St. Louis · Kansas City

Azure Chat With Your Data: Keeping Your Data in Your Own Microsoft Account

Microsoft retired a popular Azure feature on October 14, 2026, which leaves a simpler question than any setup tutorial answers. Can AI read our files and systems without the data leaving our own cloud account? And if it can, who sees what it did, and who keeps it running?
Azure chat with your data means asking questions in everyday language and getting answers drawn from your own files, databases and business systems, using Azure OpenAI models inside your Microsoft cloud account. Microsoft retired its built-in "On Your Data" feature on October 14, 2026, so today you either build with Microsoft's current tools or have a partner run a managed setup. Either way, the data can stay in your own Azure tenant.
In this article
- What "chat with your data" means in Azure
- Microsoft retired "On Your Data" on October 14, 2026
- What "inside your own Azure account" actually means
- What running in your own account does for control and logging
- Microsoft's do-it-yourself options compared
- Who handles the day-to-day work
- How ThrottleNet's Managed Intelligence does it
- Frequently Asked Questions
- Where to start
What "chat with your data" means in Azure
Chat with your data is a chat interface that answers from your content instead of from general knowledge on the internet. Someone types "Which clients renewed last quarter?" or "What does our travel policy say about mileage?" and the AI finds the answer in your own records.
In a typical setup, two pieces do the work: a language model that reads the question and writes the answer, and a search index of your documents so the model can find the right passages. A web app or a Teams app sits on top as the chat window where people type their questions.
The point for a business owner is where all of that runs. Each of these Azure resources can live inside your company's own Microsoft account, under your name, on your bill.
Microsoft retired "On Your Data" on October 14, 2026
The Azure OpenAI feature most people meant by "chat with your data" is gone. Microsoft's documentation states that Azure OpenAI On Your Data "retired on October 14, 2026, and is no longer supported by Microsoft."
Three facts matter if you built something on it, or planned to:
- Migration isn't automatic. An app that relied on On Your Data has to be moved by someone on your side or by your partner.
- Microsoft names a replacement. For apps that answer from a search index, Microsoft points to Foundry IQ, optionally paired with Foundry Agent Service.
- Ordinary Azure OpenAI chat still works. Chat requests that never used the On Your Data setup are not part of the retirement.
Microsoft's Foundry documentation also now splits into a "classic" portal and a new one. The direction of travel is toward AI agents that can do more than answer from documents. For you, that means any chat with your data solution needs someone watching for changes like this one.
What "inside your own Azure account" actually means
Running inside your own Azure account means the AI services, the search index and any stored conversations sit in your company's Microsoft tenant and subscription, in the geography you choose. Nothing is copied to a third-party vendor's servers. There is more on why your AI should run in your own Azure tenant if you want the longer argument.
Microsoft's own privacy documentation for its Azure-hosted models is specific. Your prompts, completions and embeddings are not available to other customers, are not available to OpenAI or other model providers, and are not used to improve their models or train foundation models without your permission.
Features that store data, such as uploaded files, vector stores and saved conversation threads, keep that data at rest in the Foundry resource inside your Azure tenant, in the same geography. Data stored at rest stays in the geography you designate.
Two honest caveats belong next to those promises:
- Where processing happens depends on the deployment type. With a standard deployment, prompts and responses are processed in your chosen geography, though they may move between regions inside that geography for operational reasons. A Global deployment can process in any geography where the model runs, and a DataZone deployment stays within its data zone.
- Microsoft runs abuse monitoring. When abuse indicators show up, a sample of prompts and responses may be selected for review. The FAQ below covers how that works and how some customers can opt out.
None of this needs a lawyer to read. It does need someone to pick the right deployment type on day one and write the choice down.
What running in your own account does for control and logging
Running in your own account puts the controls in your hands. You decide who gets access, whether detailed logs are kept and where they go.
Access. Access is managed in your own tenant, so your company decides who can use the AI. In ThrottleNet's setup, what the AI can reach follows each person's existing permissions, so a question from someone in sales doesn't surface payroll files they couldn't open anyway.
Logging. Azure keeps platform metrics and an activity log automatically. The detailed resource logs are a different story: Microsoft's monitoring guide says they aren't collected until you create a diagnostic setting and send them somewhere. Logging is something you turn on, and someone has to decide where those logs live.
Read-only. ThrottleNet calls its approach Zero-Write AI. The AI can read connected systems but cannot change them, and that limit is enforced at the connection layer (see the read-only MCP architecture) rather than by a setting someone could flip. ThrottleIQ also keeps a full log of every question asked, so you can see what people asked and when.
Training. On top of Microsoft's own no-training terms, ThrottleNet says the AI model behind ThrottleIQ does not train on customer data.
Microsoft's do-it-yourself options compared
Microsoft offers several routes to chat with your data, and each one suits a different team. The table below sums them up in business terms, with a managed setup alongside for comparison.
| Option | What it is | Who it suits | What your team still does |
|---|---|---|---|
| Microsoft 365 Copilot | Microsoft's packaged AI assistant inside the Microsoft products your staff already use | Companies that want AI built into the apps their staff already open every day | Rolling it out and getting people using it |
| Copilot Studio | Microsoft's low-code tool for building your own agents | Teams that want to build agents without much code | Building the agent, connecting sources, testing answers |
| Microsoft Foundry with Foundry IQ | Microsoft's platform for professional developers who want model choice and a custom back end | Companies with in-house developers or a development partner | Design, code, index upkeep, security, cost watching |
| Chat with your data solution accelerator | An open-source sample application from Microsoft's Azure-Samples on GitHub | Developers who want a starting point to adapt | Deploying it, adapting the code, checking answer quality on your own data |
| Managed setup with ThrottleNet | AI designed and run for you inside your own Azure account | Companies on Microsoft without spare technical staff | Answering questions about your business and using it |
The solution accelerator is the option people most often find first when they search. It's a sample, built from a web app, a backend and an ingestion worker that loads your files into a search index. The repository itself says you need to evaluate answer quality on your own data, and it points to Azure pricing pages instead of giving a cost.
Microsoft's own architecture guidance covers both routes, from Foundry to Copilot Studio for low-code agents.
Who handles the day-to-day work
With any do-it-yourself route, your own team or a developer you hire handles the day-to-day work. With a managed setup, the provider designs it, runs it and keeps it current.
A chat with your data application needs regular care after launch. The jobs that come up most often:
- Loading new files and keeping the search index up to date as documents change.
- Keeping permissions right as people join, move roles or leave.
- Checking that answers are accurate, and fixing the source data when they aren't.
- Watching the Azure bill, since every service deployed adds cost and log routing carries its own charge.
- Keeping up with Microsoft changes, such as the On Your Data retirement and the move to Foundry IQ.
If you have an in-house developer with time to spare, Microsoft's tools are a fair choice. Not every smaller business has that person on staff. What they want is a Monday where someone asks the data a question, gets a straight answer, and nobody has to think about index refreshes.
How ThrottleNet's Managed Intelligence does it
ThrottleNet's Managed Intelligence service runs ThrottleIQ inside your own Azure account, connects your existing systems without moving your data, and handles the ongoing work for you. Your data stays in your Azure tenant.
Connections. The Sync Engine connects to ConnectWise (Manage and Automate), Microsoft 365, Azure and Entra ID, with other connections available on request. It reads from those systems where they sit, with no data migration project first.
Readiness first. Before anyone asks a question, ThrottleNet checks that the data is in good enough shape to give useful answers. Good answers depend on good data, so that work comes first. ThrottleNet can also help you put an AI use policy in place for your staff.
Timeline. The company says a typical rollout takes six to ten weeks.
Background. Managed Intelligence is a ThrottleNet service. The parent company has supported St. Louis businesses as an IT provider since 1999, and it has a Kansas City office as well.
Frequently Asked Questions
Is Azure OpenAI chat with your data still available?
If the "Add your data" option has disappeared from your Azure portal, the retirement is why. For an app built on a search index, Microsoft's migration path to Foundry IQ can reuse the index you already have, so your documents don't need loading again from scratch.
Does Azure OpenAI train on my data or share it with OpenAI?
Microsoft says Azure-hosted models don't interact with services OpenAI runs, such as ChatGPT. The models are also stateless: Microsoft states that no prompts or completions are stored in the model itself.
Where is my data stored, and is it encrypted?
Stored data is encrypted at rest with AES-256 by default, and you can use your own customer-managed key instead, though some preview features may not support it. You can delete stored data at any time.
Can I see what people asked the AI?
Once you create a diagnostic setting, Azure can send resource logs to Azure Monitor Logs through a Log Analytics workspace, a storage account, Event Hubs or a partner system. Routing logs adds cost, so decide what you need to keep before turning everything on. Check what the logs capture before treating them as a full record of questions.
Does Microsoft see my prompts?
Microsoft's privacy page describes abuse monitoring as the case to know about. Flagged samples are reviewed by automated systems first, then by authorized Microsoft reviewers if needed, and stored separately for each customer resource in your geography. Managed customers can apply for modified abuse monitoring; once approved, Microsoft doesn't store those samples or review them by hand, though automated checks may still run.
Is the Azure chat with your data solution accelerator a finished product?
It's released under the MIT license, so you're free to change it, and you can choose Azure AI Search or PostgreSQL to hold your content. Nobody supports it for you the way a commercial product is supported, so plan for your developer to own it after the first deployment.
Where to start
The first step with ThrottleNet is a free written feasibility study. It starts with a 30-minute call about your business and the systems you use. Within five business days you get a written study of what's feasible with your data and systems. The study is yours to keep, whether you go ahead with us, build it yourself on Microsoft's tools, or decide the timing isn't right yet.

