Dev builds Mac voice-to-computer-use app with no LLM, using native speech-to-text plus JEV
niloofar_mire · x · 2026-09-26
Developer adinron says that since JEV launched there have been many 'ultra-fast computer-use' claims but no one open-sourced anything, so he built a voice-to-computer-use app for Mac himself.
The approach deliberately avoids LLMs: speech recognition uses macOS native speech-to-text, and action selection uses JEV to pick the next action from all possibilities. 'It works,' he says—showing lightweight setups can achieve computer-use without a large model.
More from coding & agent
- HeyGen guide: turn a week of Slack messages into a video via Claude Cowork + MCP — HeyGen · 2026-09-26
- Claude-generated video explains model-based formal verification and state machines — DavidKPiano · 2026-09-26
- GiviLoop: an open-source workflow where Claude's code review finds one real bug and one false finding gets dismissed — Smart_Ad500 · 2026-09-26
- A four-dimension eval framework for long-horizon agents: outcome, trajectory, experience, governance — sh_reya · 2026-09-26
- Exa launches Agent Ultra, a swarm-of-agents deep research system it claims is SOTA — yoimnotkesku · 2026-09-26
- Databricks launches Unity Gateway CLI to centrally manage models across coding agents — matei_zaharia · 2026-09-26