ON-PREMISE AI APPLIANCE · v1.0

Your business AI, in a box
that you own.

Lunara Box is a downloadable on-premise AI appliance. One Docker bundle ships a complete stack — local LLM, Whisper STT, Kokoro TTS, RAG search, omnichannel, outbound dialer, supervisor QA, video avatars and fine-tuning. Install with one command, manage from your cloud dashboard. Every byte of customer data stays inside your perimeter.

Download Lunara BoxSee what's inside

Linux x86_64 · NVIDIA optional · Mini-PC to multi-GPU rack · Air-gapped supported

Lunara Box
on-premise · v1.0
CPU
16+ cores
RAM
64 GB
LAN
1 Gb
● vLLM
● Whisper
● Kokoro
● RAG
● Dialer
● n8n
● MinIO
ONE COMMAND TO INSTALL

Install Lunara Box in 90 seconds.

The installer downloads the appliance bundle (~12 GB), validates your hardware, brings up every service, and prints the pairing code for your cloud dashboard. Works on any modern Linux. Air-gapped install also available.

install.sh
# 1. Run the installer
$ curl -fsSL https://get.lunara.box | sudo bash
# 2. Pair with the cloud dashboard
$ lunara-box pair

Or grab the air-gapped tarball: lunara-box-v1.0.0.tar (~14 GB) for offline sites.

What's inside the box

27+ services, one appliance.

Everything you need to run autonomous AI agents, augmented humans, outbound campaigns and compliance — pre-wired, monitored and updated together.

vLLM (local LLM)

Qwen2.5 / Llama-3 with OpenAI-compatible API. Prefix caching, GPU memory tuning, hot-swap of LoRA adapters.

Faster-Whisper STT

Sub-200 ms streaming transcription, 90+ languages, GPU-accelerated. Silero VAD bundled.

Kokoro TTS

Open-license neural voices with sub-second time-to-first-byte. XTTS profile for voice cloning.

RAG (pgvector + BGE-M3)

Hybrid dense + BM25 search with reranker. Ingests PDF, DOCX, HTML, Markdown.

Outbound Dialer

Bulk campaigns with retries, timezone-aware windows, list segmentation, DNC + opt-out.

Omnichannel & n8n

WhatsApp, Telegram, SMS, email + browser channels (Instagram, LinkedIn, FB Messenger). No-code workflows.

Supervisor & QA

Live whisper / barge-in, real-time risk scoring, LLM-based call scoring after each call.

Compliance & DLP

PII redactor, must-say/must-not-say rules, WORM audit chain in Postgres.

Video Avatars

MuseTalk / SadTalker / Wav2Lip lip-synced personas for video kiosks and meetings.

Fine-Tuning

Build datasets from high-QA calls, train QLoRA adapters on your GPU, hot-load into vLLM.

Keycloak SSO + mTLS

OIDC for users, mutual TLS between services, HSM/KMS-friendly key storage.

MinIO + Postgres + Redis

S3-compatible object storage, primary database with HA Patroni option, Redis Sentinel.

Architecture

Cloud panel up here. The Box down there.

Lunara Box runs entirely inside your perimeter. Your cloud dashboard talks to the Box only over signed control messages — no transcripts, no audio, no PII ever leaves the customer's network.

Cloud panel · lunara.app

Management dashboard

  • · Status & health of every service
  • · Integration & flow management
  • · Campaign control, RAG ingestion
  • · QA scores, sentiment, supervisor view
  • · LoRA training & avatar rendering jobs
HMAC-SHA256 · mTLS
Only signed control messages travel this line. Zero PII, zero audio, zero transcripts.
Customer site · Lunara Box

Your hardware

  • · vLLM, Whisper, Kokoro, RAG, n8n
  • · Postgres + pgvector, Redis, MinIO
  • · LiveKit + SIP for telephony
  • · Browser-omni, dialer, qa-scorer
  • · Prometheus, Grafana, Loki, Caddy

Capabilities

Built for real businesses — from clinics to banks.

Support & Sales agents (RAG)

  • Per-agent knowledge base from PDF/DOCX/HTML.
  • Hybrid search + reranker injects top matches per turn.
  • Source citations stored next to every reply for audit.

Massive outbound campaigns

  • Upload a CSV, pick a voice, launch — that's it.
  • Per-campaign pacing, retries, DNC, timezone windows.
  • Live progress, success-rate, and per-call recordings.

Banking & enterprise compliance

  • Voice biometrics + eKYC + AML guardrails.
  • PII redaction sidecar; WORM audit trail.
  • HSM/KMS-ready keys; SSO via Keycloak.

No-code automation (n8n)

  • Pre-built templates: 1C, Bitrix24, Jira, HubSpot.
  • Operators connect new tools without code.
  • LLM auto-discovers integrations as callable tools.

Data sovereignty

Nothing ever leaves the customer's perimeter.

Models, transcripts, recordings, embeddings — all stored, processed and indexed inside the Box. The cloud dashboard never sees a single byte of customer data. Pass your DPIA, GDPR, AML and PCI audits with the architecture, not paperwork.

100% local inference
vLLM, Whisper, Kokoro, BGE-M3 — all run on the Box GPU/CPU.
Zero egress for data
Only signed HMAC control messages travel out of the perimeter.
Air-gapped support
Restore from a single tarball. No internet required at runtime.
Hot updates
Pull new images and migrate with one make command. Rollback is one flag.

Ready to own your AI?

Download Lunara Box, sign in to your cloud dashboard, and ship customer-facing AI this week — entirely on your hardware.

Download Lunara BoxBook a live demo