@skvark

Co-founder @GitHits_com - Equip your agents to use open source right. | Scaled opencv-python to 100m+ downloads as a side hustle

 Tampere, Finland
githits.com
Joined July 2010
  • Tweets10.7K
  • Following2.3K
  • Followers1K
361Media
After ~5 years of maintainership and over 100 million downloads, I'm happy to announce that my opencv-python packaging project has been adopted by the OpenCV organization. It can be found in the future from github.com/opencv/opencv-pyt…. #OpenCV #python

GitHub - opencv/opencv-python: Automated CI toolchain to produce precompiled opencv-python,...

Automated CI toolchain to produce precompiled opencv-python, opencv-python-headless, opencv-contrib-python and opencv-contrib-python-headless packages. - opencv/opencv-python

github.com
7
1
62
What does the GitHits index actually contain? @jlitola built a quick prototype that visualizes the data we have. We parse everything down to symbol level, and have the call graphs and even ecosystem imports mapped out. We provide this for every OSS repo out there. GitHits has the map of the worlds code. Served via ultra fast GitHits index. No cloning needed. Your coding agents need this, in fact they will love it. npx githits@latest init
1
3
81
Olli-Pekka Heinisuo retweeted
Nathan Burg
@NathanBurg15h
We just mapped what six coding models actually know about six libraries they ship against every day. Codex gpt-6.1-sol · Codex gpt-6-luna · Kimi K3.1 Pro · DeepSeek V4 Flash · Qwen3 Coder Next · gpt-oss-120b vs pydantic · FastAPI · SQLAlchemy · Next.js · Zod · Tokio GitHits is the version-pin ground truth. We can finally see each model's real knowledge cutoff. Which model × library combo should we map next? githits.com/blog/model-knowl…
3
2
1
7
294
So how quickly exactly we index over 5 million lines of code? Let's take Bun as an example: the indexing estimate here was over 2 mins, but in the end we did it in 1 min 17 secs. I guess that is pretty fast. And you can do this for any commit / ref in your repo. How cool is that? Find yours here: index.githits.dev
1
75
Olli-Pekka Heinisuo retweeted
Nathan Burg
@NathanBurg23h
Want agents and software factories to start recommending your OSS project on their own? Get it in the GitHits index and grab your badge.
Olli-Pekka Heinisuo
@skvark23h
GitHits badges appearing across OSS ecosystem. Get yours now at index.githits.dev
2
4
182
GitHits badges appearing across OSS ecosystem. Get yours now at index.githits.dev
2
2
8
728
Our long running example and research tools send now these nice status events, at least Claude Code renders them nicely. Codex ignores them :(
5
6
306
X did not accept my screencapture video because of too big aspect ratio, linkedin took it in without issues :D
59
Olli-Pekka Heinisuo retweeted
Dillon Mulroy
@dillon_mulroyOct 26
in an era of constrained compute (cpu and memory crunch is coming folks) it is more important than ever to squeeze every ounce of performance and efficiency out of your services, esp if you operate at a meaningful scale
21
13
13
401
56,487
Olli-Pekka Heinisuo retweeted
Jack Timonen
@Jack_TimonenOct 26
Code Discovery for Agents and Software Factories Presentation by @NathanBurg at the SF Tech Week Agent Day organized by @yujian_tang & oss4ai youtu.be/TgwNQyQNnCs?si=lTu_…

Code Discovery for Agents and Software Factories, Nathan Burg, GitHits, SF Tech Week Agent Day 2026

Your coding agent can see your codebase, but not the mountain of op...

youtube.com
1
5
113
Great, I have a 3090
铁锤人
@lxfaterOct 24
卧槽，3090要可能要涨价了！！ 因为老黄的刀法被人破解了 有人给3090解了个锁，同样限制在 360W，PyTorch 测试从 75 TFLOPS 跑到了 135 TFLOPS，提升约 80%！ 你以为是超频，把功耗往上拉？ 但他改的是显卡里控制运算速度的寄存器，这些寄存器平时有权限保护，正常改不了 他利用 GPU 启动固件的漏洞，赶在 NVIDIA 驱动加载之前拿到写入权限，把相关运算改成全速，再交给原版驱动使用 解开的主要是 FP16／BF16 输入、FP32 累加这类 AI 运算的速度限制 但测试算力多了 80%！！ 4090 也有分支项目做到了：同类运算的指令吞吐，从约 165 TFLOPS 提升到 331 TFLOPS，接近翻倍 项目地址，附上👇 github.com/Zagitta/gpu-unloc…

GitHub - Zagitta/gpu-unlock: Unlock FP/BF16 instruction issue rate on Nvidia 3000's cards to their...

Unlock FP/BF16 instruction issue rate on Nvidia 3000's cards to their professional equivalents - Zagitta/gpu-unlock

github.com
1
175
Guess what GitHits does? Grep.
will brown
@willcbOct 24
"the bitter lesson for search is to grep over a filesystem, just like how google works" - rich sutton
5
366
Olli-Pekka Heinisuo retweeted
will brown
@willcbOct 24
"the bitter lesson for search is to grep over a filesystem, just like how google works" - rich sutton
42
45
8
1,210
59,906
Olli-Pekka Heinisuo retweeted
Jack Timonen
@Jack_TimonenOct 23
Someone's agent just asked to index @steipete's codexbar. If you want to chat with that repo, visit index.githits.dev/repositori…
1
2
70
We put our Research tool now on every repo and package page to our public index browsing site. No login gates, fully free to use (there are limits of course :D). Essentially, ask any question about any OSS repo and the Research will answer, using our search, grep and read tools. Test on any repo here: index.githits.dev/
2
1
6
218
If you have a public repo, just try it out! GitHits is a really powerful tool for organizations that do mostly OSS, as your agents can work without cloning in sandboxes, saving on disk, tokens and time. I would assume companies like @supabase and @vercel would find it extremely powerful for all of their engineers.
1
66
I think monorepos made some sense pre-AI and in the early days of AI. Harnesses don't really care where the code is in the fs, and it's better anyway security wise to keep different parts of a system logically separated in the agent age.
dax
@thdxrOct 24
i think i kept seeing people say "monorepos are good for agents" but we just switched to multi repos. i have a hard time caring about the benefits of monorepos anymore
1
172
Another argument against monorepos is that if you work in cloud sandboxes smaller units are cheaper and faster to operate on vs. massive monorepos. And we are btw solving this issue with GitHits on scale, you don't need to clone your dependencies to make agents work with them correctly.
1
37
Olli-Pekka Heinisuo retweeted
dax
@thdxrOct 24
i think i kept seeing people say "monorepos are good for agents" but we just switched to multi repos. i have a hard time caring about the benefits of monorepos anymore
229
32
31
1,920
200,005
Couldn't agree more.
Matt Pocock
@mattpocockukOct 22
The code, not the spec, should be the center of your universe
1
94
Olli-Pekka Heinisuo retweeted
Nathan Burg
@NathanBurgOct 23
Fun Easter egg in DeepSeek's new DeepInfer: 23 files say "Derived from vLLM." But the source tells a different story. vLLM is a server for any model. DeepInfer is one model, one job, no server. We traced both through GitHits. Read more: githits.com/blog/deepinfer-v…
2
4
73
Load more