Chat with Your Data: Azure RAG Solution Accelerator
Chat with Your Data is Microsoft's Azure Samples accelerator for a document-grounded chat assistant, deployed into your own Azure subscription with one `azd up` command. Reach for it if you're on Azure AI Foundry and want retrieval, ingestion, and a citation-aware chat UI wired together already. Skip it if you want a lightweight local prototype or don't want your stack tied to Azure Container Apps and Azure AI Search or PostgreSQL.
The Challenge of Unstructured Knowledge
Contracts, policies, product manuals, and benefit guides pile up as files that are hard to search and slow to answer questions from β the README calls this out directly. Wiring together document ingestion, a retrieval index, a grounded language model, and a chat frontend on Azure is not a five-minute job: it needs a retrieval store, an ingestion worker, an identity model for downstream calls, and a UI that streams cited answers. Chat with Your Data packages that wiring as a deployable template instead of a how-to guide you follow by hand.
What is Chat with Your Data?
Chat with Your Data is an Azure Samples Solution Accelerator that grounds a conversational assistant in your own documents and answers with inline citations. It runs on Azure Container Apps: a React and Vite frontend, a FastAPI backend, and an Azure Functions worker that parses, chunks, and embeds uploaded files. At deploy time, either Azure AI Search, paired with Cosmos DB, or PostgreSQL, which incorporates pgvector, can be selected for the retrieval and chat-history store.
Core Capabilities
- βAnswers stream back with inline citations to the exact source documents they're grounded in.
- βA document ingestion pipeline uploads files or indexes public web pages, then parses, chunks, and embeds them for retrieval, covering many supported file types.
- βRetrieval and persistence is a deploy-time choice: one option involves Azure AI Search in conjunction with Cosmos DB; the other is PostgreSQL, which includes pgvector.
- βTwo interchangeable orchestrators, Agent Framework and LangGraph, share the same retrieval and grounding pipeline and are picked at deploy time.
- βA collapsible reasoning panel streams the model's intermediate steps alongside the final answer.
- βAzure AI Content Safety screens both prompts and responses.
- βA single user-assigned managed identity authorizes every downstream call through Azure RBAC, so there's no Key Vault and no application secrets to rotate.
- βAn admin experience lets you ingest, inspect, and reconfigure the dataset and prompts without touching code, and voice input works through speech-to-text.
Practical Applications
- β’Contract review and summarization: legal and compliance staff query indexed contracts in natural language and get obligations, deadlines, and clauses back with citations.
- β’Employee and HR assistance: indexed policies and handbooks answer employee questions with a citation to the source document instead of a manual policy search.
- β’Customer intelligence: grounding the assistant on customer-facing notes and documents gives customer success managers and analysts instant account context.
Deploying to Azure
Deployment runs through the Azure Developer CLI: everything deploys into your own Azure subscription with one `azd up` command. There's also a one-click 'Deploy to Azure' button that skips cloning the repository, or you can follow the local deployment instructions in docs/LOCAL_DEPLOYMENT.md for a developer workflow. You need an Azure subscription with Contributor role at the subscription level plus a Role Based Access Control role at the subscription or resource group level, and a region where every required service is available β the README lists East US, East US 2, Australia East, UK South, and France Central as examples. After deployment you choose PostgreSQL or Cosmos DB and run a post-deployment setup script from Azure Cloud Shell. Pricing varies by region and usage, so the README points you to the Azure pricing calculator instead of a fixed figure, and warns that resources keep billing until you run `azd down`.
Business Value and Benefits
- βEmployees get grounded, cited answers immediately instead of manually digging through documents.
- βInline citations let readers verify an answer against the source material before acting on it.
- βNon-engineers are given the ability by the admin experience to customize personas, fine-tune system prompts, and revise the knowledge base.
- βSwapping the retrieval backend or orchestrator is a deploy-time choice, not a rewrite, per the README's description of the modular architecture.
- βDeploying into your own Azure subscription keeps the data inside your existing compliance boundary.
Important Considerations and Disclaimers
- β³The README states this is a starting point, not a turnkey production system β it asks you to evaluate retrieval quality, answer accuracy, as well as responsible-AI considerations, all measured against your own data before relying on it.
- β³It only runs on Azure: Azure Container Apps, Azure AI Foundry, and either Azure AI Search or Azure Database for PostgreSQL are all required, so there's no path to running this off Azure.
- β³The bundled sample data is synthetic, generated with Azure OpenAI Service for demonstration only, so it can't tell you how retrieval performs on your real documents.
- β³The README states the software is not subject to SOC 1 or SOC 2 compliance audits and is not a substitute for professional medical or financial advice.
- β³Costs aren't predictable upfront β the README says pricing varies by region and usage and points to the Azure pricing calculator instead of a number.
Related Azure AI Solutions
Common Questions
Chat with Your Data is licensed under the MIT License; the sample dataset in its data/ folder carries a separate CDLA-Permissive-2 License.
Chat with Your Data runs on Azure Container Apps and Azure Container Registry, calls Azure AI Foundry for models and retrieval, and can use Azure AI Search, Azure Document Intelligence, Azure Storage, Azure Functions, Azure Cosmos DB, Azure Database for PostgreSQL, Azure AI Speech, and Azure Monitor depending on configuration.
Chat with Your Data can be deployed without cloning the repository: the README describes a one-click 'Deploy to Azure' button, and cloning is only needed for the local development workflow.
The README describes Chat with Your Data as 'a starting point, not a turnkey production system,' and advises users to assess their own data for retrieval quality, answer accuracy, and responsible-AI considerations before relying on the system.
Chat with Your Data's README says pricing varies by region and usage and can't be predicted exactly, so it points you to the Azure pricing calculator and a per-service pricing table covering Container Apps, AI Foundry, AI Search, and the other resources it deploys.
Chat with Your Data supports voice input through speech-to-text on any supported browser, according to the README's feature list.
Who should try it β and who should skip
Try Chat with Your Data if your team already operates inside Azure, needs a document Q&A assistant with citations, and wants ingestion, retrieval, and a chat UI wired together instead of assembled piece by piece. Skip it if you're not on Azure, want a backend-agnostic RAG stack you can move between clouds, or just need a quick local proof of concept without provisioning Azure AI Foundry, Azure AI Search or PostgreSQL, and Azure Container Apps.