NockAI (@useNockAI) header
NockAI (@useNockAI) profile picture

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

On your machine
Joined October 2026
Photo by NockAI (@useNockAI)Photo by NockAI (@useNockAI)Photo by NockAI (@useNockAI)Photo by NockAI (@useNockAI)Photo by NockAI (@useNockAI)Photo by NockAI (@useNockAI)Photo by NockAI (@useNockAI)Photo by NockAI (@useNockAI)Photo by NockAI (@useNockAI)Photo by NockAI (@useNockAI)
Nock Hub can now turn a scan into JSON, offline. NuExtract3 by @numind_ai: name the fields you want, give it a page, get JSON back. Thanks @etiennebcp We tested 6 documents. Clean pages: near perfect. Handwritten 1906 invoice: 7 of 8. Faded 1910 invoice: 3 of 6. nock add nuextract3
2
2
9
388
Also new in nock: --glossary for Index-Translate. Give it a file of your terms and it keeps them the same in every language. In our test, agreed terms went from 81% to 98%. pipx upgrade nock-cli nockai.dev/hub
2
208
We wrote up everything nock runs on your own machine: a 1 GB coder, a video model, an agent, and now Nock Hub. New today: MiniCPM-OCR from @OpenBMB. It read a 1914 typed letter and a Chinese court notice with every character right. A 1921 Serbian telegraph form, it got badly wrong. Both results are on the listing. pipx upgrade nock-cli nock add minicpm-ocr
1
4
1
9
1,341
Also new in nock 0.6.0: --glossary for translation. Give it your own terms, one per line, and it passes only the ones in your text. On our test it took Index-Translate from 81% to 98%. pipx upgrade nock-cli
93
A typed page took about a minute on a laptop CPU and about 4 seconds on an RTX 3090. Small enough to keep on your own machine for scans you would rather not upload anywhere. nock add minicpm-ocr
Awesome to see MiniCPM-OCR integrated into Nock! 💙 Thanks for sharing both the successes and the challenging cases. Real-world feedback like this is incredibly valuable! 🙌
2
5
301
NockAI retweeted
Awesome to see MiniCPM-OCR integrated into Nock! 💙 Thanks for sharing both the successes and the challenging cases. Real-world feedback like this is incredibly valuable! 🙌
We wrote up everything nock runs on your own machine: a 1 GB coder, a video model, an agent, and now Nock Hub. New today: MiniCPM-OCR from @OpenBMB. It read a 1914 typed letter and a Chinese court notice with every character right. A 1921 Serbian telegraph form, it got badly wrong. Both results are on the listing. pipx upgrade nock-cli nock add minicpm-ocr
1
4
1
11
1,389
6,232 downloads on @huggingface. The surprise: most of them, 4,521, came from community builds of Nock Coder, not our own repos. People are running it their own way. That's the point of open weights.
3
13
438
Use Nock Coder in your editor, on your own machine. pipx install nock-cli nock add coder nock serve That gives your tools one local, OpenAI-compatible address: http://127.0.0.1:8080/v1 It listens on this machine only and logs nothing you ask. pypi.org/project/nock-cli
5
12
521
In your tool, set three values: Base URL: http://127.0.0.1:8080/v1 API key: anything Model: nock-coder Streaming works, and we tested it with the official OpenAI Python client.
1
4
162
Nock Coder is 1 GB and doesn't need a GPU. On a laptop CPU it ran at about 25 tokens per second. It answers questions and writes code, but it can't drive agent or tool modes. Review code before you run it. nock-video and nock-agent use the same address once you add them.
1
44
Thanks for trying it, and for the screenshot. Cloud mode is paused while Surplus moves to its new setup. Switch to "On this device" at the top: the model downloads once, then runs on your own laptop. Fixing the page so it starts there.
2
7
421
A GPU running 24/7 for a model almost nobody calls helps no one. It also matches how we build NockAI. Our models are made to run where you are, not on a server waiting for traffic. Nock Coder is 1 GB and runs on any laptop. Through Ollama, through our CLI, and as of this week in your browser with no install: nockai.dev/app Pick "On this device". The model downloads once, then answers on your own hardware. Nothing you type leaves the page. Glad to keep building next to @SurplusRH.
Surplus is making its biggest change yet. Hosting goes from "a GPU per model" to free inference for every graduated project. The launchpad closes today and reopens with a new fee split where 60% of every launch's fees flows back to $SURPLUS. Here's the full plan 🧵
1
2
9
492
Nock Coder now runs in your browser. nockai.dev/app Pick "On this device": the model downloads once (1 GB), then runs on your laptop. Nothing you type leaves the page.
26
2
37
820
Three models. One tool. pipx install nock-cli nock add coder video agent nock code: Nock Coder, 1 GB, runs on any laptop. Pipe in an error log and ask why. nock video: ask your videos questions, offline. 21 seconds per answer on an RTX 4090. nock fix: Nock Agent, a fine-tune of Gemma 4 31B. It shows every edit and command and waits for your yes. Experimental, needs a big GPU. Nothing leaves your machine. pypi.org/project/nock-cli
5
1
18
537
$ nock video summarize Ask your videos questions. Offline. pipx install nock-cli nock setup Then: - nock video summarize meeting.mp4 - nock video ask clip.mp4 "what error shows on screen?" - nock video find clip.mp4 "when the cake comes out" Files or links. Links are downloaded to your own machine first. Nothing is uploaded, no account needed. It runs Nock Video 4B, our open model, through llama.cpp on your own GPU. On our 20 video test it found the moment in 42 of 56, answered 56 of 60, and gave 6 of 7 good summaries. pypi.org/project/nock-cli
1
1
15
490
One link. One line. Zero uploads. Soon.
2
10
449
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-…
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-…
12
350
One link. One line. Zero uploads. Soon.
2
1
9
619