How to Build an Autonomous AI Assistant with OpenClaw on a MacBook Air

Overview

The creator walks through a fully automated AI ecosystem built on a fresh MacBook Air running OpenClaw 24/7. The system integrates multiple communication channels, AI models, databases, and external services to act as a personal CRM, knowledge base, video‑idea generator, business analyst, and more.

Hardware & Basic Setup

• MacBook Air in clamshell mode, always on, connected to the internet.
• Remote access via TeamViewer and Tailscale (SSH).
• Cron jobs keep the machine running tasks every hour or day.

Communication Interfaces

• Telegram: primary chat interface, organized into narrow topics (knowledge base, food journal, cron updates, video research, etc.). Sessions are set to expire after one year to keep context.
• Slack: limited to two private channels; only the owner can invoke OpenClaw.
• Command‑line / SSH: for direct scripting and debugging.

AI Models & Skills

• Anthropic (Opus, Sonnet, Haiku), Google Gemini, XAI Gro, XARCH, OpenAI.
• Skills include personal CRM, knowledge base, video‑idea pipeline, X/Twitter research, business meta‑analysis, HubSpot ops, humanizer (removes AI‑style phrasing), task management, YouTube analytics, etc.

Personal CRM Workflow

1. Daily cron downloads Gmail and calendar.
2. Parses senders/participants, de‑duplicates contacts, classifies roles using Gemini 2.5 Flash.
3. Updates a graph‑based database with conversation history.
4. Provides instant answers via Telegram (e.g., “Who was the last person I talked to at Grapile?”).
5. Generates meeting‑prep briefs each morning with last touch, agenda, and contact info.

Knowledge Base

• Files or URLs dropped in Telegram are automatically:
• Detected, extracted, normalized, chunked.
• Stored in a hybrid SQL + vector database.
• Natural‑language queries retrieve relevant articles with source citations.

Video Idea Pipeline

1. Article/link added via Telegram or Slack triggers an idea event.
2. System parses the topic, researches X and the web, queries the knowledge base for related content.
3. Generates video pitches, hooks, outlines, and creates an ASA task.
4. All steps complete in ~30 seconds.

Twitter Search Workflow

• Tiered fallback chain to minimize cost:
• FX Twitter API (free, single‑tweet lookup).
• twitterapi.io (≈ $0.15 / 1k tweets, supports search & thread context).
• Official X API v2 (expensive but full‑featured).
• XAI/XARCH and Grock as additional fallbacks.
• Chooses the cheapest tier that satisfies the request.

YouTube Analytics & Competitor Tracking

• Daily API pull of channel stats, video performance, and competitor uploads.
• Stores snapshots locally, computes trends, generates PNG charts, feeds insights into the business‑analysis workflow.

Business Council AI

• Aggregates ~200 high‑confidence signals (YouTube, CRM, HubSpot, Slack, meetings via Fathom, etc.).
• A council of AI agents (growth strategist, revenue guardian, skeptical operator, team‑dynamics architect) debates and produces a daily improvement report.

HubSpot Integration

• OpenClaw can query the deal pipeline via natural language (e.g., “What deals are in qualification?”).
• Primarily used for sponsorship tracking.

Humanizer Skill

• Runs on every generated text to strip AI‑style patterns (no dashes, more natural phrasing).
• Available as a reusable skill on Clawhub.

Image & Video Generation

• Integrated Nano Banana (image) and VO (video) APIs.
• Separate Telegram topics for images and videos.
• Iterative prompt‑refinement until the desired output is produced, then sent back to Telegram.

To‑Do List Management

• Meeting transcripts from Fathom are sent to Gemini 2.5 Flash to extract actionable items.
• Tasks are cross‑referenced with the CRM, displayed for approval, and pushed to To‑Doist.

Cost Monitoring

• Background logger records every AI and API call.
• Telegram queries can retrieve weekly spend, cost breakdown per workflow, and 30‑day trends.
• Monthly budget is roughly $150, considered a high ROI.

Backup & Recovery

• Code and markdown files versioned in GitHub.
• Databases backed up hourly to Google Drive with timestamps.
• Detailed restoration guide stored alongside the repo.

Memory Management

• Daily notes captured from conversations and tasks.
• Weekly synthesis distills patterns into long‑term memory.
• Learnings folder stores corrective patterns for continuous improvement.

Development Workflow

• Primary development done in Cursor (IDE) via SSH to the MacBook Air.
• TeamViewer used for full remote control when needed.
• Frequent commits, tests, and pushes to GitHub.

Markdown File Maintenance

• workspace.md holds the architecture overview and table of contents.
• OpenClaw periodically downloads best‑practice guides (OpenClaw website, Anthropic Opus prompting guide) and cross‑checks all markdown files, suggesting updates to stay aligned with model‑specific prompting rules.

Final Thoughts

The setup demonstrates how a single low‑cost laptop can become a 24/7 AI‑powered personal assistant, handling everything from contact management to content creation, business analysis, and cost control. The modular, reusable “skills” architecture makes it easy to extend the system for new use cases.

With a modest MacBook Air and a well‑orchestrated mix of AI models, automation scripts, and cloud services, you can build a self‑sustaining AI assistant that handles CRM, knowledge management, content creation, and business analytics—all while keeping costs transparent and data safely backed up.