GitHub - vignesh2027/Vilvona-AI: Self-hosted AI assistant powered by Claude Fable 5. Tamil & Hindi UI, Mobile PWA, 100+ integrations. npx vilvona

What is this?

I built Vilvona AI because I wanted a personal AI assistant that could connect to WhatsApp, Telegram, Gmail, Slack, and everything else I use — without sending my data to some cloud service.

It runs entirely on your own machine. You bring your Anthropic API key, and Vilvona handles the rest. Your conversations stay local. You pay Anthropic directly per token — no markup, no middleman.

The two things that make it different from everything else I found:

• Tamil and Hindi — I'm from Chennai. I wanted an AI assistant that actually works in Tamil, not just pretends to. Vilvona has a full Tamil and Hindi interface built in.
• Claude Fable 5 by default — Anthropic's most capable model, set as the default. 1 million token context. Computer use. The best reasoning available right now.

Start in 30 seconds

export ANTHROPIC_API_KEY=sk-ant-...
npx vilvona

Open http://localhost:4000. Done.

Get an API key at console.anthropic.com — it's free to sign up, you only pay for what you use.

Install

# npx — no install, always latest
npx vilvona

# yarn
yarn dlx vilvona

# pnpm
pnpm dlx vilvona

# global
npm install -g vilvona

Windows (PowerShell):

$env:ANTHROPIC_API_KEY = "sk-ant-..."
npx vilvona

Docker:

docker run -e ANTHROPIC_API_KEY=sk-ant-... -p 4000:4000 ghcr.io/vignesh2027/vilvona-ai

Requires Node.js 22+. If you have nvm: nvm install 22 && nvm use 22

What it connects to

100+ integrations inherited from OpenClaw. You connect them from the Settings panel in the UI.

Languages

The interface supports 21 languages. Tamil and Hindi are built in by Vilvona — no other AI assistant I found actually has these as first-class options.

Set the language in your config:

{ "ui": { "locale": "ta" } }

AI Models

export VILVONA_DEFAULT_MODEL=fable    # Claude Fable 5 — default, best reasoning
export VILVONA_DEFAULT_MODEL=opus     # Claude Opus 4.8 — balanced
export VILVONA_DEFAULT_MODEL=sonnet   # Claude Sonnet 4.6 — fast

Or set it in ~/.openclaw/config.json:

{ "defaults": { "model": "anthropic/claude-fable-5" } }

Mobile

Vilvona works as a mobile app without any app store. On your phone, open http://your-ip:4000 and tap "Add to Home Screen." It installs like a native app on both iOS and Android.

Configuration

Config file: ~/.openclaw/config.json

Pro

The free version has everything for personal use. Pro adds features for teams and businesses.

Get Vilvona Pro — one-time purchase, lifetime access

After purchase, activate with:

export VILVONA_PRO_KEY=VLVN-XXXX-XXXX-XXXX-XXXX
vilvona

How it works under the hood

npx vilvona
  └── openclaw.mjs checks Node.js 22+
      └── downloads openclaw runtime → ~/.vilvona/runtime/ (first run, ~97MB)
          └── sets OPENCLAW_DEFAULT_MODEL=anthropic/claude-fable-5
              └── starts web UI at localhost:4000

The first run downloads OpenClaw's runtime bundle once. After that, everything starts in under a second from cache.

Releases

Releases are automatic. Pushing a v* tag runs the CI checks and publishes to npm + creates a GitHub release.

# bump version and release
npm version patch
git push origin main --follow-tags

License

MIT. Built on OpenClaw (also MIT). Attribution in UPSTREAM.md.