GitHub radar
NVIDIA Nemotron Omni: video, audio and docs locally
NVIDIA's Nemotron 3 Nano Omni is now on Ollama — a 31B multimodal model that processes video, audio, images, and text together in one pass, capable of transcribing recordings, reading charts-heavy PDFs, and powering vision-capable agents.
Nemotron 3 Nano Omni is a multimodal model from NVIDIA that processes video, audio, images, and text in one inference pass. The 31-billion-parameter architecture activates only a fraction of weights per token, reducing memory pressure despite the overall parameter count. It understands video clips up to two minutes long, transcribes speech with word-level timestamps, reads documents with embedded charts and tables via OCR, and supports GUI automation for agentic tasks. The context window spans 256,000 tokens — enough for a lengthy document or a multi-turn session with multiple file inputs. NVIDIA built the model as part of the Nemotron family, drawing on Qwen3-VL-30B-A3B from Alibaba and other Qwen variants. The Ollama release makes it available locally; the compressed download weighs around 28 GB.
Why a vibe-coder should care
Being able to process meeting recordings, audio, and image-heavy documents in one place — without switching between different tools or cloud services — is useful for anyone who reviews or produces content. Running locally means the file never leaves the machine: handy for recordings, contracts, or research materials that shouldn't go to a third-party API. The 256k context window lets you send a full interview transcript alongside a PDF and ask questions across both at once.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install Nemotron 3 Nano Omni from NVIDIA locally: ollama run nemotron3 — then show me how to send it a video file and ask it to transcribe and summarize the content
A Mac with 48 GB of RAM or more, or a GPU with 32 GB — the compressed version takes around 28 GB.
Open on Hugging Face▌ More finds