Let an agent build and run its own online business
How Nat set up his OpenClaw bot to run its $4,000/week business — including a 3-layer memory system, multi-threaded chats, and security best practices. Task: How Nat set up his OpenClaw bot to run its $4,000/week business — including a 3-layer memory system, multi-threaded chats, and security best practices. Dear subscribers, Today, I want to share a new episode with my friend, Nat Eliason. Nat gave his OpenClaw bot, Felix, $1,000 to build its own business 3 weeks ago. Since then, it has made $14,718 by launching its own website, info product, and X account. Nat walked through exactly how he set up his OpenClaw bot to run its own business — from the 3-layer memory system, to multi-threaded chats, to security best practices. Watch now on YouTube, Apple, and Spotify. Nat and I talked about: (00:00) Meet Felix: The OpenClaw bot building its own business (00:00) Meet Felix: The OpenClaw bot building its own business (03:49) “I’m going to sleep. Build a product that makes money.” (03:49) “I’m going to sleep. Build a product that makes money.” (08:03) How to set up multiple OpenClaw chats to build 5 projects at once (08:03) How to set up multiple OpenClaw chats to build 5 projects at once (11:06) How Felix is able to ignore prompt injections on X/Twitter (11:06) How Felix is able to ignore prompt injections on X/Twitter (14:42) The wild story of how Felix ended up with $100K+ in crypto (14:42) The wild story of how Felix ended up with $100K+ in crypto (17:24) The 3-layer memory system that makes it all work (17:24) The 3-layer memory system that makes it all work (22:14) Heartbeat, cron jobs, and delegating to Codex (22:14) Heartbeat, cron jobs, and delegating to Codex (26:41) Ask this question to make your OpenClaw bot more capable (26:41) Ask this question to make your OpenClaw bot more capable (32:14) Recap: how to set up your own bot to build a business (32:14) Recap: how to set up your own bot to build a business I’m proud to partner with Replit Replit now makes beautiful videos for product launches, fundraise announcements, and explainers from simple text prompts. To try it out: Go to Replit and select the “App” tab Go to Replit and select the “App” tab Ask for “an animated video about…”, or use the dropdown selector under the prompt box & select “Animation” Ask for “an animated video about…”, or use the dropdown selector under the prompt box & select “Animation” It’s as simple as that. Try Replit Animation now for free via the link below. Try Replit for Free Top 10 takeaways I learned from this episode Set up a 3-layer memory system for OpenClaw first. This is the single biggest unlock. From Nat: “Get the memory structure in first because then your conversations from day one will be useful.” Here’s what Felix uses:Layer 1: Knowledge graph. A ~/life/ folder using the PARA system (projects, areas, resources, archives). Stores durable facts about people and projects with summary files for quick lookups.Layer 2: Daily notes. dated markdown file for each day logging what happened. Your bot writes to this during conversations, then extracts important stuff into Layer 1 during nightly consolidation.Layer 3: Tacit knowledge. Facts about you like your communication preferences, workflow habits, hard rules, and lessons learned from past mistakes. This is what makes your bot feel like it actually knows you. Set up a 3-layer memory system for OpenClaw first. This is the single biggest unlock. From Nat: “Get the memory structure in first because then your conversations from day one will be useful.” Here’s what Felix uses: Layer 1: Knowledge graph. A ~/life/ folder using the PARA system (projects, areas, resources, archives). Stores durable facts about people and projects with summary files for quick lookups. Layer 1: Knowledge graph. A ~/life/ folder using the PARA system (projects, areas, resources, archives). Stores durable facts about people and projects with summary files for quick lookups. Layer 2: Daily notes. dated markdown file for each day logging what happened. Your bot writes to this during conversations, then extracts important stuff into Layer 1 during nightly consolidation. Layer 2: Daily notes. dated markdown file for each day logging what happened. Your bot writes to this during conversations, then extracts important stuff into Layer 1 during nightly consolidation. Layer 3: Tacit knowledge. Facts about you like your communication preferences, workflow habits, hard rules, and lessons learned from past mistakes. This is what makes your bot feel like it actually knows you.; tools: OpenClaw. Human role: The original post does not say how a person stayed involved.; limitations: The source is a paid podcast summary and tutorial, indicating potential promotional bias or lack of rigorous verification. The $100K crypto claim lacks context on how it was earned or held. The workflow requires specific OpenClaw configurations and a 6,500+ character prompt for memory setup. Human approval processes for autonomous spending or publishing are not explicitly defined, posing operational risks..
Tools
Published on Agent Work Atlas:
Overview
How Nat set up his OpenClaw bot to run its $4,000/week business — including a 3-layer memory system, multi-threaded chats, and security best practices.
Workflow
How Nat set up his OpenClaw bot to run its $4,000/week business — including a 3-layer memory system, multi-threaded chats, and security best practices. Dear subscribers, Today, I want to share a new episode with my friend, Nat Eliason. Nat gave his OpenClaw bot, Felix, $1,000 to build its own business 3 weeks ago. Since then, it has made $14,718 by launching its own website, info product, and X account. Nat walked through exactly how he set up his OpenClaw bot to run its own business — from the 3-layer memory system, to multi-threaded chats, to security best practices. Watch now on YouTube, Apple, and Spotify. Nat and I talked about: (00:00) Meet Felix: The OpenClaw bot building its own business (00:00) Meet Felix: The OpenClaw bot building its own business (03:49) “I’m going to sleep. Build a product that makes money.” (03:49) “I’m going to sleep. Build a product that makes money.” (08:03) How to set up multiple OpenClaw chats to build 5 projects at once (08:03) How to set up multiple OpenClaw chats to build 5 projects at once (11:06) How Felix is able to ignore prompt injections on X/Twitter (11:06) How Felix is able to ignore prompt injections on X/Twitter (14:42) The wild story of how Felix ended up with $100K+ in crypto (14:42) The wild story of how Felix ended up with $100K+ in crypto (17:24) The 3-layer memory system that makes it all work (17:24) The 3-layer memory system that makes it all work (22:14) Heartbeat, cron jobs, and delegating to Codex (22:14) Heartbeat, cron jobs, and delegating to Codex (26:41) Ask this question to make your OpenClaw bot more capable (26:41) Ask this question to make your OpenClaw bot more capable (32:14) Recap: how to set up your own bot to build a business (32:14) Recap: how to set up your own bot to build a business I’m proud to partner with Replit Replit now makes beautiful videos for product launches, fundraise announcements, and explainers from simple text prompts. To try it out: Go to Replit and select the “App” tab Go to Replit and select the “App” tab Ask for “an animated video about…”, or use the dropdown selector under the prompt box & select “Animation” Ask for “an animated video about…”, or use the dropdown selector under the prompt box & select “Animation” It’s as simple as that. Try Replit Animation now for free via the link below. Try Replit for Free Top 10 takeaways I learned from this episode Set up a 3-layer memory system for OpenClaw first. This is the single biggest unlock. From Nat: “Get the memory structure in first because then your conversations from day one will be useful.” Here’s what Felix uses:Layer 1: Knowledge graph. A ~/life/ folder using the PARA system (projects, areas, resources, archives). Stores durable facts about people and projects with summary files for quick lookups.Layer 2: Daily notes. dated markdown file for each day logging what happened. Your bot writes to this during conversations, then extracts important stuff into Layer 1 during nightly consolidation.Layer 3: Tacit knowledge. Facts about you like your communication preferences, workflow habits, hard rules, and lessons learned from past mistakes. This is what makes your bot feel like it actually knows you. Set up a 3-layer memory system for OpenClaw first. This is the single biggest unlock. From Nat: “Get the memory structure in first because then your conversations from day one will be useful.” Here’s what Felix uses: Layer 1: Knowledge graph. A ~/life/ folder using the PARA system (projects, areas, resources, archives). Stores durable facts about people and projects with summary files for quick lookups. Layer 1: Knowledge graph. A ~/life/ folder using the PARA system (projects, areas, resources, archives). Stores durable facts about people and projects with summary files for quick lookups. Layer 2: Daily notes. dated markdown file for each day logging what happened. Your bot writes to this during conversations, then extracts important stuff into Layer 1 during nightly consolidation. Layer 2: Daily notes. dated markdown file for each day logging what happened. Your bot writes to this during conversations, then extracts important stuff into Layer 1 during nightly consolidation. Layer 3: Tacit knowledge. Facts about you like your communication preferences, workflow habits, hard rules, and lessons learned from past mistakes. This is what makes your bot feel like it actually knows you.
Reported result
The source claims Felix generated $14,718 in revenue over three weeks, with $3,500 in the first week, attributed to Nat Eliason's setup. It also mentions the bot accumulated over $100K in crypto, though the mechanism for this is described as a 'wild story' without detailed verification. No independent audit or failure metrics are provided. A measurable acceptance check would be verifying Stripe dashboard revenue against the claimed figures for a controlled period.
Human role
The original post does not say how a person stayed involved.
Limitations
The source is a paid podcast summary and tutorial, indicating potential promotional bias or lack of rigorous verification. The $100K crypto claim lacks context on how it was earned or held. The workflow requires specific OpenClaw configurations and a 6,500+ character prompt for memory setup. Human approval processes for autonomous spending or publishing are not explicitly defined, posing operational risks.
Try it
I want to replicate Nat Eliason's OpenClaw bot setup for an autonomous business. First, configure a 3-layer memory system in OpenClaw: Layer 1 as a ~/life/ folder using PARA for durable facts, Layer 2 for daily markdown logs, and Layer 3 for tacit knowledge like my preferences. Use the provided 6,500-character prompt to initialize this structure. Set up multi-threaded chats to manage [replaceable: 3-5 specific projects, e.g., blog, newsletter, product]. Implement security best practices to ignore prompt injections on [replaceable: social platform, e.g., X/Twitter]. Configure heartbeat and cron jobs for automation. Before any autonomous action like publishing content or spending funds, require my explicit confirmation. Start with a dry run: have the bot draft one [replaceable: content piece or product outline] and show me the output for approval. Do not execute any real transactions or public posts until I verify the memory system is working correctly and the output meets my standards. Ask clarifying questions if the initial budget or specific business niche is unclear.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.