Private LLM deployment in India.
Run an open AI model on infrastructure you control, with search over your own documents. We build it inside your accounts, with the guardrails, logs and evidence that DPDP and sector reviews ask for.
What we build
What DPDP means for a private model.
A private model helps, but it is not a shortcut. Our guide to where prompts, embeddings and logs live walks through each store of personal data in an AI feature.
It does not require Indian hosting
The DPDP Act does not say personal data must stay in India. Section 16 lets the government restrict transfers to countries it notifies. As of November 2026 none are notified. Section 16(2) keeps stricter laws in force, such as sector rules on where payment data is stored.
It still applies to a private model
Prompts, documents, search indexes, logs and caches can all hold personal data. Running the model yourself does not remove your duties as the Data Fiduciary. Most core duties start on 13 May 2027.
Processors stay your responsibility
If a provider runs the model for you, it is your Data Processor. You stay responsible for the data (section 8(1)) and you need a contract with it (section 8(2)).
Retention and erasure need a design
Section 8(7) asks you to erase personal data once consent is withdrawn or the specified purpose is no longer served, whichever is earlier, and to make your Data Processor erase it too, unless a law requires you to keep it. The Rules also ask for certain logs to be kept for at least one year. A private LLM needs both, planned together.
Where it runs, in one paragraph.
A private model can run in an Indian cloud region, on an Indian GPU cloud or on your own servers. If every part, including backups and support access, stays in India, personal data does not leave the country, so a future Section 16 restriction would not reach it, and sector rules that ask for data to stay in India are easier to meet. The right choice depends on your data, your regulator and your team. Our Own Your AI page shows the options, and this guide compares private LLM and API costs in rupees.
By sector, what we build.
The same platform fits each sector, with different rules around it. For lenders, read how RBI’s draft model risk guidance applies to generative AI, and see our DPDP work for BFSI.
NBFCs, lenders and banks’ tech vendors
What drives itCredit files, KYC records and customer chats that you would rather not send to a public AI service. RBI’s model risk guidance is still a draft, so we build for a model inventory, testing, human review and logs, which are sound practice whichever way the final text lands.
What we buildDocument analysis, a private assistant for operations and a model register, with every request logged.
Insurers
What drives itClaims files with medical and financial details, and long documents that staff read again and again.
What we buildClaims summaries and policy search, with a person approving anything that affects a claim.
Hospitals and healthcare
What drives itPatient records and clinical notes are personal data. We build for tight access and a clear record of use. We don’t give clinical advice.
What we buildDrafting of notes and discharge summaries for a clinician to check, on infrastructure you control.
CA firms and law firms
What drives itClient confidentiality, and matter files that must not mix between clients. Your professional rules and your counsel decide what is allowed.
What we buildSearch and drafting over each matter’s documents, with separate access per client or matter.
Buyers who ask for “sovereign AI”
What drives itThe phrase means different things to different buyers: where data sits, who runs the model, who can get access, or which law applies.
What we buildA short written answer to each of those four questions, backed by how the system is set up.
How to evaluate a private LLM provider.
Ask these of any provider, Mono included. The answers should be in writing, and they should match how the system is built.
Where does each part run?
Model, search index, logs, backups and any support access, with the region for each.
Whose account is it in?
Yours, or the provider’s. Ask who holds the keys and who can switch it off.
What does the provider see?
Prompts, outputs and logs, and under which contract.
Which model, under what licence?
Check the licence allows your use, and who updates the model later.
How is it tested on your documents?
Results on your own samples, not a vendor demo.
How do you erase data?
From prompts, indexes, logs, caches and backups, and how you can prove it.
What happens when we leave?
Export, handover and deletion, in writing.
How it runs.
Phases, not promises of dates. We agree the plan with you once we know your data and your systems.
- Step 1
Scope
We agree the use cases, the data involved and the rules that apply to you.
- Step 2
Design
We choose where it runs, which model fits, and how it signs in and logs.
- Step 3
Build
We set up the model, the search and the guardrails inside your accounts.
- Step 4
Test
We check answers on your own documents, with a small group from your team.
- Step 5
Hand over
Documentation, training and the evidence file, with everything running in your accounts.
Related services.
DPDP compliance
Notice, consent, rights, retention and security, set up in your systems.
See DPDP implementationRBI and FREE-AI readiness
Governance and evidence for lenders and fintechs using AI.
See RBI readinessISO 42001 readiness
A management system for AI that buyers and auditors recognise. Mono prepares you; it does not certify.
See ISO 42001 readinessPrivate AI in India FAQs.
Does DPDP require our AI to run in India?
No. The Act does not require personal data to stay in India. Section 16 lets the government restrict transfers to notified countries, and as of November 2026 none are notified. Sector rules can be stricter, so check yours. Hosting in India avoids future Section 16 restrictions and makes sector localisation rules easier to meet.
Is a private LLM automatically DPDP compliant?
No. It gives you control over where data goes, which helps. You still need notice, consent where it applies, a way to handle rights requests, security safeguards, retention and erasure. We build the engineering and the evidence. We don’t certify compliance, and there is no government DPDP certificate.
Has RBI issued final rules on AI for lenders?
As of November 2026, RBI has not issued final model risk management guidance. The June 2026 draft is the latest text. The FREE-AI report from August 2025 is a committee report, not a binding direction. We build to the themes in both and update when the final text arrives.
Which models do you use?
Open models such as Llama, Qwen, Mistral, Gemma and DeepSeek, chosen by task, language, licence and how they perform on your own documents. We check each licence against your use before we build.
Is “sovereign AI” a certification you can give us?
No. It is a procurement term, not a certificate. We help you answer what a buyer usually means by it: where the data sits, who runs the model, who has access, and which law applies.
Is this legal advice?
No. We work on the engineering and the evidence: systems, settings, logs and documents. Contract terms and legal questions go to your counsel, or to our partner law firm if you want us to involve them.