Why Google Execs Are Rushing to Deploy Clawdbot: The AI Agent Sensation

What is Clawdbot?

Clawdbot is an open‑source AI‑agent gateway installed locally on a computer. It runs with full system privileges and can be invoked at any time through WeChat, Telegram or Discord. The core architecture follows a "message app ⇄ gateway ⇄ AI model + tools" pattern.

Runtime logic : a message sent from a phone is received by the local gateway, which calls a large model (e.g., Claude or GPT) to interpret the intent, translates the intent into a shell script or browser operation, and executes it directly on the host machine.

Data sovereignty : all conversations, memories and logs are stored as Markdown files on the local disk, ensuring that no user data is uploaded to external services.

Why it draws strong interest

From advisor to actor : instead of merely describing how to register a domain, Clawdbot can open the browser, fill the form, complete payment and return the credentials. It can read/write files, run Bash commands, control browsers, manage calendars and even operate smart‑home devices. The author labels this paradigm “Software as Conversation (SaaC)”.

Extreme privacy : only the model inference requires network access; all routing, automation and memory storage remain on the local machine, giving programmers, lawyers and finance professionals full control over their data.

Zero‑friction interaction : no additional app or browser tab is needed. While coding, a user can request documentation or restart a server by sending a single message in WeChat/Telegram and receive an immediate response.

Self‑evolving capability : the platform supports plugins and skill extensions. When asked to fetch Twitter trends, the agent can study the Twitter API, generate a scraping script, and notify the user that the new skill has been added.

Notable endorsements

Andrej Karpathy (former Tesla AI director) publicly expressed enthusiasm for Clawdbot, calling it a future form of AI agents.

Logan Kilpatrick (Google AI Studio lead) ordered a Mac mini specifically to try Clawdbot.

Dave Morin (venture investor) said it felt like the first time since ChatGPT that he truly felt “living in the future”.

Federico Viticci (founder of MacStories) reported burning 180 million Anthropic tokens in a week to test the system and replacing Zapier with Clawdbot for automation.

Functional categories

Office automation : automatically sort tens of thousands of emails, draft replies, generate weekly reports, monitor RSS feeds and sync results to Notion.

DevOps : merge pull requests, fix bugs, restart servers from a phone, and scrape 4 million tweets for analysis.

Life concierge : adjust boiler temperature based on weather, send scheduled good‑morning/evening messages to family, and place restaurant reservations automatically.

Cost‑killer : replace paid automation platforms such as Zapier with local scripts that incur no additional runtime cost.

Step‑by‑step deployment guide

Three main deployment streams are common in China:

One‑click deployment on Tencent Cloud or Alibaba Cloud (convenient and stable).

Docker local/VPS manual deployment (flexible and controllable).

VM isolation deployment (security sandbox).

The deployment formula is:

Gitee mirror source + Docker isolation + domestic large‑model API (MiniMax/GLM) + Matrix/Feishu channel

Purchase a server.

Select an image: search for “Moltbot” or “Clawdbot” in the cloud provider’s application marketplace and choose the official image.

Initialize the instance:

Model selection: MiniMax or Kimi.

API Key: obtain the key from the chosen model provider and paste it into the configuration.

Channel: beginners may skip this step and configure it later.

Log in to the console → “OrcaTerm” password‑less login.

Run the onboarding command: clawdbot onboard Install and start the daemon:

# Exit temporary run (Ctrl+C)
clawdbot daemon install  # Install daemon
clawdbot daemon start   # Start daemon

Open a browser and navigate to http://YOUR_PUBLIC_IP:18789 to access the web management dashboard.

When the agent responds, the digital employee is online.

Cost consideration

API token consumption is very fast; heavy usage can burn 180 million tokens in a week, so users must be prepared for the associated expense.