Descriptions:
Nate Herk walks through how he built a fully autonomous AI executive assistant named Klouse using Claude (via the Moltbot open-source framework) running 24/7 on a server. The flagship demo shows Klouse building a complete YouTube analytics dashboard overnight while Herk slept — the agent gave itself six follow-on tasks after initial setup and logged activity continuously from midnight until 7 AM, tracking 7-day and 30-day stats, comment sentiment, competitor video breakouts, and AI-generated video title ideas.
Herk explains the underlying architecture in detail: a soul.md file defines the assistant’s identity and behavioral guidelines, while a user.md file captures personal and business context gathered through a structured onboarding interview. Klouse connects to GitHub, Brave Search, OpenAI, ElevenLabs, Apify, ClickUp, Google Workspace, and YouTube APIs. Running exclusively on Anthropic’s API with Opus 4.5, Herk reports spending approximately $223 over three days for roughly 250 million tokens — and notes that switching to Sonnet 4.5 or a locally hosted model would substantially reduce costs. He also warns that using a Claude Max subscription plan for Moltbot violates Anthropic’s terms of service, requiring an API key instead.
The video shares five practical lessons for building a smarter assistant, including creating dedicated accounts for the AI to contain security risk, prompting the agent to interview you before configuration, and using Claude as an active teacher — asking it to explain its own reasoning — rather than treating it purely as a task executor.
📺 Source: Nate Herk · Published January 30, 2026
🏷️ Format: Hands On Build
