Leave Azure. Keep your code.
One command reads your Azure resource group, prices the equivalent Drift slice, and pulls your code and data into a portable folder you own. Free. Read-only. Honest about the 20% it can't move.
From 12 January 2027, the EU Data Act makes leaving your cloud a legal right. This is how you exercise it.
Three commands. Nothing hidden.
Every Azure call is your own az login, printed before it runs. Drift never holds your credentials and never receives your data: it flows from Azure to your laptop, full stop. The whole migration is free; only hosting on Drift costs, like any Drift slice.
# 1. read-only: what does this cost on Azure vs. Drift?
drift migrate azure estimate -g my-resource-group
# 2. read-only: pull code + data into a vendor-neutral folder
drift migrate azure snapshot -g my-resource-group -o ./azure_export
# 3. offline: rewrite it into a deployable Drift project
drift migrate azure transform -i ./azure_export -o ./drift_workspace
# then it's just a normal Drift project: see what it costs, then apply it
cd drift_workspace && drift plan
drift applyPrerequisites. The Azure CLI (az), logged in, plus mongoexport if your group has a Cosmos account, and squashfs-tools to unpack Linux Consumption source. snapshot checks these against your resource group before it does any work and fails with the exact install command if one is missing. No silent half-exports.
What it pulls out: your data, in formats you own
A plain folder of JSONL, JSON and source, with no Drift-specific binary anywhere. You can read it, diff it, hand it to a lawyer, or take it elsewhere. Verified end-to-end against a live Azure tenant.
| From Azure | You get |
|---|---|
| Your handler source Functions, Atomic | Python and Node Function Apps, including the Linux Consumption case where the code is a SquashFS package in storage, not a downloadable zip. |
| Documents to JSONL Cosmos, Backbone | Cosmos DB (Mongo API) collections exported document for document, ids preserved. |
| Blobs and queues Storage, Backbone | Blob containers downloaded byte for byte, nested paths intact. Queue messages are peeked, never dequeued: your live queue is untouched. |
| Static websites $web, Canvas | A Storage static website ($web) becomes a Canvas site, content and all. |
| App settings to secrets Config, Backbone | Captured as secret names by default; values only with an explicit flag, written 0600 with a loud warning. |
| A deployable Driftfile Offline, deterministic | transform emits a Drift project and scaffolds and validates the Driftfile against the real parser before writing it. Same input, identical output. |
What it will not move
Each of these is a line in REFUSED.md with a stable code and a reason: on the record, never faked. The first refusal is the tool working, not failing.
- .NET, Java, PowerShell function apps. No Drift SDK for these runtimes: they stay on Azure (or a Studio rewrite).
- Durable Functions, Service Bus topics and sessions, Event Grid and Event Hub, blob and Cosmos triggers. No Drift equivalent: Drift's primitives are small and honest, never invented.
- AAD and Easy Auth identity. A Drift Function is gated by an API key or not at all; per-tenant identity is a security-semantics change, not a flatten.
- Cosmos SQL, Gremlin and Cassandra, Table Storage, Azure SQL. Mongo-API Cosmos moves today; the rest are refused or on the roadmap.
- Static Web App (managed) content. Its build-pipeline output isn't retrievable from Azure: rebuild from your repo into a folder and declare it with a
siteblock. ($webstorage sites are captured.)
Where you stay in the loop
We automate most of the lift. These are the steps that are yours, stated up front and never papered over.
- If source retrieval is blocked (hardened SCM, private endpoints), pass
--source <app>=<dir>: you already have your code. Works on any plan type. - Two prerequisite tools, checked up front:
mongoexportfor a Cosmos account,squashfs-toolsfor Linux Consumption source. The CLI fails loudly with the exact install command if either is missing. They are as required asazitself, never a silent half-export. - Queues with a live consumer drain themselves. Pause the consumer before you snapshot, or you'll capture an empty queue.
- A snapshot is "approximately now," not a frozen instant. Great for an estimate or a dry run; don't cut a live system over on one until point-in-time replay lands.
- Timer triggers are captured but not mapped: a Drift schedule is the
cronattribute of a Function, so they're refused with guidance and the source stays in the snapshot for a hand-port, never silently mis-billed. - Memory bookings are a starting point.
transformbooks 128MiB for each function and sizes each collection at twice its data (5MiB at least), because a Driftfile states every size and nothing here has been measured yet. Rundrift planto see what that costs, anddrift sizeafter the first deploy to write what each function actually holds. - Read
REFUSED.mdandREPORT.mdbefore you deploy. The hard 20% (AAD, .NET rewrites, query translation) is where Drift Studio helps.
Built for the Data Act
From 12 January 2027, EU Regulation 2023/2854 makes switching cloud providers a right: your provider must hand your data over in structured, machine-readable formats, and can no longer charge you to leave. This tool is the front door to that right. And because "your data is yours" has to work both ways, a Drift-native app runs on your own metal with one command, drift file run. We'll help you off Azure for free, and we can't trap you either.
Start with Getting started, or read the source.
The tool never changes your Azure tenant
--teardown-azure. Decommissioning Azure is your call, in your portal, after you have verified the slice.