Ton Claude Code oublie TOUT ? Tout savoir sur sa mémoire ! (Guide complet)

Item: Ton Claude Code oublie TOUT ? Tout savoir sur sa mémoire ! (Guide complet)
Rating: 4
Author: PostSingularity Institute

by Aurélien FAGIOLI

Formal & Physical Sciences AI — Models & Research UYQArtificial intelligence (AI)UYQMMachine learning

🎙 Aurélien FAGIOLI 👥 21K 📅 March 22, 2026 ⏱ 46 min 👁 6K 📄 tutorial 🧭 2026-08-28

Available in: English (current) Français

Keywords

tokencontext windowcompactmemoryClaude Code

Summary

This video by Aurélien FAGIOLI is a comprehensive tutorial on how memory works in AI models like Claude, GPT, and Gemini, with a focus on Claude Code. The creator explains the concept of tokens, comparing them to currency or calories for AI, and details how text and images are converted into tokens. He then discusses the context window, emphasizing that the available space is not fully usable due to system prompts, tools, and other overhead. The video covers the transition from 200K to 1 million tokens in Claude Opus 4.6, highlighting the benefits of a larger context but also the risks of ‘context rot’ and the ’telephone game’ effect during compaction. The creator demonstrates the /compact command and its negative effects, such as information loss and distortion. He also presents a comparison of memory capabilities between different models and explains the three layers of memory in Claude Code. Practical advice is given on when to use /clear versus /compact, and a summary of best practices for managing memory in Claude Code is provided. The video is aimed at both beginners and intermediate users, with a pedagogical approach using analogies like chocolate bars and suitcases.

194 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides valuable insights into the practical workings of AI memory, especially for users of Claude Code. The creator uses clear analogies (chocolate bars, suitcases) to explain complex concepts like tokens and context windows, making them accessible. The argumentation is structured and logical, moving from basic definitions to advanced topics like compaction and context rot. However, the creator claims to have conducted ’exclusive studies’ and references ‘scientific research’ without providing specific citations, which weakens the argumentative rigor. The practical tips, such as using /compact and understanding the three memory layers, are useful and actionable.

Scientific Rigor, Source Quality, Title Accuracy

The video lacks rigorous scientific sourcing; the creator mentions ‘studies’ and ‘research’ but does not provide specific references or links. The description contains promotional links (formation, booking, hosting) but no direct sources for the technical claims. The title accurately reflects the content, which is a comprehensive guide to Claude Code’s memory. The video is well-structured with clear chapters, and the creator’s pedagogical approach is effective, but the absence of verifiable sources reduces its scientific credibility.

185 words

Title / Content Match

The title accurately reflects the content, which is a comprehensive guide to Claude Code's memory, including token usage, context windows, and compaction.

Quality & Reliability

6/10

The video provides a clear and structured explanation of tokens, context windows, and memory mechanisms in Claude Code, but relies on analogies and personal claims rather than verifiable scientific sources. The creator mentions 'exclusive studies' and 'scientific research' without providing specific references, limiting the overall reliability.

Chapters

Cited Sources

Formation n8n (early access) — Mentioned as a resource for learning n8n automation.
Calendly - Book a call — Mentioned for booking a call with the creator.
Hostinger VPS (affiliate) — Mentioned as a hosting recommendation for n8n.
Skool Community — Mentioned as a free community with resources and live sessions.

Concurring Sources

Anthropic Claude Code documentation — Official documentation on Claude Code features, including memory and context management.

Dissenting Sources

Lost in the Middle: How Language Models Use Long Contexts — This study suggests that LLMs may not effectively use all information in long contexts, which aligns with the video's 'context rot' concept but also indicates that larger context windows may not always improve performance.

Contribution & Novelties

The video offers a practical, user-focused explanation of AI memory mechanisms, specifically for Claude Code, with a unique pedagogical approach using analogies. It highlights the often-overlooked overhead in context windows and the risks of compaction (telephone game effect). The ’three layers of memory’ concept is presented as a framework for understanding Claude Code’s memory management.

Pour aller plus loin :

Tokenization in LLMs — Provides background on how text is split into tokens.
Context window in LLMs — Explains the concept of context windows and their limitations.
Lost in the Middle: How Language Models Use Long Contexts — A study on how LLMs handle long contexts, relevant to context rot.
Claude Code documentation — Official documentation for Claude Code, including memory and context management.

124 words

Radar Profile

The radar profile shows high scores in quantity of information and technical level, reflecting the video's comprehensive coverage and technical depth. However, the lower scores in quality of information and global reliability indicate that the content, while informative, lacks verifiable sources and scientific rigor.

Reliability 5/10

💬 Sur les 0 commentaires analysés, aucune tendance n'est disponible.