Death Of Cha

modified frame from naruto shippuden

When I first started working on Cha back in early January 2024, I built it because I needed it. There was no real good CLI terminal tool that integrated LLMs into the terminal at the time that fit what I was looking for. I tried many tools, with chatblade being the closest option, but none of them really worked for my needs.

So one day, I decided to build it myself. I started coding with the help of GPT-4 through ChatGPT to make a draft and outline, and went from there. I kept adding features to Cha based on what I needed, and over time, I really enjoyed working on it to the point where it became an addictive passion project on the side; see my first blog post about Cha for more details on the origins.

What made it so fun was that this was actually a tool I used every day. So every time I worked on it, it usually led to improvements that directly benefited my workflow and productivity.

Cha did find a place in the market of LLM tools. It got 64 stars on GitHub and 6 forks with no marketing and just word of mouth through Reddit. I had features that most CLI tools like this could only dream of at the time while staying simple and easy to iterate on. It got to the point where I thought Cha had a real chance to become a serious product for developers rather than just this fun side project I made for myself.

But when 2025 rolled in, everything changed. If 2024 was the year of AI-powered IDEs, then 2025 was the year of AI-powered CLI tools. Anthropic released Claude Code CLI along with updated and refined Sonnet and Opus models, showing the world that they had the best coding models and the tools to back them up. Within a few weeks, Claude Code CLI took a significant chunk of market share from Cursor IDE, and I had to try it out. I was immediately hooked. It was the most powerful AI coding tool I had ever used, and it changed how I code and build projects forever. Then with the success of Claude Code CLI, a flood of AI CLI tools started getting released like Gemini CLI, OpenAI Codex CLI, GitHub Copilot CLI, Cursor CLI, Kiro CLI, Amazon Q CLI, OpenHands CLI, Qwen Code CLI, Aider, and many many more.

Out of all of them, only Claude Code CLI, Gemini CLI, and OpenAI Codex CLI really stuck around. Claude Code CLI took most of that market.

With all of this and the power of Claude Code CLI, Cha almost died at that point. But what Cha had that none of these other CLI tools really offered was that it was simple, fast, and much cheaper to use. It also allowed users to easily switch between different LLM providers like OpenAI, Anthropic, Google, and xAI. The main value was that it was faster to load, gave users full control without the vibe coding feel, and was significantly cheaper overall. The core thing keeping it alive was that it was still pretty powerful for the price, especially compared to the exorbitant prices for tools like Claude Code CLI and Gemini CLI at the time.

But later into 2025, everything changed again when Anthropic released the claude-haiku-4-5 model. This model costs just $1 per million input tokens and $5 per million output tokens while being incredibly smart and effective. It seemed specifically tailored for tool use in Claude Code CLI. This changed everything because now Claude Code CLI was both fast and not that expensive. My costs for Claude Code dropped to about a third of what they were before because this model allowed me to do 90% of what I had Sonnet and Opus do.

This truly killed Cha. Sure, Cha was still faster and cheaper than Claude Code CLI, but not by much anymore. The performance and price of Claude Code CLI destroyed the last real value Cha provided. It didn’t help that mega-billion dollar companies like Google were now entering the AI CLI market, and the model providers themselves like OpenAI and Anthropic were too. With these two factors, my personal usage shifted dramatically. I used to split my time roughly 50/50 between Cha and other CLI tools, mainly Claude Code CLI. Now it’s more like 8% Cha and the rest is spent with other AI tools, still mainly Claude Code CLI.

This in turn killed Cha. By mid 2025, I had migrated Cha from Python to Go, renaming the vision to Ch to reflect the fresh start; see my blog post on migrating from Cha to Ch, the Ch project blog, and the Ch website for more details on this transition. Ch even picked up 10 stars and 1 fork with no marketing at all. But even with this rewrite, it couldn’t overcome the momentum that Claude Code CLI and other major players had already built. Unless Cha became something entirely different beyond this Go migration, it had no real chance to compete. And even then, the odds of gaining traction against mega-corporations and model providers are pretty very slim and almost impossible.

So yeah, this sucks. But what can you do? This industry with tech, especially in LLMs and AI, moves incredibly fast. It taught me that platforms win in the end, which aligns with my personal theory: The Barbell of Software.

But honestly, I’m not too sad about this because I really enjoyed working on this project. It helped me relax during some of the toughest moments in my life when things were really difficult in late 2024 and early 2025. It also allowed me to make a tool I actually love and still use. So I’m grateful for the fun time I had working on this project and for everything I learned from it.

Going forward, I’ll think more carefully about what I work on and which projects I commit to this deeply. Passion is great, but I’m at a point in my life where I want my projects to have real impact. So I’ll still maintain Cha, but I won’t go overboard with it. I’ll just update packages and fix small bugs here and there (probably using Claude Code CLI, ironically). Over time, it’ll likely get deprecated in the next year or two as these other AI tools keep improving and getting cheaper.

With Cha, I won’t force it to stay alive and I’ll focus on the next chapter of my life instead. A lot has changed for me personally since I started Cha, both in my career, health, and personal life. Just like most things, the best times come to an end and a new chapter begins.

To anyone who used Cha, thank you! I’m grateful. The code is open source, so do whatever you want with it, and I’d love to hear about the projects you all build with it.

Thank you all!