The library page provides a local command, ollama run nemotron-3.5-bolt:e4b, as well as API examples for the Ollama client and chat endpoint.
Ollama describes Bolt E4B as an 8-billion-parameter model with 4 billion effective parameters, post-trained by NVIDIA from Gemma 4 E4B. It is positioned for local conversations, instruction following, and multi-turn tool use. The E4B download is listed at 7.2 GB to 7.9 GB, accepts text and image input, and has a 128K context window. Ollama says it is supported on Windows, Linux, and Apple Silicon, and identifies a DFlash draft model for speculative decoding.
The same page also lists the larger Bolt 74B-A9B variant, with 74 billion total parameters and 9 billion active per token. Its stated context limit is 256K, and Ollama says Apple Silicon and Metal are not currently supported for that variant.
The page does not provide independent benchmark results, latency measurements, or a fixed hardware-memory requirement for E4B. It states that usable context depends on available memory and that images, history, tool results, and output share the context budget. The listed license is OpenMDW License Agreement version 1.1.