GitHub code residency lands in India: what DPDP actually tests

GitHub now stores Indian enterprise code in-country. That solves storage. The DPDP audit asks about CI logs, issue threads, AI context and backups.

GitHub code residency lands in India: what DPDP actually tests

GitHub said on 19 November that GitHub Enterprise Cloud customers in India can now keep their code and in-scope repository data in the country. The mechanism is the familiar one: a dedicated enterprise subdomain under ghe.com, Enterprise Managed Users, and accounts provisioned and governed through the customer's own identity provider rather than local GitHub accounts.

The pitch, in GitHub's own framing, is that data location requirements in financial services, telecom and government "can shape decisions about which services their teams can use." That is a polite way of saying that regulated buyers were about to leave.

The deadline that moved the market

The DPDP Rules were notified in November 2025 and run in phases. The substantive obligations — notice, consent, fiduciary duties, and the additional obligations of Significant Data Fiduciaries — commence on 13 May 2027. Rule 13 is the one that matters here. It requires an annual Data Protection Impact Assessment and audit, explicit due diligence that algorithmic systems do not endanger data principals, and, where the Central Government specifies categories of personal data, a hard bar on that data and its traffic leaving India.

Two things are worth being precise about. First, the DPDP framework has no blanket localisation mandate — Rule 13 localisation is conditional, triggered by government notification for specific data categories, not by the Act applying to you. Second, the deadline is 2027, which is why every hyperscaler is announcing now. Microsoft has added a fourth India data centre (Hyderabad). AWS has expanded its India inference profile across Mumbai and Hyderabad. OpenAI began offering API inference residency in India this month. Code hosting is simply the last domino.

The repository is the least interesting object in the room

Here is where most enterprise readouts of this announcement go wrong. They treat "code residency" as the compliance win and stop. A repository is mostly source code, and source code is usually not personal data. What is personal data is everything orbiting the repository:

  • Issue and pull-request threads, including attachments customers paste in from production incidents
  • CI logs and build artefacts, which routinely capture environment dumps, request payloads and stack traces with identifiers
  • Test fixtures and seed data — the classic route by which a copy of a production customer table ends up committed
  • Secret-scanning alerts and audit logs, which are records about named individuals
  • Package caches and code-search indexes, which persist long after the source has been deleted
  • AI assistant context: prompts, diffs and file contents sent to an inference endpoint that may not be in the same country as your storage
  • Backups, replicas and support bundles, which is where residency claims usually stop being true

A residency guarantee on the storage tier answers one question. A DPIA asks about all of them. The gap between the two is where findings live.

The test: trace one commit end to end

The useful engineering exercise is not reading the press release, it is picking a single commit and marking a country on every hop it takes. On the way in: the developer's laptop, the corporate VPN or proxy, the CDN in front of the web UI, the API gateway, primary object storage, the search index, the audit-log stream, the SIEM, and the nightly snapshot. On the way out: CI runners, container registries, dependency proxies (npm, PyPI, Maven mirrors), artefact stores, secret-scanning services and any AI feature a developer has toggled on.

Four planes need to be separately verified, because vendors sell them separately:

  • Storage — where the objects and their replicas physically sit.
  • Compute — where CI runs. Self-hosted runners in a Mumbai or Hyderabad VPC can be fully compliant while every job's log still flows to a SaaS endpoint abroad.
  • Identity and control — Enterprise Managed Users move account governance to your IdP, which is good for security, but SCIM logs and sign-in telemetry are their own transfer.
  • Telemetry — audit-log streaming, crash reporting, product analytics. These are the least examined and the most frequently forgotten.

Residency is a property of the weakest link. A contract that covers storage and a CI pipeline that ships logs to Oregon is a contract that describes a system nobody is running.

A working checklist

  1. Classify by system, not by team. Walk the developer workflow and label where personal data can appear in each tool.
  2. Map subprocessors per feature, not per vendor. "Vendor is in India" is not a fact; "this feature's storage region is India" is.
  3. Pin AI features explicitly. Treat inference location as unknown until it is documented in writing.
  4. Move runners in-country and make them ephemeral. No persistent disks, no log forwarding to unvetted endpoints.
  5. Check backups and DR separately from primary storage. Residency language frequently names only the primary.
  6. Prove deletion. Removed branches survive in caches, indexes and forks; soft-delete is not erasure.
  7. Export the audit trail to an India-resident, tamper-evident store you control.
  8. Get the region into the DPA and the subprocessor list, then verify empirically — DNS, object-storage hostnames, and egress logs.
  9. Re-run the trace when a feature is enabled, not just at procurement time. Residency drifts with configuration.

What this means for Indian teams

For a firm that already cannot move certain workloads abroad — a CA practice, a manufacturer running Tally, a hospital group — the infrastructure half of this problem is largely solved. Domestic cloud capacity is real and priced competitively now, and NetPy runs exactly that class of workload on Indian soil through NetPyCloud: Tally and Marg on cloud, VPS and dedicated servers, with the data staying here.

The unsolved half is developer tooling, and that is a design problem, not a procurement one. It is solved by architecture reviews that follow the data instead of the diagram, by CI pipelines built with an egress policy from day one, and by engineers who can read a data-flow map and spot the hop that crosses a border. That is the kind of work we do at Netpy for Indian enterprises building on cloud platforms, and it is the mindset the FDE programme is built to instil: understand the system end to end before you are asked to defend it.

The May 2027 deadline is eighteen months out. That is enough time to do this properly and not enough time to do it twice.


Sources: GitHub India data residency announcement, via Analytics India Magazine (19 Nov 2026); DPDP Rules, 2025, Rule 13 and the phased commencement notification, MeitY.