DPDP Act and AI: run LLMs on Indian infrastructure, with personal data kept in India

What the DPDP Act means for AI projects

The Digital Personal Data Protection Act, 2023 applies to digital personal data processed in India, and to processing outside India connected with offering goods or services to people in India. The DPDP Rules were notified in November 2025 with a phased timeline: the Data Protection Board first, with most fiduciary obligations following over the next 18 months.

Four parts of the Act matter most for generative AI:

• Purpose and consent (Sections 4–7): personal data used to ground an LLM needs a lawful purpose. Keeping retrieval inside the system that already holds the consent record makes purpose checks a query, not a vendor review.
• Data Processors (Section 8(2)): the fiduciary stays responsible for processors it engages, and needs a valid contract with each. A hosted embeddings API, vector database and LLM API are three processors.
• Security safeguards (Section 8(5)): reasonable safeguards to prevent a breach, with penalties of up to ₹250 crore for failing to take them. A smaller footprint is easier to secure and to evidence.
• Erasure (Sections 8(7) and 12): data must be erased when the purpose is served or consent is withdrawn, and that includes copies in vector indexes and vendor logs.

Cross-border transfers and data localisation

Section 16 lets the central government restrict transfers to specific countries. Transfers elsewhere are allowed under the Act, but sector rules can be stricter, and the Act defers to them. The DPDP Rules also let the government require Significant Data Fiduciaries to keep specified categories of personal data, and traffic data about their flow, inside India.

Reference setup: an India-only AI stack

SynapCores ships native Linux binaries (x86_64 and aarch64) and a Docker image. It runs on servers in your own data centre or on VMs in any Indian cloud region. The default AI provider loads a quantised GGUF model in-process with llama.cpp, and Ollama works as a drop-in alternative. Nothing in this configuration needs a connection outside the host.

# /etc/synapcores/gateway.toml
# Option 1: in-process GGUF model via llama.cpp (default, fully offline)
[query.ai_service]
provider        = "native"
model           = "llama-3.2-1b-instruct-q4_k_m"
embedding_model = "minilm"

# Option 2: a local Ollama on the same host or network
# [query.ai_service]
# provider = "ollama"
# base_url = "http://localhost:11434"
# model    = "llama3.2"

# Anonymous install telemetry is opt-out; switch it off for air-gapped sites
[telemetry]
enabled = false

RAG on customer data without an overseas API

With a local model, embeddings and semantic search are SQL functions. Customer records, their embeddings and the retrieval query sit in one database, under one set of access controls and one erasure process.

-- Embeddings are computed by the local model, inside the database
CREATE TABLE policies (
  id        INT PRIMARY KEY,
  clause    TEXT,
  embedding VECTOR(384)
);

INSERT INTO policies (id, clause, embedding)
VALUES (1, 'Personal data is processed only within the EU.',
        EMBED('Personal data is processed only within the EU.'));

-- Semantic search: no external embeddings API, no export to a vector SaaS
SELECT id, clause,
       COSINE_SIMILARITY(embedding, EMBED('cross-border transfer')) AS score
FROM policies
ORDER BY score DESC
LIMIT 5;

Where Indian teams use it

Checklist for your DPO and CISO

• Map where personal data goes in the AI design today: embeddings, vector store, LLM, logs, evaluation tools.
• Mark which data falls under sector localisation (RBI, IRDAI, SEBI) or may be specified for Significant Data Fiduciaries.
• Move inference and retrieval onto infrastructure in India and remove the overseas hops.
• Turn telemetry off for regulated sites ([telemetry] enabled = false).
• Keep AI activity logs in India for at least 180 days to meet the CERT-In directions.
• Test erasure: delete a data principal's rows and confirm their embeddings and AI outputs go with them.

DPDP Act and AI: frequently asked questions