Open-Source AI: Full Computer Access Explained

Peter Steinberger
, the engineer who founded and sold PSPDFKit (now Nutrient) for a substantial sum, was inspired to create Clawdbot. This open-source AI assistant gained immense popularity over the weekend, quickly amassing 15,000 to 20,000 stars on GitHub in a matter of days.

Clawdbot operates locally, integrating models like Claude with the capacity to control browsers, manage files, and execute shell commands, acting as a privacy-focused digital assistant across platforms like WhatsApp, Telegram, Discord, and Slack.

The installation process is straightforward (available on Mac, Windows, Linux, and even Raspberry Pi). A single curl command for installation followed by quick onboarding grants it access to your preferred messaging services like WhatsApp, Telegram, Signal, Discord, Slack, iMessage, and Teams. It is powered by a single Gateway process that manages channels via WebSocket and lightweight nodes for efficient processing.

The appeal lies in its autonomy: Clawdbot can operate Chrome for various tasks such as booking flights, automating check-ins, scraping websites, filling out forms, reading and writing files, executing shell commands, and even drafting and publishing blog posts without prompts. Its persistent memory retains context across days, allowing it to switch between models like Claude, OpenAI, or local options seamlessly. Users can also add over 50 integrations (Gmail, GitHub, Spotify, Home Assistant) or train it to perform new tasks—sometimes simply by requesting it.

Some critics have remarked that it serves merely as a “wrapper with browser control.” While that may hold some truth, no other tool has accomplished this level of proactive and cross-channel execution in a self-hosted, free format. A few setup hiccups exist (such as Claude token peculiarities and security compromises related to system access), but the sandboxing and local-first design have kept many advanced users engaged.

Clawdbot is not your well-mannered Siri awaiting instructions; it is a dynamic automation tool devoid of guardrails. It is advisable to test it with caution, ideally on a dedicated machine!

→ clawd.bot

→ github.com/clawdbot/clawdbot

Anthropic is gearing up to introduce the Security Center for Claude Code. This centralized dashboard will streamline the process of scanning, flagging issues across repositories, and allowing developers to conduct manual checks by repository or branch.

Previously named AutoPatch, this feature will integrate directly into Claude Code, enhancing security workflows for teams and individuals alike. The rollout coincides with OpenAI’s enhancements to Codex as they reach “Cybersecurity High,” although a launch date remains to be announced.

Claude’s Excel integration, which was previously in beta, has now become accessible to all Pro subscribers, embedding AI assistance into everyday tasks.

The Claude CoWork feature has significantly increased attention towards Claude Code. According to Similarweb, there has been notable growth in searches, website traffic, app downloads, and daily active users in the last month, a surge likely attributed to the upcoming Security Center and expanded integrations. This indicates a growing interest in Anthropic’s AI solutions among developers managing complex, security-focused code. Read more.