@useNockAI

Open models that run on your machine. Nock Coder: 1 GB, offline, no API key. 0xb0f444a069d9264017610e0674ebf7e1d6e19cbb

 On your machine
nockai.dev/
Joined October 2026
  • Tweets144
  • Following28
  • Followers344
63Media
$NOCKAI is live on @SurplusRH. The model behind it is Nock Coder: open weights, Apache 2.0, a 986 MB file that runs on the laptop you already have. → code first answers, plus Solidity for Robinhood Chain builders → Windows, macOS, Linux, no GPU needed, works with the Wi-Fi off → every raw test answer public, misses included → Surplus serves the exact Hugging Face commit we launched Trading fees pay for the GPU that serves it. CA: 0xb0f444a069d9264017610e0674ebf7e1d6e19cbb
46
11
8
97
16,449
NockAI
@useNockAI14h
One link. One line. Zero uploads. Soon.
1
6
215
Run 2 of Nock Video 4B is up, and two of run 1’s misses are fixed: family videos 7 to 9 of 12, and the basketball summary now sticks to what’s on screen. Before training it, we wrote down 6 goals to drop the “preview” label. It hit 5. Sport came in at 5 of 8, one short. So it ships today. huggingface.co/Nock-AI/nock-…
NockAI
@useNockAIOct 20
The GGUF is up! Nock Video 4B, preview: ask about your own videos, offline. When something happens, what's on screen, or a short summary. One 2.7 GB file for llama.cpp. Your video never leaves your machine. huggingface.co/Nock-AI/nock-…
9
172
One link. One line. Zero uploads. Soon.
2
1
6
334
The GGUF is up! Nock Video 4B, preview: ask about your own videos, offline. When something happens, what's on screen, or a short summary. One 2.7 GB file for llama.cpp. Your video never leaves your machine. huggingface.co/Nock-AI/nock-…
NockAI
@useNockAIOct 20
Replying to @useNockAI
Next: real sport and family clips in the training mix, kept apart from the test, so the pitch we draw isn’t the only sport it ever sees. Then a GGUF, so it runs in llama.cpp like the others.
2
2
1
18
828
more replies
Where it misses: slides went 10 to 9, family stayed at 7 of 12, and on a long basketball video its summary made up interviews that weren't there. Questions about what's on screen: 56 of 60, same as the base. All of it is on the model card.
1
3
61
To run it: llama-server -hf Nock-AI/nock-video-4b-GGUF --jinja -c 32768 -ngl 99 python ask_video.py my_video.mp4 "When does the error show up?" --moment ask_video.py is the exact setup we scored with. It's in the repo.
43
Sport barely moved: 1 to 2 of 8. Saying so, as promised. The rest did. First training run, same 20 test videos, same frames, same prompt: Find the moment: 29 to 38 of 56 Screen recordings: 5 to 11 of 12 Questions about the video: 52 to 53 of 60
1
1
11
351
Not all of it went up. Family videos dropped, 9 to 7 of 12. And on held-out videos drawn like the training data it jumped 32 to 55 of 62. That mostly tells us the drawn data is learnable, not that real footage is solved.
2
4
151
Next: real sport and family clips in the training mix, kept apart from the test, so the pitch we draw isn’t the only sport it ever sees. Then a GGUF, so it runs in llama.cpp like the others.
1
1
676
The prompt fixed screen recordings. Sport it couldn’t: it actually dropped to 1 of 8. So this week’s training data has matches in it. We draw them ourselves, so every answer is exact to the frame, like this one. 240 videos, 1,575 questions. First run is training now.
1
12
314
Why draw instead of scrape: most public data for “find the moment” is non-commercial, and hand labels drift by a second or two. The test set stays apart from all of this: different apps, people and content.
1
4
159
When the run finishes, it goes through the same 20 test videos as last week. If sport doesn’t move, we’ll say so.
30
One week of NockAI, in 42 seconds. Three open models that live on your machine: a 1 GB coding model, a coding agent, and a video model on the way. Everything we shipped is in the replies.
1
11
373
more replies
What we did behind the releases: Every test answer is public on the model cards. nockai.dev has a live demo. GGUF builds run in llama.cpp and Ollama. And the first weekly burn: 4,564,757 $NOCKAI sent to the dead address.

NockAI: open coding models that run on your machine

An open coding model in 1 GB that runs on the laptop you already have.

nockai.dev
1
1
41
1,100+ downloads on Hugging Face in the first week. All open weights, Apache-2.0. Week two: train the video model and run it on the same test. The score goes on the model card either way.
1
34
Load more