opencode-parakeet v0.5.0

Windows server builds for local OpenCode voice dictation, mirrored from parakeet.cpp v0.5.0 (MIT — by the LocalAI team). Byte-identical to the upstream assets.

Assets

Models are downloaded straight from Hugging Face by the installer: mudler/parakeet-cpp-gguf.

Install (Windows, PowerShell 5.1+ or 7+)

irm https://raw.githubusercontent.com/moesuito/opencode-parakeet/main/install.ps1 | iex

Options like -Backend cpu, -Model q8_0 and -Configure: see the README.
The installer also installs an OpenCode plugin that starts the server automatically on every OpenCode boot (and skips starting when one is already listening).

This mirrors the latest parakeet.cpp release that ships binaries (upstream v0.6.0 currently has no assets).

opencode-parakeet 0.2.0

Local speech-to-text for OpenCode voice input — now with a lazy server and a bundled auto-start plugin.

What's new

• The server starts without the model (~8 MB, instant). POST /warmup loads it on demand; if no warmup arrives, the first transcription loads it automatically (fallback). GET /health reports {"status":"ok","model_loaded":true|false}.
• The model unloads right after each transcription, so memory is only used while you dictate. A warmed model that never gets used is unloaded after an idle timeout (default 120 s).
• New flags: --preload (eager loading), --keep-loaded, --idle-timeout N. The OpenAI-compatible surface is unchanged (POST /v1/audio/transcriptions, WAV, processed in memory).
• Plugin bundled (opencode-parakeet.ts): starts the server with OpenCode and pre-warms the model when a recording starts (subscribes to the voice.recording event).
• No temporary files: audio never touches disk, end to end.

Assets

Based on parakeet.cpp v0.5.0 (MIT — by the LocalAI team) with opencode-parakeet patches: lazy loading, POST /warmup, unload after request, idle timeout.

Install (Windows, PowerShell 5.1+ or 7+)

irm https://raw.githubusercontent.com/moesuito/opencode-parakeet/main/install.ps1 | iex