Your enterprise AI.
Sovereign by architecture.

No — and this is one of the most important things to understand. The model is already trained on general knowledge by its creator (Microsoft, Google, Meta). Your data plays no part in this training at any point.

What Altern8 does instead is called RAG (Retrieval Augmented Generation). Your documents are indexed into a private vector database on your server. When an employee asks a question, the system retrieves the relevant section and the model reads and answers from it — in real time, on your server.

This means: add a new policy today, it is searchable today. Update a document, the new answer is available immediately. No retraining, no waiting, no GPU costs. And your data never gets permanently baked into any model anywhere.

Every Altern8 deployment includes a live /sovereignty-check endpoint. Call it at any time and it returns a real-time JSON report showing exactly how many external network calls have been made since the server started.

In Sovereign Mode, this number is always zero. You can run this check during an audit, share the output with your compliance team, or demonstrate it live to a regulator.

Additionally every query is logged in a tamper-evident local audit file with SHA-256 hashing — metadata only, never content — giving you a full compliance audit trail that lives on your infrastructure.

No. Ollama is an open-source application with no telemetry and no external calls during inference. The model weights are downloaded once at initial setup from the model provider. After that, Ollama operates with zero internet dependency.

Think of it like downloading Microsoft Word — it connects once to install, then runs entirely locally forever. Once installed, you can block all internet access on the server and Altern8 continues to work identically.

For air-gapped deployments, we pre-download model weights on a connected machine and transfer them via internal network or USB. The customer's server never touches the internet at all.

No. There are no per-token charges, no per-query fees, no per-user fees, and no usage limits. You pay a one-time fixed deployment cost based on your use case. After that, your team can send 10 queries a day or 10,000 — the cost is identical.

The only recurring cost is an optional annual maintenance plan (included free in Year 1, roughly 20% of the deployment cost from Year 2) covering software updates, model upgrades, security patches, and support. If you don't renew, your deployment keeps working — you just stop receiving updates.

At enterprise scale, cloud AI API costs typically run Rs 40 lakh to Rs 4 crore (approx. $48,000 to $480,000) per year. Altern8 replaces that with a one-time fixed cost and unlimited usage. Note that the server, GPU, and on-premise infrastructure that Altern8 runs on is provisioned and paid for by your company — our pricing covers software, deployment, and maintenance only.

All Altern8 services are deployed as permanent systemd services. They start automatically when the server boots and restart automatically if they crash — with no manual intervention required.

The document index, CRM data, and audit logs all persist on disk. If the server restarts, all previously indexed documents are immediately available — nothing needs to be re-uploaded. The system is designed to be genuinely always-on without any human babysitting.

Standard enterprise hardware — no specialised GPU servers required for most deployments. Minimum 8GB RAM, any modern CPU. 16GB RAM recommended for production use with multiple concurrent users.

It runs on Windows Server, Ubuntu Linux, or any modern Linux distribution. Many customers run it on an existing office server or a dedicated mini-PC. For large-scale deployments with hundreds of concurrent users we recommend a GPU-equipped server — we assess your infrastructure during the deployment consultation. All server, GPU, and hardware costs are borne by your company; Altern8 pricing covers software, deployment, and maintenance only.

Yes — by architecture, not by contract. The DPDP Act 2023 requires data fiduciaries to process personal data within India. RBI's data localisation directive requires payment data to stay in India. SEBI's cybersecurity circular requires data not to leave India without approval.

When Altern8 is deployed on-premise in India, all data processing happens within your Indian infrastructure. No data crosses a border. No external API is called.

This is the fundamental difference from using ChatGPT or Copilot for internal work — where every query is processed on servers in the US, a potential violation of all three regulations above.

Yes. Altern8 provides an OpenAI-compatible API — any application or tool that currently works with OpenAI works with Altern8. You change one line of code pointing to your server instead of OpenAI's servers.

Current integrations include Zoho CRM, CSV data uploads from any platform, a REST API, and Python SDK. SharePoint, Salesforce, SAP, and Microsoft 365 integrations are in development.

You cannot run ChatGPT on your own server — OpenAI does not allow this. ChatGPT and GPT-4 always run on OpenAI's servers in the US. There is no on-premise version at any price.

Altern8 uses fully open-source or openly-licensed model weights (Microsoft Phi-3, Google Gemma, Meta LLaMA) that can be legally deployed on your own infrastructure. Beyond the model, Altern8 adds the enterprise layer — document intelligence, CRM integration, sovereignty audit, persistent services, and compliance documentation — that you would otherwise need to build yourself.

For general open-ended tasks, GPT-4 is more capable than the smaller models Altern8 runs by default. We are transparent about this.

However for document-specific questions — "what does our leave policy say?", "what are the capital requirements in Basel III?" — Altern8 is more accurate than GPT-4. GPT-4 answers from training data which may be outdated or wrong. Altern8 reads your actual document and cites the specific section.

For enterprise use cases involving internal documents and proprietary data, specific accuracy from your actual files is more valuable than general capability. And as open-source models improve rapidly, the capability gap continues to narrow.