at 8am, fetch + summarize + notify.
"Every 10 minutes, explain a Python concept with an example" Educational agent with one concept + code example per run.
"Watch my GitHub repo for new issues and summarize" Polling agent → GitHub API → LLM summary → your chat.
"Bitcoin price every hour" Price agent using a free API, results in chat.
"Remind me to drink water every 2 hours" Simple reminder agent on a schedule.

Agents are saved as YAML under configs/telegram/ or configs/whatsapp/. You can edit them by hand or create everything from chat.

Why does agent creation take 30–60 seconds?
ClawBlink uses an LLM to turn your message into a YAML config. You get "Building your agent…" immediately; "Agent created!" when the LLM finishes. Use a faster model (e.g. Gemini or a smaller Ollama model) for quicker creation.

📦 Install

git clone https://github.com/Merwynkumar/clawblink.git
cd clawblink

pip install -r requirements.txt
pip install -e .    # installs the `clawblink` CLI
cp .env.example .env

Edit .env:

  • TELEGRAM_BOT_TOKEN from @BotFather
  • One LLM: OLLAMA_MODEL=... or GEMINI_API_KEY=... or OPENAI_API_KEY + base URL + model

Update later: git pull then pip install -e .

🚀 Quick Start

Telegram (recommended) 
clawblink gateway
  1. Create a bot with @BotFather, put the token in .env.
  2. In Telegram, open your bot → /start.
  3. Send a plain-English request (e.g. "Every morning, BBC news digest").

Commands: /list · /run <name> · /stop <name> · /config <name> · /status

WhatsApp (Web bridge)

Requires Node.js ≥18.

  1. Link device: clawblink channels login → scan QR in WhatsApp (Settings → Linked Devices).
  2. Run (two terminals): clawblink channels gateway and clawblink whatsapp-bridge.
  3. Chat from that WhatsApp number – same as Telegram: /start, then plain-English requests.

Agents created from WhatsApp are stored in configs/whatsapp/ and only appear in WhatsApp /list.

⚙️ How It Works

User message     "Every morning, BBC news digest"
       │
       ▼
ConfigBuilder    Plain English → YAML (with validated, intent-based URLs)
       │
       ▼
ConfigValidator  Structure + logic check
       │
       ▼
AgentRunner      Scheduler + triggers (scheduled / polling / manual)
       │
       ▼
Actions          http_request → llm_analyze → notify_telegram | notify_whatsapp
       │
       ▼
Result           Message in your chat

Data is always fetched by http_request first; the LLM only sees that text. No raw network access for the model.

📁 Project structure

clawblink/
├── main.py                 # Telegram entry point
├── interfaces/             # Chat interfaces (CLI, Telegram, WhatsApp bridge)
├── builder/                # LLM: message → YAML config + validation
├── engine/                 # Agent runner, triggers, and actions
├── providers/              # LLM providers (Ollama, Gemini, OpenAI-compatible)
├── configs/                # Saved agent configs (telegram/ and whatsapp/)
├── bridge/whatsapp/        # Node.js WhatsApp Web (QR) gateway
├── pyproject.toml          # Packaging + CLI entrypoint
├── requirements.txt        # Python dependencies
└── .env.example            # Example environment config

🧱 Architecture

High-level flow from your chat message to a running agent:

ClawBlink architecture

Key pieces:

  1. Interface layer (interfaces/)

    • Telegram bot or WhatsApp bridge receives your message.
    • Commands like /start, /list, /run <name> are parsed here.

  2. Config builder (builder/)

    • config_builder.py asks the LLM to turn plain English into a YAML agent config.
    • URL helpers fix news / jobs / prices sources and validate that endpoints actually return data.
    • config_validator.py checks the structure (trigger + actions are valid) before saving.

  3. Agent runner (engine/)

    • runner.py loads configs from configs/telegram or configs/whatsapp, starts a background scheduler, and tracks each agent’s runs.
    • Triggers (scheduled, polling, manual) decide when an agent should fire.

  4. Actions pipeline (engine/actions/)

    • http_request fetches data from APIs or websites.
    • llm_analyze turns raw data into a chat-sized summary, list, or explanation.
    • notify_telegram / notify_whatsapp send the final message back to the right chat.

  5. LLM providers (providers/)

    • SmartProvider picks the best available LLM (Ollama, Gemini, or any OpenAI-compatible endpoint) and handles timeouts and fallbacks.

LLMs never call the network directly: all external data goes through http_request, then llm_analyze works only on that text. This keeps agents predictable and debuggable.

⏱️ Triggers

Type Description Config
scheduled Fixed interval or daily at a time interval_minutes and/or time_local: "08:00"
polling GitHub issues or URL change check source, repo or url, interval_minutes
manual Only when you run /run <name> —

🔧 Actions

Type Description
http_request GET/POST URL, store in variable. Headers can use env:VAR for API keys.
llm_analyze Run LLM on fetched data; output chat-readable text.
notify_telegram Send message to Telegram (chat_id auto-filled when created from Telegram).
notify_whatsapp Send message via WhatsApp Web bridge (to set when created from WhatsApp).

Pattern: http_request → optional llm_analyze → notify_telegram or notify_whatsapp.

📋 Dependencies

  • requests – HTTP client
  • python-telegram-bot – Telegram bot API
  • pyyaml – YAML config parsing
  • flask – WhatsApp Web bridge server

🤝 Contributing

  • New triggers – RSS, webhooks, email, or other event sources.
  • New actions – Slack/Discord notifications, email, file writes, etc.
  • Smarter builder – More intent→URL mappings, better validation heuristics.
  • Example agents & docs – Great starter agents or walkthroughs.

If you build something with ClawBlink, share it in an issue or PR.

📄 License

MIT

About

ClawBlink : No‑code AI agent builder from chat. Describe an agent in plain English; ClawBlink builds and runs it.

Topics

automation telegram-bot whatsapp-bot no-code awesome-bot ai-agent llm awesomeai

Resources

Readme

License

MIT license
Activity

Stars

4 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages