fcitx5-vinput: pricing, features, alternatives

fcitx5-vinput is a voice-input addon for Fcitx5. It records from the keyboard, runs local or cloud speech recognition, then optionally rewrites the transcript with an LLM before inserting text. The repo xifan2333/fcitx5-vinput is C++, GPL-3.0, created 2026-03-06. On 2026-09-20 GitHub listed 451 stars and 45 forks. 451 stars is a heat signal, not traffic. Docs live at xifan2333.github.io/fcitx5-vinput. Packages exist for Arch, Fedora COPR, Ubuntu PPA, Nix, Flatpak, and GitHub Releases (full and lite).

It is an input method plugin, not a coding agent. Compare Lark CLI if you wanted Feishu from a terminal, or a speech model page if you only needed weights. There is no macOS or Windows IME in the README.

Key Features

Two triggers: tap the trigger key (default Alt_R) to toggle recording, or hold for push-to-talk.
Local and cloud ASR: offline sherpa-onnx models, or cloud providers the README names: Doubao, Aliyun Bailian, ElevenLabs, and OpenAI-compatible endpoints. Switch at runtime from the command palette (Shift_R then /asr).
LLM scenes: post-process for correction, formatting, or translation.
Command mode: select text, hold the command key (default Control_R), speak an instruction, release to apply.
GUI and CLI: vinput-gui for setup, vinput for full control. Full builds ship a local ONNX runtime. Lite builds are cloud-only.

Limitation: you need Fcitx5 on Linux, PipeWire, and a working microphone. Cloud ASR and LLM rewriting send audio or text to the provider you configured. Local ASR quality depends on the model you download in Resources → Models. This is not a general dictation app for macOS or Windows.

Use Cases

Linux desktop users who already live in Fcitx5 and want talk-to-type.
People who want offline ASR on the full package, or cloud-only on lite.
Skip if your desktop is not Fcitx5, or if you cannot send audio to a cloud vendor and do not want a local model.

Pricing

Piece	Price	Notes from first-party pages 2026-09-20
Addon	$0	GPL-3.0. Distro packages and GitHub Releases. About 451 stars.
Local ASR	Your disk and CPU/NPU	sherpa-onnx models downloaded in the GUI.
Cloud ASR / LLM	Provider bill	Doubao, Bailian, ElevenLabs, or an OpenAI-compatible key.

Getting Started

Install the package for your distro (Arch fcitx5-vinput, Fedora COPR, Ubuntu PPA, Nix flake, or Flatpak). Lite if you do not want sherpa-onnx.
systemctl --user enable --now vinput-daemon.service then fcitx5 -r
Open Vinput GUI → Resources → Models, download a model, tap Alt_R to record.

First-party start: README and the install docs.

Frequently Asked Questions

Full or lite?

Full includes the local sherpa-onnx runtime. Lite is cloud ASR plus LLM, with no local ONNX dependency.

Can I change the hotkeys?

Yes. Trigger, command, and palette keys are Fcitx5 settings. Defaults are Alt_R, Control_R, and Shift_R.

Does it work on Wayland?

The project is a Fcitx5 addon over PipeWire. Use a Fcitx5 build that already works on your compositor; the README does not claim a separate Wayland stack.

Alternatives

Desktop dictation on macOS: not this addon. Stay on the platform IME.
mlx-serve: local LLM server on Apple Silicon, not an IME.
Cloud ASR APIs alone: use them without Fcitx5 if you only needed a file transcriber.

Tips

Start local if the transcript must stay on the machine.
Use command mode on already-typed text instead of re-dictating a paragraph.
Keep the daemon enabled, or the trigger key will do nothing after a reboot.

Conclusion

fcitx5-vinput is the Linux Fcitx5 path for voice input with optional LLM cleanup. Install the distro package if you already use Fcitx5. Pick full for offline ASR, lite for cloud-only.