Full docs and setup guide here: molt.bot
If you found this useful, give it a repost — more people should know how to build their own AI agents
And follow for more breakdowns like this
Security is the real consideration here
This tool gets full access to your system — files, shell, browser, credentials
Researchers have already found hundreds of exposed instances leaking API keys online
Run it on isolated hardware, start with limited permissions, expand slowly
Here's what people are doing with it
→ One user negotiated a $4,200 discount on a car — Moltbot emailed dealers and handled back-and-forth
→ Another configured it to catch and fix production bugs via Slack
→ Someone has it managing their home heating based on live weather
→ Inbox automation
50+ skills come built-in
→ Calendar and email management
→ Browser automation via Puppeteer
→ GitHub commits and deployments
→ File system and shell access
→ Smart home integrations
But here's the thing —
It doesn't wait for you to ask and can initiate convos with briefings, alerts, and reminders
If moltbot throws an error, try clawdbot instead
The rename happened days ago and the old command still works as a fallback
After setup, you're chatting with your bot directly in terminal
To connect via Telegram, search for your bot's username and send any message
It responds with a pairing code
Run this:
moltbot pairing approve telegram <CODE>
Now you can message it from your phone
When the wizard runs, here's what works:
→ Choose "Quickstart"
→ Select your LLM and paste the API key
→ Enable Brave search when prompted
→ Pick Telegram as your messaging channel
→ Skip Google Places unless you need weather
→ Enable all memory hooks
Start with one skill — you can add more later
Installation takes two commands
First, make sure you're on Node 22+
Then run:
curl -fsSL molt.bot/insta… | bash
Followed by:
sudo moltbot onboard --install-daemon
The setup wizard handles the rest
Before you install, collect these keys
→ Brave API key (free — enables web search)
→ API key from your LLM provider (OpenAI, Anthropic, or Google)
→ Telegram bot token from BotFather
Create a new bot in Telegram, grab the token, save it somewhere safe
What you'll need
→ Claude Pro, ChatGPT Plus, or Google AI Pro ($20/month)
→ Mac, Linux, Windows via WSL2, or a Raspberry Pi
Critical: This is experimental software with full system access
Use a secondary machine or VPS — not your daily driver
Who should try this?
→ Anyone with a laptop and an AI subscription
→ People curious about building their own AI agents
→ Those tired of assistants that suggest but never do
No coding background required
Moltbot just crossed 60,000 GitHub stars in under a week
It's a self-hosted AI assistant that actually executes tasks — not just talks about them
Here's how to set it up from scratch:
Why it matters:
Fewer tokens = faster inference, lower costs at scale
In production, repetition errors dropped 33% compared to their previous model
Code and weights are open source: github.com/deepseek-ai/...
DeepSeek’s 2025 run was historic. Their 2026 is starting the same way
The results:
→ 91.09% on OmniDocBench (standard doc parsing benchmark)
→ Beats GPT-4o, Gemini 2.5 Pro, Qwen3 235B, InternVL 241B
→ Uses only 256-1120 visual tokens/page (competitors use 6000-7000+)
→ Reading order accuracy jumped — Edit Distance dropped 0.085 to 0.057
More accurate and efficient
Current vision models read images like a typewriter — left to right, top to bottom
But humans don’t read that way. We follow meaning, not coordinates
DeepSeek-OCR 2 lets the AI decide the reading order based on what it sees: tables, formulas, columns, whatever
They call it “Visual Causal Flow”
DeepSeek dropped another one!
DeepSeek-OCR 2 just mass-outperformed GPT-4o, Gemini 2.5 Pro, and Qwen 72B on document parsing
While using 5-6x fewer visual tokens
The whale keeps shipping
Here’s what’s actually new:
The creator Peter Steinberger (ex-PSPDFKit founder) built this as a personal project.
It’s now at 18,700 GitHub stars with 30+ PRs a day.
Not plug-and-play. But for devs who understand tradeoffs?
This is Jarvis demo that actually works.
The lobster 🦞 is loose. Handle with care.
But this power comes with real risk
It can read your files, run commands, and send emails. Prompt injection could trigger unintended actions.
One tester asked it to run find ~ and it dumped his directory structure to a group chat
If you try it: run on dedicated hardware with low-risk tasks
One founder’s Clawdbot “Henry” did 48 hours of non-stop coding while he didn’t touch Claude Code once
Another guy rebuilt his entire website via Telegram while watching Netflix in bed
A Redditor ported a CUDA backend to AMD ROCm in 30 minutes
Clawdbot can run on a dusty laptop, a $5/mo VPS, a Raspberry Pi, anything with 1GB RAM and Node.js.
So what’s the hype about?
Clawdbot can control your computer:
• Sends, responds to emails
• Manages calendar
• Runs terminal
• Writes, tests, deploys code
• Messages WhatsApp before you even ask
An open-source project just made Mac Mini sales spike to Black Friday levels.
Clawdbot — a personal AI assistant — went viral this week
Google’s Head of AI products ordered a Mac Mini for it. One developer bought 12. Another bought 40
The funny part? You don’t need a Mac Mini at all
💬 Which story do you think has the biggest long-term impact?
🔄 Repost if you’re tracking AI daily
👀 Follow for tomorrow’s update
🎮 TextArena Launches on Hugging Face
An open-source suite of 57+ games designed to test LLM behaviors: negotiation, deception, and more.
Basically, it’s an AI testing ground for agentic skills.
Read more: lnkd.in/gajaR6FC
🎙️ PlayAI’s Wild Voice Changer
With just 10 seconds of audio, you can speak in anyone’s voice.
It even keeps tone and style — and it works on mobile.
Read more: lnkd.in/gTG2gfRh
🧠 ElevenLabs Enables Agent Transfers
You can now pass conversation (and related context) between multiple specialized agents — perfect for support workflows and handoffs.
Read more: lnkd.in/g9ztFDz2
💸 Windsurf Simplifies Pricing
No more flow action credits. Now it’s just 1 prompt = 1 message to Cascade.
They’ve also merged all individual plans into a $15/month Pro plan (500 credits).
Read more: lnkd.in/gCQmhum4
🎬 AI Officially Allowed in the Oscars
The Academy says AI use in film production will “neither help nor harm” chances of a nomination.
It’s a major shift — and a big greenlight for AI’s role in storytelling.
Read more: lnkd.in/gBy9zHZ7
💰 Elon Musk Seeks $25B+ for xAI-X JV
He’s reportedly raising massive capital for the combined xAI–X venture, aiming at a $150-200B valuation.
Another big bet in the AI arms race.
Read more: lnkd.in/gsjTEXZs
🧭 Anthropic Maps Claude’s Morality
After analyzing 300K+ chats, Anthropic found Claude upholds 5 core values — like being practical and protective.
The work will help make AI more aligned for real-world use.
Read more: lnkd.in/g_WkdFPP
🧬 Google DeepMind’s Big Vision
Nobel laureate and DeepMind CEO Demis Hassabis said on 60 Minutes that AGI could arrive in 5–10 years — and may become conscious at some point.
He also believes AI could fast-track drug discovery enough to eradicate all disease.
Read more: lnkd.in/g3hNPmtZ