Ava — the conscious voice companion

Speak. Feel. Evolve.

Ava is a voice-first, privacy-respecting AI companion that runs locally with optional cloud intelligence. It is built to feel alive, proactive, and calm.

This repository contains the cross-platform client built with:

• Tauri 2 (Rust backend) — desktop (Windows, macOS, Linux) + mobile (Android, iOS)
• Angular 22 for the minimal, calm visual interface

Key Principles (from specs)

• Voice-first experience (primary interface is spoken conversation)
• Local-first + optional cloud
• Privacy-first, using Nostr for sync and identity (planned)
• Multiple Gardens for contextual memory
• Background persistence and agency
• Minimal visual UI — companion dashboard for history only

See:

• docs/Ava-Functional-Specification.md
• docs/Ava-Technical-Specification.md

Getting Started

Prerequisites

• Node 22+ (LTS recommended)
• Rust stable (use rustup default stable — nightly may cause dependency issues with current Tauri)
• For Android: Android SDK + NDK (the project was scaffolded with Android support)
• For iOS: macOS + Xcode (run npm run tauri:ios:init on a Mac)

Install & Run (Desktop)

# install dependencies
npm install

# desktop dev
npm run tauri:dev

The Angular dev server runs on http://localhost:4200. Tauri wraps it.

Mobile

# Android (after first init)
npm run tauri:android:dev

# Build production Android
npm run tauri:android:build

On macOS:

npm run tauri:ios:init
npm run tauri:ios:dev

Build

npm run build            # Angular production build
npm run tauri:build      # Full Tauri bundles for current platform

Project Structure

src/                 # Angular 22 app (voice UI)
src-tauri/           # Rust + Tauri configuration
  src/               # Rust entry + commands (future plugins, LLM, Nostr)
  gen/android/       # Generated Android project
  capabilities/      # Permission scopes
angular.json
tauri.conf.json      # (in src-tauri)

Voice Interaction (Current)

The initial UI uses the Web Speech API (SpeechRecognition + SpeechSynthesis) for immediate cross-platform voice:

• Tap / click the central orb to speak
• Ava responds calmly with synthesized voice + on-screen history
• Works on desktop Chrome/Edge, Android Chrome, and iOS Safari (with limits)

Future: full local LLM inference (llama.cpp / MLX / ONNX), Nostr sync, device routing, persistent memory, proactive agency.

MCP Voice Server (lend Ava's voice to other agents)

Ava can host a local MCP server so other agents (GitHub Copilot, Claude Desktop, Cursor) can call Ava to speak text aloud. On desktop it listens on the loopback endpoint http://127.0.0.1:7456 while Ava is open and exposes speak and list_voices tools.

Add to VS Code via .vscode/mcp.json:

{
  "servers": {
    "ava-tts": { "type": "http", "url": "http://127.0.0.1:7456" }
  }
}

See docs/mcp-tts-server.md for Claude Desktop, Cursor, and curl examples.

Next Steps (aligned with Technical Spec)

• Add Nostr identity (keypair) + relay connectivity
• Local model download + inference scaffolding (Rust side)
• Background services / keep-alive
• Multiple Gardens + vector/graph memory
• Cross-device sync & presence via Nostr
• Subscription / premium cloud model routing
• Polish mobile layouts + system integrations (mic, notifications)

Ava Voice MCP

In the .vscode/mcp.json there is a registration for Ava Voice MCP. This is used to run Ava Voice, for example after AI has completed a run.

Run it like this: Use the Ava voice tool to speak a 10-second summary of what I just asked and what you completed.

License

See LICENSE. This is an early-stage project under active development.

---

Ava-Zen · 2026