> ## Content Index
> Fetch the complete content index at: https://thenitai.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# How to save time and money with Codex and Claude
- URL: https://thenitai.com/how-to-save-time-and-money-with-codex-and-claude/
- Published: 2026-06-14T18:09:49.000Z
- Updated: 2026-06-14T18:09:49.000Z
- Description: I often see many users expressing frustration about running out of tokens due to their 5-hour session limits or weekly quotas. In contrast…
- Author: Nitai
- Tags: #Import 2026-09-21 03:50

I often see many users expressing frustration about running out of tokens due to their 5-hour session limits or weekly quotas. In contrast, I have been using both Claude Code and Codex extensively on the $100 monthly plan, which includes Opus 4.8 and GPT-5.5, and I’ve made a point of integrating these tools into my daily workflow.

Every day, I immerse myself in four distinct projects, dedicating 12 hours or more to their development and execution. During this time, I frequently manage 4 to 6 different chat sessions with either Claude or Codex. While some might consider this level of engagement modest — explaining the need for higher-tier plans — I’d argue that it’s a scenario many users face. Yet, despite this intense usage, I have never actually hit my token limits. I’ve come perilously close on a few occasions, but I’ve never found myself compelled to upgrade my plan.

I attribute my ability to sustain this level of productivity to my use of Kumbukum, which serves as a powerful knowledge and memory layer for all my AI tools. With Kumbukum, platforms like Claude, Codex, ChatGPT, FluentApp, and others no longer need to sift through the entire repository or rely on complex regex searches to locate files. This not only streamlines the process but also conserves both time and tokens, which is quite remarkable.

Moreover, as the system has begun indexing emails, its efficiency has notably improved. Now, when I encounter a customer issue, it swiftly retrieves and presents the relevant notes or memories alongside the associated email. This seamless integration allows me to address concerns more effectively and enhances my overall productivity.

Here are some compelling benchmarks that highlight the improvements we’ve achieved:

![](https://thenitai.com/content/images/cdn-images-1-medium-com/max/800/1-sqs3448xntb3xxgznjcemw.png)

Overall Combined Performance:  
From a cumulative usage of 14,272 tokens, we have reduced this to 1,964 tokens, representing an extraordinary 86.2% decrease.

In a typical production MCP workflow, three separate retrieval calls generated an estimated 14,272 tokens sent back to the AI model. After the integration of Kumbukum, this process now only transmits 1,964 tokens, yielding a staggering reduction of 12,308 tokens per retrieval set.

It is essential to clarify that the baseline we measured was not a scenario where “no search” was conducted. With Kumbukum in play, the system performs a semantic search that prioritizes returning locator metadata first. This enhancement allows the AI to access specific content only when necessary, streamlining the retrieval process.

The primary benefit we observe is a dramatic reduction in the number of tokens and contextual data processed, rather than a decrease in raw search latency, which has remained relatively consistent. The real-time savings arise from fewer manual searches through repositories, a significant decrease in the number of irrelevant files opened, and a reduction in the need for follow-up retrieval calls. This creates a more efficient and effective workflow overall.

[Kumbukum is an open-source system](https://github.com/kumbukum/kumbukum?ref=thenitai.com) that I’ve built and that has helped my AI workflow tremendously. Once hooked up, it automatically stores notes, memories, git repositories, URLs, and now also email (for triaging your emails). As outlined above, the addition of customer emails yields astonishing results, where my AI tools even highlight an issue that was also mentioned in an email.

In conclusion, using Kumbukum has greatly improved my workflow. It helps me get the most out of my AI tools while reducing token usage. I’ve saved time and increased efficiency by leveraging technology effectively.