Guides

Guides.

Practical, tested walkthroughs. The built-in tools get honest coverage first; Edict appears where it changes the answer.

Speech to text on a Mac

The whole field on one page: built-in dictation, the apps, file transcription, and where your audio goes in each.

Apple’s built-in dictation end to end: enabling it, the commands that work, its real limits, and when an app makes sense.

Voice-driving a coding agent without your audio leaving the Mac: setup, the wrong-window guard, auto-send, and SSH.

Every part of Cursor, not just the Agent prompt: setup, the wrong-window guard, and connecting Cursor to Edict over MCP.

Setup, connecting Codex over MCP (the one TOML config of the set), and the nvm trap that makes it vanish from detection.

The 1907 trademark that became the word, and the record-then-transcribe workflow that does its job today.

Zoom only transcribes cloud recordings on paid plans; here’s the on-device route for everything else, with speaker labels.

Record it with permission, then turn an hour of audio into searchable, timestamped text in under half a minute, offline.

The memo backlog nobody replays, turned into searchable text: Apple’s routes first, then the on-device batch workflow.

Category verdicts with the vendors’ own claims linked. Competitors win where they win, including the free ones.

Any file to words in seconds: the free routes, the formats, speaker labels, and when a cloud service actually helps.

Four routes that are free without an asterisk, ranked by effort, including the open-source one and where free runs out.

Every coding agent ships voice now. What Claude Code’s /voice and Cursor’s voice mode do, and where each one stops.

Category verdicts for talking instead of typing, plus why nobody can honestly tell you which one is most accurate.

What the research actually measured, why the 3x figure is a phone study, and the typing record that turns out to be talking speed.