Claude Code : j’ai réduit mes tokens de 90% grâce à RTK 💸

Item: Claude Code : j’ai réduit mes tokens de 90% grâce à RTK 💸
Rating: 4
Author: PostSingularity Institute

Claude Code: I reduced my tokens by 90% thanks to RTK 💸

Title history

Claude Code : j’ai réduit mes tokens de 90% grâce à RTK 💸 — observed on 20261110-0313 UTC (YouTube)

by Pentiminax · View the channel on PSI

Applied Sciences & Engineering Computing & Cybersecurity UComputing and Information Technology UDDigital Lifestyle and online world: consumer and user guides

🎙 Pentiminax 👥 30K 📅 March 19, 2026 ⏱ 10 min 👁 8K 📄 tutorial 🧭 2026-11-10

Available in: English (current) Français

Keywords

token reductionRTKClaude Codecontext pollutioncost saving

Summary

In this tutorial, the presenter introduces RTK (Rust Token Killer), an open-source tool that filters command outputs for AI coding agents, specifically aiming to reduce token consumption. The video explains the problem of context pollution, where verbose command outputs waste tokens and slow down agents. RTK uses source filters to compress outputs while preserving essential information, promising up to 89% noise reduction and longer sessions. The presenter demonstrates installation via curl or Homebrew, and shows how to initialize hooks for Claude Code or OpenCode. He walks through examples like cargo test, git status, git diff, and docker ps, illustrating token savings (e.g., from 146 to 50 tokens for docker ps, and from 84 to 17 for git status). The tool automatically rewrites shell commands through hooks, making integration seamless. The video also covers advanced usage, such as creating custom filters and referencing the architecture.md file. The demonstration is hands-on, with real command examples and token counts, though the presenter notes that some savings vary. The overall value lies in practical advice for developers using AI coding assistants to optimize their workflows and reduce costs.

184 words

Critical Evaluation

Value of the Information & Strength of the Argument

The value of the information is high for developers using AI coding agents: it presents a working solution to a common problem (token waste) with concrete examples and measurable savings. The argumentation is built on practical demonstrations: the presenter shows exact token counts for standard commands versus RTK-filtered ones, making the benefits tangible. However, the evidence is anecdotal and based on the tool’s own examples; no independent benchmark is provided. The argument is persuasive for the target audience, but the lack of comparative analysis limits its rigor.

Scientific Rigor, Source Quality, Title Accuracy

The video’s scientific rigor is moderate. The presenter references the tool’s official site and GitHub, but does not cite independent studies or external comparisons. The demonstration is reproducible, which is positive, and the tool’s open-source nature allows verification. The adéquation between the title and content is strong: the video indeed explains how to reduce tokens by 90% (though the actual savings vary by command, as shown). The title might overpromise, but the content clearly addresses the claim. No external sources are provided beyond the tool’s site and the presenter’s personal site.

193 words

Title / Content Match

The title claims a 90% reduction in tokens, which matches the video's message about typical savings, though the exact percentage varies by command. The content directly addresses this.

Quality & Reliability

7/10

The video provides concrete demonstrations of token savings for several commands, and the tool is open source with a transparent filter system. However, it relies heavily on the tool's own claims and does not offer independent benchmarking or comparative analysis.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

RTK — Official website of the RTK tool, presented in the video with installation and usage examples.
Pentiminax — Personal site of the creator, mentioned at the end of the video as a place for more content.

External References

Contribution & Novelties

The video introduces a novel open-source tool (RTK) that addresses token inefficiency in AI coding workflows. Its original contribution is the concept of filtering command outputs to preserve only essential information, thereby reducing token consumption without losing critical data. The presentation includes concrete token-saving examples and installation steps, making it immediately applicable. The novelty lies in the tool’s approach of injecting hooks to transparently rewrite commands, requiring no changes to the existing agent workflow.

Pour aller plus loin :

Large language model - Wikipedia — Provides background on how LLMs process tokens and why context length matters.
Prompt engineering - Wikipedia — Related to optimizing inputs to AI models, which complements the tool’s goal of reducing token usage.
Context window - Wikipedia — Explains the concept of context limits in LLMs, directly relevant to the token reduction problem.

138 words

Radar Profile

The radar shows a balanced profile: the video scores highest on information quantity and reliability, with slightly lower technical depth. This indicates a practical, accessible tutorial that prioritizes usefulness over deep technical exploration.

Reliability 7/10