LFM3-Omni-2B runs locally on iPhone 17 Pro Max doing agentic web search with native speech generation

fuzhongkai · reddit · 2026-11-23

A developer added Liquid AI LFM3-Omni support to TensorSharp, an open-source .NET local inference engine and agentic runtime, demoing LFM3-Omni-2B (Q4KM) running on an iPhone 17 Pro Max.

The demo asks the agent to search recent Anthropic news and assess the stock: the model decides to search, invokes a web search tool, processes results, then generates a native female voice reply streamed alongside text — via LFM3-Omni's own audio decoder, not iOS TTS or an external API. Only the search needs internet; all inference and speech run on-device.

The author frames this as local multimodal models moving beyond chat to agentic tool use with speech on phones, with speed, memory and latency still to optimize. Engine and model are both open source; investment-content disclaimer included.

Original post →

More from coding & agent

coding & agent channel →