NemoClaw Knowledge Wiki

Use clawdcode Matthew Be

4 min read

https://www.youtube.com/watch?v=MUDvwqJWWIw

Here is a summary of the video on Clawdbot, the open-source personal AI assistant.

Clawdbot: The “Real” Siri We Were Promised

Clawdbot is an open-source, self-hosted personal AI assistant that runs locally on your computer but is accessible via standard chat applications. It is described as “what Siri should have been,” offering full computer access, persistent memory, and agentic capabilities.

🔑 Key Features

  • Runs Locally: You host it on your own machine (Mac, Windows, or Linux). It is fully open-source.
  • Chat Anywhere: Connects to apps you already use like Telegram, WhatsApp, Slack, Discord, and iMessage. You can chat with your computer remotely from your phone.
  • Full Computer Access: Unlike standard chatbots, Clawdbot can read files, write code, execute terminal commands, and browse the web via Chrome extensions.
  • Persistent Memory: It learns your preferences, schedule, and work habits over time.
  • Proactive Agents: It can set up cron jobs to run tasks in the background (e.g., checking emails every 10 minutes) without you initiating the chat.
  • Customizable Personality: Uses a soul.md file where you can define the bot’s “vibe,” boundaries, and core truths (e.g., “be concise,” “have opinions”).

🛠️ Integrations & Setup

🚀 Real-World Use Cases Demonstrated

The video showcased several powerful examples of Clawdbot’s agentic capabilities:

  1. YouTube Research: Connected to Grok (for real-time X/Twitter access) and Asana to research, outline, and populate project management tasks for the video itself.
  2. File Management & Repair: The user had a failed bulk upload to Google Drive. Clawdbot:
    • Scanned the local folder and the Drive folder.
    • Identified the 212 missing files.
    • Wrote code to upload the missing files one by one (handling Google API rate limits automatically).
  3. Proactive Email Assistant: Created a cron job to check 3 different inboxes every 5 minutes, identify only urgent emails based on user-defined criteria, draft replies, and send a summary to Telegram for approval.
  4. Social Media Management: Monitored Twitter replies, drafted responses in the user’s voice, and waited for user approval to publish.

💻 Local Models & Cost Management

A major issue with agentic AI is cost (the creator spent $130 in one day using Claude Opus). To solve this:

  • LM Studio Integration: Clawdbot can download and run local models (like Qwen 3).

ee0e5799-ec7d-46e2-a517-37b1726cec10

⚠️ Risks & Limitations

  • Security: You are giving an AI access to your credentials, files, and terminal. It is a non-deterministic system, meaning it can make mistakes or delete things unintentionally.
  • Memory Issues: Like all LLMs, “memory compaction” (compressing history to fit context windows) can cause it to lose specific details over time.
  • Stability: The project is only ~2 months old. It creates bugs, gets stuck in loops, or crashes, requiring manual restarts.
  • Cost: Without careful management of which models are used, API costs can skyrocket due to the sheer volume of tokens used for background processing.

🏁 Conclusion

Clawdbot represents the future of AI assistants—moving away from chat windows and into full OS integration. While it is currently for “power users” due to security risks and setup requirements, it offers a glimpse into a future where AI handles tedious digital chores autonomously.

Graph View