Welcome to rtk Discussions!

[pszymkowiak]

👋 Welcome!

We’re using Discussions as a place to connect with other members of our community. We hope that you:

• Ask questions you’re wondering about.
• Share ideas.
• Engage with other community members.
• Welcome others and are open-minded. Remember that this is a community we
  build together 💪.

To get started, comment below with an introduction of yourself and tell us about what you do with this community.

[alterphp]

Hi,

I'm a web developer, currently using Claude Code, Github Copilot, but also having some intents to use smaller LLM. I'm really interested in decreasing AI coding impact, from money and resource savings to reduced dependency to a remote tool !

If some of you code with locally hosted models, please let me know, my first tries have been frustrating 🥲

[relmaazouz]

Very nice tool! how about to generalize this for other languages ? (Python, Go ...)

[pszymkowiak]

hi !
Language support is growing fast — we recently added filters for Python (ruff, pytest, pip), Go (go test, golangci-lint), and Docker Compose. More coming via community PRs.

A plugin system is the most requested feature (#139) — not planned yet, but contributions and ideas are welcome.

Questions, ideas, PRs — all welcome!

[gjkim42]

Thanks for sharing this great tool!

I understand that applying rtk reduces a lot of tokens by compressing the output.

What about the performance?
Do we have some performance metrics of this? This compressed output should not be trained by LLM. I'd like to make sure that this doesn't reduce the coding performance of the agents, or only reduce minimally.
Or possibly, it may outperform the original outputs by filtering out the unnecessary outputs.

[pszymkowiak]

there is rtk gain command with options

[gjkim42]

we can get token saving performance using rtk gain.
what about the coding benchmark score?
Given that the LLM hasn't been trained on RTK-compressed output, there's a risk of degraded benchmark scores. That said, performance could remain on par, or potentially even improve if the compression effectively filters out irrelevant output.

[gjkim42]

Do you have a plan to support other coding agents not only claude code?

[pszymkowiak]

yes ! sorry but we have had a lot of users / issues / PRs . w re looking for maintainers .
roadmap > gemini , toml
w r looking for a tool to discuss about this : #238