Q: Would GitIngest be good for coder bots?

[BradKML]

GitIngest can be used for regular LLM, but what are the tailoring needed to make it suitable for more advanced tools like Aide or OpenHands or Agentless?

Cross-ref: All-Hands-AI/OpenHands#2185

P.S. How is GitIngest different from the other tools like

• https://github.com/mufeedvh/code2prompt
• https://github.com/kamilstanuch/codebase-digest
• https://github.com/raphaelmansuy/code2prompt

[cyclotruc]

@BradKML
Yes! It's actually what gitingest aims to become: a way for agents to ingest code easily

The main difference between this project and the ones you mentionned is basically the ecosystem it focuses on: Python

https://github.com/yamadashy/repomix is more focused on NPM ecosystem
https://github.com/mufeedvh/code2prompt is Rust based so will probably offer the best perfomances

Gitingest will soon have an API that will be a convenient way to get ingests from every source through simple http requests

[BradKML]

@cyclotruc let's say OpenHands and other tools are written in Python, BUT the agent is designed to be programming language-ignostic, so should gitingest be used, or mufeedvh's code2prompt?

[ODAncona]

@cyclotruc Python sdk for Code2prompt is available now. Since this commit 😉

[cyclotruc]

@ODAncona Good to know, thanks!
I was wondering if I should start using code2prompt to improve the performances of gitingest's web UI, I guess time will tell

Looking forward to see how those 2 projects and the others will fit in the big puzzle of code ingestion that we're trying to solve

[ODAncona]

@cyclotruc @BradKML

I ran benchmarks on 12 repositories on my computer using basic command-line tools, and the results were interesting. Code2Prompt is still single-threaded.

[BradKML]

So code2prompt is unironically 10x fast? Surprising unless we are talking about the Python vs Rust gap

[ODAncona]

Updated on 2025-02-17

@BradKML @cyclotruc

I found another repo, doing the same thing and here's my findings:

Star History

Performance

Features

Aspect	REPOMIX	Code2prompt	Gitingest
Language	TypeScript/Node.js	Rust (with Python bindings available)	Python
User Interface	Command‑line (CLI)	Command‑line (CLI) and Python API	Web‑based (FastAPI web server) + CLI
Output Formats	Plain text, XML, Markdown (customizable via templates)	Markdown, Handlebars‑rendered text prompt	Text digest
Concurrency/Performance	Uses worker threads (Piscina) for parallel processing	Compiled in Rust for high performance	Python-based; asynchronous Git cloning and processing
Git Integration	Supports remote repo processing, Git commands	Includes Git diff extraction; has Git module	Clones repositories, parses Git URLs, supports branch/commit extraction

[filipchristiansen]

@ODAncona

Thanks for sharing those benchmarks—although I have to say, shaving off a second or two in code ingestion doesn’t make much difference when the LLM itself could be chugging away for half a minute or more! Sure, Rust is compiled and typically faster than Python at raw text processing, but in practical terms, the user experience isn’t about whether something takes one vs. a few seconds to parse code.
We focus on providing a smooth Python-based workflow and a user-friendly interface, because that’s where most of our users find real value. But hey, if microseconds are a deal-breaker, feel free to give code2prompt a spin 😄—it’s always great to see the ecosystem growing. In the meantime, we’ll keep doing what we do best: making code ingestion simpler and more accessible for everyone.