Claude, ChatGPT, Codex : ne change plus jamais de mémoire

Claude, ChatGPT, Codex : ne change plus jamais de mémoire

Claude, ChatGPT, Codex: never switch memory again

Title history
  • Claude, ChatGPT, Codex : ne change plus jamais de mémoire — observed on 20261104-1054 UTC (YouTube)

by Melvynx · View the channel on PSI

Applied Sciences & Engineering AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence
🎙 Melvynx 👥 67K 📅 November 4, 2026 ⏱ 13 min 👁 112 📄 tutorial 🧭 2026-11-04
Available in: English (current) Français

Keywords

switching costlocal-firstCLI toolsAI agentsMarkdown files

Summary

This tutorial addresses the common problem of being locked into a single AI assistant (like ChatGPT or Claude) because conversations, memory, projects, and tools are stored in proprietary apps. The creator argues that the real solution is to make the switching cost zero by building a resilient, local-first system. Three pillars are presented: (1) a ‘Brain’ folder on your computer containing all memory (agent.md, notes, projects, skills) in plain text files; (2) connecting tools via CLI instead of proprietary connectors, ensuring they are accessible across all AI agents; (3) using ‘skills’ (instructions) that any agent can read. Demonstrations show how the same memory and tools work seamlessly across Claude Code and ChatGPT Codex. The approach emphasizes simplicity: avoid databases and complex RAG systems, stick to Markdown files, and store everything locally and optionally on GitHub. This way, the AI becomes just a ‘motor’ that drives your existing tools and workflows, allowing you to switch agents freely. The video includes practical examples and offers free resources, including a ‘Connect Tool’ skill. Ultimately, the goal is to create an ecosystem where a single prompt can execute complex multi-tool workflows, making AI a true assistant rather than a dependency.

196 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides a clear, actionable method to decouple AI tools from personal data and workflows. It convincingly argues that storing memory in plain text files on the local machine is robust, portable, and future-proof. The demonstrations are concrete and show real usage, which strengthens the practical value. The argumentation is logical: by standardising memory and tools via CLI and skills, the user gains flexibility and avoids vendor lock-in. However, the approach relies heavily on the creator’s own experience and does not reference comparative studies or external validation. The suggestion to avoid sophisticated database or RAG systems is pragmatic but might not suit every use case. Overall, the value is high for practitioners but the scientific foundation is thin.

Scientific Rigor, Source Quality, Title Accuracy

The video does not cite academic or scientific sources; it references the creator’s own website and tools. The content is based on practical experience rather than rigorous experimentation. The title is an accurate summary of the content, and the video stays on-topic throughout. The demonstrations are reproducible, but the claim that ‘all AI agents’ can use the same skills and CLI tools requires verification across different platforms. The video is well-structured and clear, but the lack of external references limits its scientific rigor. The creator’s own links (blog, GitHub, newsletters) are provided as resources, but they are not independent sources. Overall, the video is a useful tutorial but not a scientific study.

246 words

Title / Content Match

The title accurately reflects the content: the video explains how to avoid being tied to a single AI's memory by making the memory portable, so users can switch between Claude, ChatGPT, and Codex without losing context.

Quality & Reliability

7/10

The video is a practical tutorial based on the creator's personal experience. It offers clear demonstrations and a coherent method, but lacks external scientific sources or peer-reviewed references. The claims about portability are plausible but not empirically validated. Technical details are accurate regarding CLI and Markdown, but the overall reliability is limited by anecdotal evidence.

Chapters

Cited Sources

Contribution & Novelties

The video presents a practical method to avoid AI lock-in by creating a local, file-based memory system and using CLI tools and skills that are agent-agnostic. It combines existing concepts (Markdown, CLI, skills) into a coherent workflow, making it easy for users to switch between AI assistants without losing context. The original contribution lies in the ‘Connect Tool’ skill that automates the linking of external services across different AI agents, and the emphasis on treating AI as an interchangeable motor rather than a singular platform.

Pour aller plus loin :

  • Vendor lock-in — The concept of being dependent on a single provider, which the video directly addresses.
  • Markdown — The text format recommended for storing memory and documentation; its simplicity and longevity are key to the method.
  • Command-line interface — CLI tools are central to the approach for tool connectivity, ensuring cross-agent compatibility.
  • Local-first software — A design philosophy that prioritises local storage and control, aligning with the video’s core ideas.

161 words

Radar Profile

The radar profile shows strong scores in quantity of information (8) and moderate scores in quality (7), technical level (5), and reliability (6). This indicates a content-rich tutorial with practical demonstrations, but with limited technical depth and scientific backing. The low technical level reflects accessibility for a general audience interested in AI workflows.

Reliability 6/10