vedantnimbarte

IRA — AI skill for Claude Code

AI community

Voice-first AI assistant in Rust.

How to install IRA

This entry records only its repository, not the path inside it, so there is no exact command to give. Open vedantnimbarte/IRA and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What IRA does

Voice-first AI assistant in Rust. Wake word, full-duplex speech with barge-in, hybrid local/cloud inference. Currently the voice loop: openWakeWord to Silero VAD to Whisper to streaming Claude to Piper TTS.

Alternatives in AI

Inference Serving — ai-research-skills vLLM, SGLang, TensorRT-LLM, llama.cpp 5.4k ★
Deepreasoning — A high-performance LLM inference API and Chat UI that integrates DeepSeek R1's CoT reasoning traces with Anthr 5.4k ★
Azure OpenAI .net — GPT-4, embeddings, DALL-E, and Whisper client 1.8k ★

README

IRA

A voice assistant you can interrupt.

Wake word → endpointing → speech-to-text → streaming model → streaming speech, with barge-in. It calls tools, asks before it changes anything, and shows you what it is doing on a screen.

The bet is that **turn-taking is the feature**. Speech recognition has been good enough for years; what stays broken is being cut off when you pause to think, and having to wait out an answer you already know is wrong. A slower model that yields the floor correctly beats a smarter one that talks over you.

Install

Download an installer from [Releases](https://github.com/vedantnimbarte/IRA/releases), or build her from source below.


Windows	`ira--x64.msi`	Installs for you only — no administrator rights, no UAC prompt. She lands in the Start menu, and `ira` works in any terminal.
Linux	`ira__amd64.deb`	`sudo apt install ./ira_*.deb`. No orb and no settings window here: `ira set` configures her, and the screen is at `http://127.0.0.1:8180`.
macOS	`ira--aarch64.dmg`	⚠️ Unverified. It compiles, and that is the whole of what is known about it. See what has not been verified.

**Nothing is signed.** Windows will say "Windows protected your PC" — choose **More info → Run anyway**. macOS will refuse outright: right-click the app and choose **Open**. Both are what an unsigned installer looks like, and both stop once there is a certificate to sign with.

The first start

She downloads the wake models, the voice and piper — about 85 MB — the first time she runs, and local speech-to-text with them: 80 MB on most machines, or about 900 MB where an NVIDIA driver makes the GPU build and a bigger mode