Hey everyone,
In last session we talked about DeepSeek V4 — the most powerful open source AI model on the planet right now. Free, MIT licensed, runs locally, rivals GPT-5.4 on coding benchmarks.
But here's the thing nobody talks about after "here's a great model" — what do you actually do with it?
Today I want to show you exactly that. We're going to connect DeepSeek V4 to WhatsApp using an open source tool called OpenClaw. The result is a personal AI agent that lives in your phone, remembers everything, and actually does things for you — not just talks.
No subscriptions. No data leaving your machine. Just your own private JARVIS running 24/7.
Let's get into it.
🦞 What is OpenClaw?
OpenClaw started life in November 2025 as a side project by Austrian developer Peter Steinberger. He called it Clawdbot. Within two months it had been renamed twice, survived a trademark dispute with Anthropic, and exploded to over 100,000 GitHub stars. Developers were calling it the closest thing to JARVIS that actually exists.
Here's why it's different from every other AI tool you've tried.
Most AI tools are reactive. You open a browser tab, you type something, you get an answer, you close the tab. There's no memory. There's no action. It just responds and forgets you exist.
OpenClaw is a proactive agent. It lives inside the messaging apps you already use — WhatsApp, Telegram, Slack, Discord, iMessage. It remembers your context across every conversation. And it doesn't just talk — it actually executes tasks. Sends emails. Runs terminal commands. Manages files. Monitors your GitHub repos. Schedules meetings. All from a simple message on your phone.
Think of it less like a chatbot and more like a coworker who's always available, never forgets anything, and works while you sleep.
🧠 Why pair it with DeepSeek V4?
OpenClaw is model-agnostic — you bring your own AI brain. You can connect it to OpenAI, Anthropic, or any local model running via Ollama.
DeepSeek V4 is the obvious choice right now for three reasons.
First, it's free. MIT licensed, self-hostable, no API bill. Second, it's private — your conversations stay on your machine, not some company's server. Third, after last issue you already know what it can do. Coding, reasoning, long document processing — it handles all of it.
Together they give you something genuinely new: a powerful AI agent running entirely on your own infrastructure, accessible from your phone, that costs you nothing per month.
⚙️ How to set it up — the simple version
This isn't a full tutorial but here's the overview so you know what you're getting into before you commit an afternoon to it.
Step 1 — Install Ollama If you followed Issue 03 you might have this already. If not, go to ollama.com and install it. One command pulls DeepSeek V4 Flash locally — ollama pull deepseek-v4-flash. The Flash version is smaller and faster, perfect for agent use.
Step 2 — Install OpenClaw Run npx openclaw onboard in your terminal. This is the guided setup that walks you through configuring the gateway, workspace, and channels step by step. It takes about 20 minutes if you follow the prompts carefully.
Step 3 — Connect your channel Pick WhatsApp or Telegram — both work well. Telegram is easier to set up, WhatsApp feels more natural if that's your main app. Follow OpenClaw's channel docs for the specific steps — they're well written and kept current by the community.
Step 4 — Point it at DeepSeek In your openclaw.json config file, set the model to your local Ollama endpoint. The config is straightforward JSON — even if you're not a backend developer you can follow the example in the docs.
Step 5 — Send your first message Open WhatsApp. Message your new agent. Say hello. It responds. That first moment genuinely feels like magic — a local AI running on your laptop, talking back to you through your phone.
🛠️ Three things to build with it this week
Morning briefing agent — Ask it every morning to summarise your unread emails, check your calendar, and flag anything urgent. Set it up once with a cron job and it messages you automatically before you even open your laptop.
GitHub PR monitor — Connect it to your repos via the GitHub skill. It watches for new pull requests, failed CI builds, or open issues assigned to you and pings you on Telegram the moment something needs your attention.
Personal knowledge assistant — Point it at your Obsidian vault or notes folder. Ask it questions about your own notes in plain English. It searches, summarises, and connects ideas across everything you've written — like having a second brain that actually talks back.
⚠️ One honest warning
OpenClaw is powerful and it's also genuinely risky if you're not careful. It can run terminal commands and access your file system — which means a misconfigured skill or a malicious plugin could do real damage. Security researchers have already flagged vulnerabilities in third-party skills.
Run it in Docker or on a dedicated VPS rather than your primary machine. Don't install community skills you don't understand. Use allowlists to restrict which contacts can interact with your agent.
The tool's own maintainer said it plainly — if you can't read a command line, this might not be for you yet. That's not a reason to avoid it — it's a reason to learn before you deploy.
🛠️ Dev Tip of the Week
When setting up any AI agent, start with the smallest possible permission set. Don't give it file system access, email access, and terminal access all at once on day one. Add one capability at a time, test it, understand what it's doing, then expand. Agents that can do everything are also agents that can break everything.
This is where AI is actually going — not smarter chatbots, but agents that live in your existing tools and work alongside you. OpenClaw plus DeepSeek V4 is the clearest example of that future you can run today, for free, on your own hardware.
Tried OpenClaw yet, or planning to? Hit reply and let me know — I read every message.
— Dhanush from Tech Zenith