Private AI that never
leaves your building.
A complete AI platform that runs inside your data centre or server room. Your team gets a chat interface, document Q&A, workflow automation, and model management — without a single byte leaving your network.
What you are dealing with today.
Cloud AI is off-limits for sensitive data
Regulatory requirements, classification rules, or internal policy prevent your team from sending documents to external AI services. The result: your people cannot use AI at all.
Setting up open-source LLMs takes months
Downloading models, configuring GPU drivers, building a usable interface, and managing updates requires a dedicated ML engineering team that most organisations do not have.
Shadow AI usage is growing unchecked
Employees use personal ChatGPT accounts to summarise sensitive documents because no internal alternative exists. Data leaks are a matter of time, not probability.
From setup to results.
We assess your infrastructure
Our team evaluates your existing servers or recommends hardware. On-Prem OS runs on standard Linux servers with NVIDIA GPUs. We size the setup based on your user count and workload.
System Check
All passed| Component | Requirement | Your Setup | Status |
|---|---|---|---|
| GPU | NVIDIA A100/H100 | 4× A100 80GB | Exceeds |
| RAM | 256GB minimum | 512GB DDR5 | Exceeds |
| Storage | 2TB NVMe | 4TB NVMe RAID | Exceeds |
| Network | 10GbE | 25GbE | Exceeds |
Platform is deployed on your network
We install and configure On-Prem OS on your servers. Models are loaded, access controls are set, and the system is tested end to end. Typical deployment takes 5 to 10 working days.
Deployment Status
CompletedYour team starts using AI from day one
Users log in through a browser-based chat interface. They can ask questions, search documents, generate summaries, and automate workflows — all processed locally on your hardware.
AI Chat
New ChatSources: HR-Policy-2026.pdf (p.42), Leave-Rules-Amendment.docx
We handle updates and support remotely
New models, security patches, and feature updates are delivered as packages you approve and apply. Our support team is available for troubleshooting via secure remote sessions.
System Health
All healthy| Service | Version | Status | Next Update |
|---|---|---|---|
| Inference Engine | v4.2 | Healthy | Oct 15 |
| Document Pipeline | v2.8 | Healthy | Oct 15 |
| API Gateway | v3.1 | Healthy | Oct 15 |
What you get.
Chat interface with document Q&A
A clean, browser-based chat where users ask questions in plain English. The system searches your internal documents and returns answers with source citations. Supports PDF, Word, Excel, and plain text.
Document Q&A
A: Enterprise refund policy allows full refund within 30 days...
Source: Enterprise-Terms-2026.pdf (p.18)
Semantic search across your document library
Upload thousands of documents. On-Prem OS indexes them using vector embeddings and lets users find information by meaning, not just keywords. Results are ranked by relevance.
...leave encashment of up to 30 days earned leave...
Relevance: 96%
...calculation based on basic pay + DA...
Relevance: 91%
Workflow automation
Build automated workflows that use AI to classify emails, extract data from forms, generate reports, or route requests. No coding required — configure through the admin panel.
Active Workflows
| Workflow | Trigger | Runs Today |
|---|---|---|
| Invoice extraction | Email attachment | 124 |
| Contract review | Upload to /contracts | 18 |
| Meeting summary | Calendar event end | 8 |
Model management dashboard
Run multiple LLMs side by side. Compare outputs, swap models without downtime, and control which teams have access to which models. Supports Llama, Mistral, and other open-weight models.
Running Models
| Model | VRAM | Requests/min |
|---|---|---|
| Llama 3.1 70B | 40GB | 240/min |
| Whisper Large | 3GB | 120/min |
| Embeddings v3 | 2GB | 800/min |
Role-based access and audit logs
Administrators control who can access the platform, which documents are visible to which teams, and what actions are logged. Every query and response is recorded for compliance.
User Roles
| Role | Users | Access Level |
|---|---|---|
| Admin | 3 | Full |
| Manager | 12 | Department |
| User | 33 | Standard |
Runs entirely on your hardware
No cloud dependency. No external API calls. All computation happens on your servers inside your firewall. Works in air-gapped environments with offline model loading.
Data Privacy
Watch a quick demo.
On-Prem OS in 60 seconds.
What it looks like in practice.
Defence research laboratory near Hyderabad
The lab had over 20,000 classified documents across multiple research programmes. Researchers spent hours manually searching through archives. Cloud-based AI tools were prohibited due to security classification requirements.
After deployment, researchers could search and query the full document archive in seconds. Document discovery time dropped from hours to under a minute. The platform now handles over 300 queries per day.
"We went from not being able to use AI at all to having 50 researchers rely on it every day. The fact that nothing leaves our network made approval straightforward." — A system that respects classification boundaries while giving researchers modern tools.Dr. R. Venkataraman, Project Director, Defence Research Lab
Works with your existing tools.
Enterprise-ready from day one.
Deployment options
- Installs on your existing Linux servers with NVIDIA GPUs
- Supports air-gapped networks with offline model loading
- Docker-based deployment for consistency
- Multi-node clustering for large teams
- Typical setup: 5 to 10 working days
Security and compliance
- All data stays on your servers — zero external calls
- Role-based access with Active Directory integration
- Full audit trail of every query and response
- Encrypted data at rest and in transit within your network
- Compatible with ISO 27001 and SOC 2 security frameworks
Clear pricing. No surprises.
Pricing depends on your team size, deployment, and support needs.
Starting price includes up to 50 users. GST extra.
INCLUDES
- Platform licence for up to 50 users
- Chat interface and document Q&A
- Workflow automation engine
- Model management dashboard
- Remote support and quarterly updates
- Deployment assistance
Built for specific needs.
Good fit
- Defence and government research organisations
- Financial institutions with data residency requirements
- Law firms handling confidential case files
- Healthcare organisations with patient data restrictions
- Any enterprise that cannot send data to cloud AI providers
Not designed for
- Small teams under 10 users (cloud AI tools are more cost-effective)
- Organisations without GPU-equipped servers or budget for hardware
- Teams looking for a plug-and-play SaaS chatbot
What hardware do we need to run On-Prem OS?+
A Linux server with at least one NVIDIA GPU (A100, A10, or RTX 4090 class), 64 GB RAM, and 1 TB SSD storage. We can also supply pre-configured hardware through our On-Prem AI Server.
How long does deployment take?+
Typical deployment takes 5 to 10 working days, including model loading, directory integration, and user testing. Air-gapped setups may take an additional week.
Which AI models are supported?+
On-Prem OS supports open-weight models including Llama 3, Mistral, Phi, and Gemma families. We pre-load models suited to your use case and can add new ones as they become available.
Can it work in a fully air-gapped network?+
Yes. Models, updates, and documents are loaded via secure offline transfer. No internet connectivity is required after initial setup.
How is pricing structured?+
Pricing starts at ₹65,000 per month for up to 50 users. Additional users and advanced features are priced based on your requirements. Hardware costs are separate if you use our servers.
Is our data used to train models?+
No. Your data never leaves your servers. Models run locally and are not fine-tuned on your data unless you explicitly request it. There is no telemetry or data collection.
What kind of support is included?+
Remote support via secure sessions, email, and phone during business hours. Quarterly model and platform updates are included. On-site support is available at additional cost.
Can we integrate it with our existing document management system?+
Yes. On-Prem OS connects to SharePoint, network file shares, S3-compatible storage, and databases. Documents are indexed automatically as they are added or updated.
Related guides.
On-premise vs cloud AI: when to keep it in-house
A clear framework for deciding where your AI workloads should run.
Read the guideEnterprise knowledge that stays useful
How to build internal knowledge systems that people actually use.
Read the guideFrom AI prototype to production
What changes when you move from demo to daily use.
Read the guideLooking for pricing and comparisons? See our detailed guide: On Premise Ai Server India
On-Prem OS.
Let's set it up.
Book a 20-minute demo, get a quote, or just ask a question.
Our team in Chennai responds within one working day.