Features

Run speech models locally or connect your own service. Dictate, transcribe files, generate speech, and manage your runtimes in one workspace.

Customize the overlay

See recording and processing status without bringing the main window forward. The native overlay stays out of the way of your clicks and typing.

Layout, position & size: Choose Minimal, Capsule, Meter, or Detailed, then adjust its screen position and scale.
Appearance & behavior: Adjust the surface, visualizer, opacity, glow, motion, and which phases are visible. Or turn the overlay off.
Preview before saving: Try unsaved changes in the app without recording. Live dictation can use a single-row caption strip instead.

Overlay settings →

Overlay appearance

Illustrative preview. Adjust the real overlay in Settings → Overlay.

Run locally

Install a runtime, download a model, and start it from Freehand. Its Connection appears automatically, ready to select for your task.

Local setup guide →

NeMo-Speech.cpp

Windows · macOS 13+

Nemotron 3.5 Streaming, Parakeet TDT v3, and MagpieTTS

Transcription with Nemotron or Parakeet, plus optional MagpieTTS speech in the same runtime.

Set up NeMo-Speech.cpp →

whisper.cpp

Windows only

Whisper Tiny through Large, including Turbo and quantized variants

Completed recordings and audio files. Live dictation is not available.

Set up whisper.cpp →

llama.cpp

Windows · macOS 13.3+

S1-mini by Superwhisper

English transcript cleanup after recognition, with raw text kept as the fallback.

Set up llama.cpp →

Use CPU or supported GPU acceleration: NVIDIA CUDA on Windows and Metal on Apple Silicon. Intel Macs use CPU. Freehand checks your host before offering an installation.

Start and stop runtimes and follow explicit downloads in a catalog grouped by Transcription, Text to speech, and Cleanup. NeMo loads one transcription model and optional MagpieTTS together. Quitting Freehand stops the runtimes it owns.

Models are separate downloads. Managed NeMo covers recognition and MagpieTTS speech, and llama.cpp covers cleanup. Each task uses its selected connection; choosing remote cleanup sends the transcript to that server.

Your desktop workspace

One workspace for speech

Move between tasks, Connections, local runtimes, and History from the activity rail. Resize sidebars, open settings beside your work, and find commands from the title bar.

Explore the workspace →

CPU, RAM & GPU at a glance

Watch activity and memory in the status bar, with adapter details and high-use indicators. Readings cover all apps on this computer; GPU metrics depend on driver support. Remote server usage is separate.

Resource readings →

Startup & runtime output

Connection checks wait while a local runtime loads, then refresh when it is ready. Inspect output in the shared panel with search and log highlighting while keeping your task open.

Startup and output →

Recording controls

Toggle or hold to talk

Press a shortcut to start and stop, or hold it while speaking. Choose your microphone and recording limit independently.

Recording controls →

Speech detection

Trim silence, stop toggle recordings after a pause, or split longer recordings at checkpoints. These controls apply to completed dictation, not realtime.

Speech detection →

Shared vocabulary

Keep names and specialist terms in one list for Voice and audio files. Supported models receive them as recognition hints, not guaranteed corrections.

Vocabulary support →

Results & recovery

Current result & recovery

Copy a completed transcript without enabling history. If the original destination is no longer safe for insertion, the result stays available for explicit copy.

Results and delivery →

Optional session history

Revisit recent transcripts and compare raw and cleaned versions. History is off by default, stays in memory, and clears when you quit. It never retains audio.

History details →

Files & speech playback

Transcribe an audio file without a microphone shortcut. With text to speech enabled, listen to results or generate and save speech from text you enter.

File and speech workflows →

Model-specific controls and realtime availability depend on your selected model and backend.

Model support →