Have you ever experienced this when using AI to write code?

You’re working along nicely, and then the AI suddenly says, “Sorry, I don’t remember what we discussed before.” Didn’t you just want to throw your computer out the window at that moment?

Don’t rush—it’s not that the AI is trying to annoy you. It genuinely doesn’t remember.

Goldfish Memory: AI Programming’s Open Secret

AI programming tools are exploding in popularity—Codex, Claude Code, Gemini CLI… they all claim to help you write code. But they all share a common weakness: their memory is too short.

Think of it this way: they have about as much memory as a fishbowl. You finish discussing a requirement with them, close the session, open it again next time—to the AI, you’re strangers. It completely forgets the requirements you discussed, the pitfalls you encountered, and the decisions you made.

How painful is this? Let me walk you through a few scenarios:

Scenario One: Branch Switching Loses Your Progress You’re using AI to help fix a bug, halfway through, when you need to switch to another branch to fix another urgent issue. When you come back, the AI doesn’t recognize you anymore. You have to explain everything again—which solutions you tried, which ones you rejected…

Scenario Two: Picking Up Where You Left Off Yesterday You spent half a day yesterday writing code with AI, and today you open it to continue. The problem is, the AI has no idea what you were working on. You have to explain the whole context again, and sometimes you even forget where you left off yourself.

Scenario Three: Cross-File Context Loss You asked AI to handle logic connections across three or four files simultaneously. After it finishes one and turns to ask “What did that file look like again?"—it forgot.

These aren’t bugs—they’re architectural limitations. AI models have context window constraints, and each session needs to rebuild context from scratch. It’s like hiring an assistant with a 7-second memory—brilliant, but they can’t remember anything.

So Is There a Solution?

There wasn’t a good one before. Now there is.

A tool called RexCLI (also known as AIOS) adds a “memory upgrade layer” to these AI programming assistants. It doesn’t require you to change any code or learn new commands—it just silently remembers things for you in the background.

What does it remember?

  • What project you’re currently working on
  • How far along your previous tasks have progressed
  • What requirements were discussed in previous sessions
  • Which solutions were tried and which were rejected

The next time you open AI, it automatically loads the previous work state. It’s like telling a colleague “continue with that requirement from yesterday”—they can pick up right where you left off without you having to explain everything again.

How Does It Work?

The principle is actually straightforward. It creates an invisible “work log” in your project directory. Every time you finish chatting with AI, it automatically records: what stage the current task is at, what needs to be done next, and anything special to watch out for.

The next time you start AI, this tool—before AI even says anything—feeds the previous context to it. AI sees it, understands the situation, and continues working.

The whole process is automatic. You don’t need to manually save anything, don’t need to write documentation, and don’t even need to change your habits. After installing the tool, it just runs quietly in the background.

Does It Actually Work?

I’ve been using it for a while, and two things stand out:

First, no more repeating explanations. Previously, every time I opened AI, I had to explain the background first. Now I don’t need to—it already knows what I’m working on.

Second, I dare let AI handle more complex tasks. Before, I was hesitant to let AI handle multiple files simultaneously, because I was worried it would lose track of the relationships between them. Now with this memory layer, AI can handle longer task chains.

Of course, it’s not a magic solution. AI still makes the mistakes it always makes, and the model’s limitations remain. But this memory capability does solve the biggest pain point—“not remembering.”

How to Try It?

Installation is simple—one line of code:

curl -fsSL https://github.com/rexleimo/rex-cli/releases/latest/download/aios-install.sh | bash

Or if you’re on Windows:

irm https://github.com/rexleimo/rex-cli/releases/latest/download/aios-install.ps1 | iex

After installation, you can keep using Codex, Claude Code, or whichever AI tool you prefer—no habit changes needed. The tool automatically adds memory capabilities to them.

For more details, check out: https://cli.rexai.top

In simple terms: AI used to be a goldfish, now it becomes an assistant that holds grudges. Worth trying, right?

FAQ

Q: Which AI programming tools does RexCLI support? A: Currently, it supports Codex CLI, Claude Code, and Gemini CLI—the three main AI programming tools. After installation, it automatically adds memory capabilities to all of them.

Q: Do I need to change my code to use RexCLI? A: No. It’s a background tool that works automatically after installation. You keep using your original commands, and the memory feature runs silently in the background.

Q: Does RexCLI record private content? A: No. It only stores task progress in local project directories and doesn’t upload any data. It also provides a Privacy Guard feature that can automatically filter sensitive information.