Show HN: Pasteface - Record, transcribe and copy speech locally in your terminal

Developer blairjordan released Pasteface, an open-source terminal application that records, transcribes, and copies speech locally using whisper.cpp or OpenAI's transcription API. Pasteface installs via a Rust toolchain and FFmpeg on Linux and macOS, organizes recordings into named transcript tabs, and transcribes earlier chunks while recording continues; Windows is not currently supported.

Record, transcribe, and copy speech from your terminal. Organize recordings into named transcript tabs, keep recording while earlier chunks transcribe, and copy the results to your clipboard. Transcribe on your machine with whisper.cpp or use OpenAI’s transcription API. Install a current stable Rust toolchain and FFmpeg. Local transcription also requires a separate whisper.cpp executable and model; see Transcription settings transcription-settings . Install the native dependencies for the default build, which includes the tray: Arch / Manjaro sudo pacman -S --needed base-devel pkgconf alsa-lib gtk3 xdotool libayatana-appindicator ffmpeg Debian / Ubuntu sudo apt install build-essential pkg-config libasound2-dev libgtk-3-dev \ libxdo-dev libayatana-appindicator3-dev libwayland-dev libxkbcommon-dev ffmpeg Clone and install: git clone https://github.com/blairjordan/pasteface.git cd pasteface ./scripts/install-linux.sh The installer builds Pasteface and installs the executable in ~/.local/bin and a desktop launcher in ~/.local/share/applications . Ensure ~/.local/bin is on your PATH . The tray requires a StatusNotifier/AppIndicator host. On i3, snixembed can bridge it to an enabled i3bar tray. Recording and the terminal interface also work without a tray host. Install the Xcode command-line tools, Rust, and FFmpeg: xcode-select --install brew install ffmpeg git clone https://github.com/blairjordan/pasteface.git cd pasteface ./scripts/bundle-macos.sh open target/Pasteface.app The script creates an ad-hoc signed application bundle that opens the TUI in Terminal. Grant microphone access when prompted; Terminal may also need microphone permission. This is a source build, not a notarized release. macOS compilation is covered by CI; microphone, menu-bar, and permission behavior still need testing on a Mac. cargo build --release --locked ./target/release/pasteface Omit the tray and its GTK dependencies. cargo build --release --locked --no-default-features The tray-free build still requires the platform audio libraries and FFmpeg. Linux and macOS are the target platforms; Windows is not currently supported. Run pasteface to open the terminal interface and start the tray companion. | Key | Action | |---|---| | Space | Start or stop recording; stopping enqueues the chunk | | C | Copy the selected tab’s transcript | | E | Export the selected tab to a .txt file | | Ctrl+C | Copy only the latest completed chunk | | X | Clear the selected tab’s recordings and transcript after confirmation | | S | Open settings, including microphone selection | | Tab / Shift+Tab | Switch transcript tabs | | + | Create a transcript tab | | F2 | Rename the selected tab | | ↑ / ↓ | Scroll the transcript | | Esc | Cancel active transcription and all queued chunks | | Q | Close the terminal interface | Click + to create a named transcript tab. New recordings append to the selected tab; recording a chunk never creates a tab. Click a tab to switch transcripts, or click a queue block to open its owning tab. Tabs also support / navigation. Right-click a tab and choose Rename , or press F2 , to give it a persistent name. Close tab in the same menu hides the tab while retaining its transcript and audio. Use Reopen beside + to restore closed tabs. Click the footer controls to record, copy, clear, cancel, or open settings. Settings and the microphone picker support clicks and the mouse wheel. Scroll over the transcript with the wheel. Drag across transcript text to highlight and copy it; hold Shift while dragging to use native terminal selection instead. Press E , or right-click a tab and choose Export… , to save its completed transcript as plain UTF-8 text. Edit the suggested path and press Enter to save. Relative paths use the terminal’s working directory; ~/ is supported. Existing files are never overwritten. The microphone meter responds to your voice while recording. Queue blocks show each chunk's duration and whether it is queued, transcribing, completed, failed, or canceled. Cancel preserves existing text and recorded audio. It does not stop a recording in progress. Clear is available once recording and transcription have stopped, and does not erase the system clipboard. In a picker or settings dialog, Esc closes the dialog instead. The terminal and tray control the same background recorder. Closing either interface leaves the recorder running. To reopen the terminal, run pasteface again or choose Open terminal from the tray menu. The tray circle changes with recording and transcription status. pasteface tui --no-tray Terminal interface without starting the tray pasteface tray Tray interface only pasteface start pasteface stop Enqueue the recording and return immediately pasteface toggle pasteface cancel Abort active transcription and cancel the queue pasteface copy pasteface clear pasteface status JSON state, including transcript and chunk status pasteface devices pasteface doctor pasteface shutdown Stop an idle recorder pasteface retry retries the oldest failed chunk. No global shortcuts are registered; a window manager or other automation can invoke the CLI commands. To restart the terminal interface, press Q and run pasteface . To restart the recorder, first finish or cancel its work, run pasteface shutdown , then reopen the app. Press S to choose a provider, configure the local executable and model, or enter an OpenAI API key. Settings are saved separately from recorded audio. Provider changes apply to subsequent transcription jobs; an active job keeps the settings with which it started. Local transcription processes audio on your machine. Install an executable and model using the whisper.cpp instructions https://github.com/ggml-org/whisper.cpp , then set their paths in Pasteface. The default paths are: ~/.local/lib/whisper.cpp/build/bin/whisper-cli ~/.local/lib/whisper.cpp/models/ggml-large-v3-turbo-q5 0.bin Acceleration is a runtime setting: | Mode | Behavior | |---|---| | Auto | Let the configured whisper.cpp executable choose its available backend | | CPU | Disable GPU inference for the configured executable | | Vulkan | Use a separately installed Vulkan-capable whisper.cpp executable | Pasteface does not bundle Whisper models, Vulkan drivers, or a GPU runtime. Vulkan support belongs to the external whisper.cpp installation, so changing it does not require rebuilding Pasteface. Select the appropriate executable in settings and install the Vulkan driver for your GPU. A CPU-only executable does not gain GPU support by selecting Vulkan. Advanced backend environment overrides, such as GGML VK VISIBLE DEVICES , can be set in the environment object in transcription.json . Supported prefixes are GGML , VK , and MESA . Select OpenAI in settings and enter your API key in the masked field, or set OPENAI API KEY before starting the recorder. The default model is gpt-transcribe ; the model name is editable. When selected, this provider sends completed recordings to OpenAI's audio transcription endpoint. API usage is billed to your OpenAI account. Local mode does not upload audio. Recordings must fit the API's 25 MB file limit after conversion; use shorter chunks for longer sessions. Keys entered in the app are stored in a separate secrets.json file with owner-only permissions 0600 . They are not included in status responses or transcript files. A saved key takes precedence over the OPENAI API KEY environment variable. See the official file transcription guide https://developers.openai.com/api/docs/guides/speech-to-text and gpt-transcribe model documentation https://developers.openai.com/api/docs/models/gpt-transcribe . Capture uses the selected microphone's native sample rate. After stopping, FFmpeg converts the recording to mono 16 kHz PCM. Each chunk is transcribed as a complete recording. Text appears when the chunk finishes; live partial transcripts are not currently displayed. State is stored in a private application directory: normally ~/.local/state/pasteface on Linux or the platform application-data directory on macOS. It contains transcription.json , the optional secrets.json , recorded chunks, processing logs, and the combined transcript. Each recording has its own directory. One worker drains the queue in order while microphone capture remains available. There is no fixed chunk-count limit; recordings use available disk space. Queued work survives recorder restarts, and a failed chunk retains its audio without blocking later chunks. Canceled chunks remain canceled after a restart. | Environment variable | Purpose | |---|---| | WHISPER BINARY | Override the configured whisper.cpp executable path | | WHISPER MODEL | Override the configured local model path | | OPENAI API KEY | OpenAI key fallback when no key is saved | | AUDIO DEVICE | Override the saved microphone selection | | TERMINAL | Terminal executable for the tray's Open terminal action on Linux | | PASTEFACE AUTO COPY=0 | Disable automatic clipboard updates | | PASTEFACE REDUCED MOTION=1 | Disable the recording shimmer | | PASTEFACE STATE DIR | Override the application state directory | Environment variables are read by the recorder process. Restart an idle recorder after changing them. Desktop launchers inherit the graphical session's environment, which may differ from a terminal's. See CONTRIBUTING.md https://github.com/blairjordan/pasteface/blob/main/CONTRIBUTING.md and AGENTS.md https://github.com/blairjordan/pasteface/blob/main/AGENTS.md . All commits must follow Conventional Commits. cargo fmt --check cargo clippy --all-targets -- -D warnings cargo test cargo check --no-default-features Tests use isolated recording fixtures and mock HTTP responses. They do not record from your microphone or upload your audio to OpenAI.