Offline Research ($OFFR) OFFR is an on-device research assistant: a 35B-parameter model that runs locally on your phone with no network connection. There is no cloud inference, no API calls and no server dependency. Your questions and the model’s answers stay on the device. Why offline Most AI research tools depend on a remote model and live web search. That means connectivity, per-query costs, rate limits, and sending your queries to a third party. OFFR removes all of that, so it works on a plane, in a remote area or anywhere signal is poor. How it works Local inference: a 35B model running entirely on the phone [add: quantization format, runtime/framework, memory footprint]. Cited answers: responses come with citations to sources so you can verify claims [add: how sources are stored, such as a bundled offline knowledge base, its size and its coverage]. Zero network: no outbound requests during use. Benchmarks (early, small sample) Measured against Claude Opus 5.5 with web search on a set of 61 questions, graded blind: Build 0.7.0: 34% Build 0.7.2: 45% Build 0.8.4: 54%