I gave up using OpenClaw

70 will be done. OpenClaw takes a few hundred lines of seven files, the same effect。

Claude Code automatically reads this file every time he starts a new conversation, which is like AI always knows who you are, who you are, what to do. You can also change styles, change roles, add rules and enter into force after the next dialogue。

Recapturing "inter-dialogue memory"

OpenClaw has three layers of memory: long notes, daily logs, full text searches. With Claude. What about the mark

• Long-term memory: straight to Claude. RememberIn Claude.ai, you're going to say, "Remember that I'm doing xx making intern," "Remember that I prefer Chinese" and "Remember that my writing style is simple and has a point of view" and it's going to be permanently stored, and every new conversation is automatically brought in. You can also say "forget XX" deleted at any time, go to Settings and see everything it remembers. And when you open the "Generate memory from chat history" from Settings, Claude automatically draws the focus from your usual chat: your role, project, habits, you don't have to say it's sorted. The biggest difference with OpenClaw is that Claude's memory is independent and not lost when the conversation is too long. OpenClaw's problem is that half of the memory is condensed and deformed, Claude。

  If you use Claude Code, it also has an automatic memory, and every time you run out of it, you automatically write down what you learned (your habits, project structure, pedals), it's local, and the next time you start, it's automatically loaded, showing "Recalled X memories." You don't have to do anything. Remember to accumulate。

• Dialogue history: If you can find it, you can't forget itClaude can search all your historical conversations. Just ask, "What did we talk about last week," "Find out the conversation about the tokenomics analysis," and it will find a direct reference to the content, without you defending anything。

Progress: using Obsidian as a memory center

It's about Claude's built-in memory, enough for most people. But if you have as much notes, project information, research material as I do, and want AI to really understand your entire knowledge base, there's a stronger approach: connect Obsidian。

https://x.com/obsdmd/status/20374633568638245?s=20

Obsidian 1.12 just released the official CLI -- you can run your notebooks directly in the terminal: search, read and write notes, check labels, check reverse links, anything. The point is that Claude Code is in the terminal. Two times, your entire Obsidian library becomes AI's memory。

Set method:

1. Update Obsidian to 1.12 and above

2. Settings, General, Open CLI

3. YoursCLUDE.mdRiga:

Done. After you talk to Claude Code, it'll go straight to your Obsidian to find the answer, and you don't have to post the notes。

What is better than OpenClaw's memory system? OpenClaw's memory is a bunch of markdown text files. Searching is a blurry match. Obsidian has its own search engine -- labels, reverse links, full text searches -- and you're using it to manage knowledge without having to maintain an extra set of things. It has been measured, as well, to find a note that would burn 7 million tokens with original file scans, 100 with Obsidian CLIs。

If you don't want to use CLI, you can also install MCP plugins to connect Claude directly to Obsidian. The community already has an off-the-shelf program to put on Claude Desktop and Claude Code to read and write your notebook。

SCRIPT "A DIALOGUE WITH AI ON THE PHONE"

OpenClaw is the best thing about talking directly to AI in Telegram, which can do anything for you. Claude has two ways:

• Mode I: Remote Control (official programme)Claude Code recently released/remote-controlFunction: Open a session on the computer and connect it with Claude App on the phone. Typing, talking, computer Claude Code execution. Messages are almost seconds of synchronization, and the notebook covers wake up and reconnect automatically. The good thing is to completely walk the Anthropic official passage, secure, without having to build anything. The disadvantage is that your computer has to be on。

• Mode two: Telegram Bot (community programme)If you're going to use it at Telegram, there's an open source solution available:

  1. Loadthis post is part of our special coverage syria protests 2011:uv tool installget+https://github.com/ RichardAtCT/claude-code-telegram

  2. Telegram @BotFather

  3. Set your bot token, Telegram ID, directory

  4. Starts, in Telegram, directly with Claude Supports voice, pictures, documents, each item's conversation record is kept。

I'll write "doing your job automatically."

OpenClaw automatically wakes up every 30 minutes to see if there's anything to do. Claude 's Cowork function can do similar things: set a schedule/schedule) make Claude run at the specified time。

My set:

• Every morning at 9 a.m.: Monitor DeFi bid, grab the 24-hour high-level interactive tweet

• monday 10 o'clock: generate last week's marketing performance summary

• Every day at 6:00 p.m.: Collapse important news today

Like OpenClaw's cron, Cowork is the time you set it up to do it, not AI's own judgment. But OpenClaw's heartbeat automatically checked 90% of the time to see nothing done and burn money for nothing. Why don't we just set when to do something。

Cost

• OpenClaw:API, AT A CHARGE, I'LL BE $150 + A MONTH. YOU DON'T KNOW HOW MUCH NEXT MONTH

• Claude Pro$20/month. For most people

• Claude Max$100-200/month. Heavy user, choose this with Claude Code

Monthly fee, billing is predictable。

Finally

OpenClaw's brilliant idea: Everyone should have a strong AI assistant。

But at this stage it is a cool but untrustworthy experiment。

And Claude's ecosystem, CLUDE.md, gives it personality, Memory, makes it remember you, Obsidian, when it's knowledge bank, Remote Control, makes you use your phone to manipulate it, and Cowork to make it work on time. Together, they can do 80% of OpenClaw, and they're more stable, safer, cheaper。

INSTEAD OF SPENDING TIME FIXING YOUR AI ASSISTANT, LET THE AI ASSISTANT DO YOUR JOB。

That's why I turned OpenClaw off。