TL;DR. Neo is a voice-controlled computer interface for macOS. You hold a push-to-talk key, speak naturally, and Neo executes your intent — dictating into the focused field, switching apps, clicking buttons, or calling MCP servers. Every model runs on your device. Median command latency is 60–80ms; full round-trip from key release to action is 200–300ms.
Neo is built for developers and knowledge workers with RSI, and for anyone who wants full computer control by voice without cloud dependency.
Every Neo command runs through three stages, all on-device:
Hold your push-to-talk key and speak naturally. No wake words, no command syntax.
On-device speech-to-text and intent classification execute in 60–100ms on Apple Silicon.
Neo dispatches via macOS accessibility APIs, browser automation, or MCP servers — no extension required.
Cloud voice assistants ship every word you speak — including code, customer data, and credentials — to a third-party server. That's a meaningful attack surface. Neo's audio never leaves the machine: the speech model runs inside the app process, voice profiles are encrypted with AES-256 and stored in the Secure Enclave, and there is no upstream telemetry of utterance content. For teams in regulated industries (HIPAA, SOC 2), this changes what voice tooling is allowed at all.
"I built Neo because I refused to choose between programming and the health of my hands — and I refused to send every command I spoke to a cloud I didn't control."
Latency targets are measured against real-time PTT sessions on M-series Macs. The Princeton "GEO" paper on generative-engine ranking (2024) found that response latency is a primary driver of perceived assistant quality, and HCI research consistently places the threshold for "instant" feedback below 100ms. Neo is engineered against that bar.
Free forever tier. No credit card. macOS 12+ · Apple Silicon.
Download Free for macOSRelated: For developers · For RSI · Compare Neo