Author: Jooooooe | NodeZ

I started playing with it at the beginning of the year when it was called Clawdbot, and it went through several name changes, including Moltbot and then OpenClaw. For more than a month, it ran 24 hours a day on my Mac Mini, connecting Telegram, email, and calendar, trying to create an AI assistant that could do everything.

I turned it off last week. Not because it wasn't cool, but because "cool" and "usable" are two different things.

Why give up?

• Memory: OpenClaw claims it can remember what you've said and automatically bring it up in the next chat. But after using it for a while, you'll discover something really stupid: it will still forget what you said.

• Burning money: OpenClaw automatically wakes up every 30 minutes to check if there's anything to do. Each check costs money in API fees—reading emails costs money, checking the calendar costs money, and checking Telegram costs money. A simple task often requires multiple calls to the AI ​​to complete. My monthly bill exceeded $150, more expensive than a direct subscription to Claude Max.

• Security issues: Kaspersky discovered 512 vulnerabilities. Researchers found over 40,000 OpenClaw applications with their API keys and chat logs completely exposed online. Cisco tested a community plugin and found it secretly transmitting data to external servers without the user's knowledge.

• You probably don't need AI 24/7 at all. That's something I only realized recently. All the truly valuable work is what I do proactively. The robots running in the background mostly do trivial tasks; there's no need to use AI for those.

Breaking it down, OpenClaw actually does two things.

Stripped of all the fancy features, the core of OpenClaw is:

• Remember you—know who you are, what you do, and what you like.

• It takes the initiative to handle things for you without you having to say a word.

Once this is understood, the question becomes: Can Claude do it?

Recreate "Remember You": Create Your SOUL.md

Many people who use OpenClaw are most excited about writing SOUL.md, a manual that tells the AI ​​"who you are." OpenClaw breaks this down into seven or eight files: SOUL.md defines personality, IDENTITY.md defines identity, USER.md describes who you are, AGENTS.md writes behavioral rules, MEMORY.md stores long-term memory, and there are a bunch of skill files. At the start of each conversation, it reads all of them in order, so the AI ​​"knows" who it is, who its owner is, and how to act.

In Claude Code, all of these are combined into one file: CLAUDE.md.

Create a CLAUDE.md file in your project folder and clearly write the following:

#JARVIS — My Personal AI Assistant

## identity

You are JARVIS, my personal AI assistant. You speak concisely and with clear opinions.

Reply in Chinese, and you may include English terms if necessary.

## About Me

My name is Joe, I'm 21 years old, and I studied International Business at National Chengchi University.

- Working as a Marketing Intern in xx

- Also serves as the Taiwan Business Development (BD) for xx, focusing on the xx ecosystem.

- Co-founded @Node_Z_

- I usually need to write tweets and marketing copy.

## Rules of Conduct

Answer directly, don't be long-winded.

- When writing marketing copy, adopt the tone and style of a KOL (Key Opinion Leader), don't sound robotic.

- When dealing with crypto projects, check the latest information before answering.

- If something is uncertain, say it's uncertain; don't make things up.

## Memory Management

- Actively write important things into memory

- What I said "remember" you must remember.

- Before each conversation ends, save the key points that are worth remembering.

It only took 70 lines. OpenClaw requires seven files and hundreds of lines, but the effect is the same.

Claude Code automatically reads this file every time a new conversation begins, meaning the AI ​​always knows who you are, who you are, and how to act. You can also change the style, switch roles, and add rules at any time, and the changes will take effect in the next conversation.

Replicating "cross-dialogue memory"

OpenClaw's memory is divided into three layers: long-term notes, daily logs, and full-text search.

How do I benchmark it using Claude?

Long-term memory: Just call him Claude and remember him.

In Claude.ai (web/mobile version), you can speak directly in the conversation:

• "Remember that I worked as a marketing intern at xx."

• "Remember I prefer Simplified Chinese."

• "Remember my writing style: concise and to the point."

It will be permanently saved and automatically included in every new conversation. You can also delete it at any time by saying "Forgot XX". You can see everything it has remembered in Settings.

Additionally, by enabling "Generate memory from chat history" in Settings → Capabilities, Claude will automatically summarize key points from your everyday chats: your role, projects, and habits, without you having to actively mention them; it will organize them on its own.

The biggest difference between Claude and OpenClaw is that Claude's memories are stored independently and won't be lost when the conversation is compressed due to its length. OpenClaw's biggest problem is that the memories from earlier parts of the conversation are compressed and distorted halfway through, but Claude doesn't have this problem.

If you use Claude Code, it also has an automatic memory feature. Every time you use it, it automatically records what you've learned (your habits, project structure, pitfalls you've encountered), stores it locally, and automatically loads it the next time you start working, displaying "Recalled X memories". You don't have to do anything; the memories accumulate on their own.

Dialogue with History: If you can find it, you won't forget it.

Claude can search all your historical conversations. Just ask "What did we talk about last week?" or "Find that conversation from the previous tokenomics analysis," and it will find the relevant content and quote it directly, without you having to maintain anything yourself.

Advanced: Using Obsidian as a memory center

The above describes Claude's built-in memory, which is sufficient for most people. However, if you, like me, have a large amount of notes, project materials, and research resources, and want AI to truly understand your entire knowledge base, there is a more powerful approach: connect it to Obsidian.

https://x.com/obsdmd/status/2027416335689638245?s=20

Obsidian 1.12 just released its official CLI—allowing you to directly manipulate your note library from the terminal: search, read and write notes, check tags, view backlinks, and more. Importantly, Claude Code also runs in the terminal. Connecting the two transforms your entire Obsidian note library into AI-generated memory.

Setup method:

• Update Obsidian to version 1.12 or later.

• Settings → General → Open CLI

• Add the following to your CLAUDE.md file:

## Obsidian Integration

- Use `obsidian search` to search notes instead of manually flipping through files.

- Use `obsidian files read` to read notes.

- Use `obsidian files append` to append content to notes.

- Use `obsidian daily` to manage daily notes

- When searching for information, prioritize searching for Obsidian; only search the internet if you can't find it there.

Done. From now on, when you talk to Claude Code, it will directly search your Obsidian for the answer, without you having to manually paste notes.

How is this superior to OpenClaw's memory system? OpenClaw's memory is just a bunch of Markdown text files; its search is fuzzy matching, and it becomes inaccurate with a large number of notes. Obsidian has its own search engine—tag, backlink, and full-text search all have sub-second response times, and since you're already using it to manage knowledge, you don't need to maintain an additional system. Someone has tested it: to find the same note, scanning with the original file requires 7 million tokens, while using the Obsidian CLI only requires 100.

If you don't want to use the CLI method, you can also install the MCP plugin to allow Claude to connect directly to Obsidian. There are already ready-made solutions in the community; once installed, both Claude Desktop and Claude Code can read and write your note library.

Replicating the experience of "talking to AI on your phone"

The best thing about OpenClaw is being able to talk directly to the AI ​​on Telegram; it can do anything for you. Claude does this in two ways:

Method 1: Remote Control (Official Solution)

Claude Code recently introduced a /remote-control feature: create a session on your computer and connect using the Claude app on your phone. Type or speak on your phone, and Claude Code on your computer executes the commands. Messages sync almost instantly, and the laptop will automatically reconnect when you close the lid and wake up.

The advantage is that it uses the official Anthropic channel, which is secure and doesn't require you to set up anything yourself. The disadvantage is that your computer needs to be on.

Method 2: Telegram Bot (Community Solution)

If you intend to use it on Telegram, there are readily available open-source solutions:

• Install claude-code-telegram: uv tool install git+ https://github.com/RichardAtCT/claude-code-telegram

• Telegram @BotFather Create Bot Get Tokens

• Set up your bot token, Telegram ID, and working directory.

• Start by talking to Claude directly on Telegram.

It supports voice, images, and files, and retains the conversation history for each project.

Replicating the "automatic tasks for you" feature.

OpenClaw automatically wakes up every 30 minutes to check if there's anything to do. Claude's Cowork feature can do something similar: set a schedule (/schedule) so that Claude runs automatically at a specified time.

My settings:

• Every morning at 9 AM: Monitor DeFi competitor accounts and capture highly interactive tweets from the past 24 hours.

• Every Monday at 10 AM: Generate a summary of the previous week's marketing performance.

• Every day at 6 PM: Organize the day's important news.

Like OpenClaw's cron, Cowork requires you to set a time for it to perform actions, rather than relying on AI for automatic decisions. However, OpenClaw's automatic heartbeat check does nothing 90% of the time, wasting money. It's better to directly set when to do what.

cost

• OpenClaw: API is billed on a pay-as-you-go basis. My monthly cost is $150+, you don't know how much it will be next month.

• Claude Pro: $20/month. Enough for most people.

• Claude Max: $100-200/month. Choose this if you're a heavy user or need to use Claude Code.

Monthly fee system, bills are predictable.

at last

OpenClaw's idea is awesome: everyone should have a powerful AI assistant.

But at this stage, it's a cool but untrustworthy experiment.

Claude's ecosystem—CLAUDE.md gives it a personality, Memory lets it remember you, Obsidian serves as its knowledge base, Remote Control allows you to control it with your phone, and Cowork lets it perform tasks on a schedule—combined together, it can achieve 80-90% of what OpenClaw does, and it's more stable, more secure, and cheaper.

Instead of spending time fixing your AI assistant, let it do the important things for you.

That's why I turned off OpenClaw.