# Microsoft Fabric Data Agent — A governed, natural-language Q&A agent built into Microsoft Fabric — turns plain-English questions into SQL, DAX or KQL over your own lakehouse, warehouse and Power BI data, read-only and permission-scoped.

> Source: The Agents Index — https://theagentsindex.com/microsoft-fabric-data-agent (structured, researched, re-verified)
> Facts last verified: 2026-08-01

Fabric data agent is a generally-available feature inside Microsoft Fabric that lets anyone ask plain-English questions over an organization's own lakehouse, warehouse, Power BI and Eventhouse data and get back a governed, read-only answer — no SQL, DAX or KQL required. Configured similarly to publishing a Power BI report, it runs on Azure OpenAI's Assistant APIs, translating each question into the right query language for whichever connected source actually holds the answer, then enforces the requesting user's own permissions and any Microsoft Purview governance policy before returning a result. It ships as part of an existing Fabric or Power BI Premium capacity, not as a separate product.

| Fact | Value |
| --- | --- |
| Website | https://learn.microsoft.com/en-us/fabric/data-science/concept-data-agent |
| Pricing | ~$0.36/hr / pay-as-you-go (~$263/mo always-on); ~$156/mo on a 1-year reserved commitment — third-party estimate, since Azure's own calculator requires sign-in for a quote |
| API | Yes |
| Best for | Organizations already running Microsoft Fabric or Power BI Premium capacity who want employees asking plain-English questions over governed lakehouse/warehouse/BI data without writing SQL, DAX or KQL themselves — and developers who want to embed that same governed data access into a larger agent system via Azure AI Foundry, Copilot Studio or the new public API. |
| Not for | Teams without an existing Fabric or Power BI Premium capacity investment (there is no standalone or free way to try it), anyone needing to query unstructured documents or non-English content, or buyers who want to choose or fine-tune their own underlying model. |

## Pricing

| Tier | Price |
| --- | --- |
| Fabric F2 capacity (minimum required) | ~$0.36/hr / pay-as-you-go (~$263/mo always-on); ~$156/mo on a 1-year reserved commitment — third-party estimate, since Azure's own calculator requires sign-in for a quote |
| Power BI Premium P1+ (alternative path) | Existing capacity |

## Verdict

Fabric data agent is the one listing in this index whose "agent" ships embedded inside a full enterprise data platform rather than as a standalone product to buy — and that changes what it's actually good at. Its real strength is governance you don't have to build yourself: every query runs on the requesting user's own credentials, respects Power BI Row-/Column-Level Security, and is automatically subject to Microsoft Purview DLP and access-restriction policies, while purpose-built NL2SQL/NL2DAX/NL2KQL routing lets one configured agent span up to five structured sources at once. A public REST API, Python SDK and MCP-based invocation from Copilot Studio and Azure AI Foundry — all made generally available in 2026 — mean it's no longer just a Fabric-portal chatbot; it can be a governed data tool inside a larger multi-agent system. The honest limits are real, though: you cannot change or fine-tune the underlying model, it only reads structured data (no PDFs, no non-English questions), responses are capped at 25 rows and 25 columns, and there is no way to try it without an existing paid Fabric or Power BI Premium capacity — independent estimates put the minimum entry (an always-on F2) at roughly $263/month regardless of whether the data agent itself is ever queried. Pick it if your organization already runs on Fabric or Power BI Premium and wants a governed, natural-language layer over that data without buying a separate product; look elsewhere (or at a general enterprise search platform like Glean) if you need to search unstructured documents, swap models, or don't already have Fabric capacity to run it on.

## Strengths and weaknesses

- ✓ Every query runs using the requesting user's OWN credentials and is automatically subject to Power BI Row-/Column-Level Security plus Microsoft Purview DLP and access-restriction policies — governance is inherited from the platform, not a separate layer to configure.
- ✓ Routes each question to a purpose-built query generator per source type — NL2SQL for lakehouses/warehouses, NL2DAX for Power BI semantic models, NL2KQL for Eventhouse/KQL databases, plus Microsoft Graph queries — and can combine up to 5 sources in one configured agent.
- ✓ A public REST API and Python SDK for creating, configuring and publishing data agents went generally available in 2026, alongside an OpenAI-compatible runtime endpoint and MCP-based invocation from Copilot Studio and Azure AI Foundry — real programmatic and multi-agent embedding, not just a portal chatbot.
- ✓ Configuration is Git-integrated with deployment pipelines and built-in diagnostics, so a data agent can be version-controlled and promoted dev-to-production like a real software artifact rather than a one-off chatbot setup.
- ✗ The underlying LLM cannot be changed — Microsoft's own documentation states this directly. It runs on Azure OpenAI's Assistant APIs, with independent write-ups pointing to a GPT-4-series model — a fixed choice where several rivals in this index let you swap or bring your own model.
- ✗ No support for unstructured data at all (no PDF, DOCX or TXT) and no non-English language support — a narrower scope than a general enterprise knowledge/search agent.
- ✗ Not a standalone purchase — it requires an existing paid Fabric F2-or-higher capacity or Power BI Premium P1-or-higher capacity. Independent pricing analyses put the cheapest entry (an always-on F2) at roughly $263/month before any other Fabric workload cost; Microsoft's own pricing page lists only SKU names with no listed price.
- ✗ Responses are capped and summarized to at most 25 rows and 25 columns, and conversation history is not guaranteed to persist across backend or model updates — built for a conversational insight, not for pulling a complete dataset.
- ⚠ No support for unstructured data (PDF, DOCX, TXT) and no non-English language support.
- ⚠ The underlying LLM cannot be changed or swapped.

## Key features

- **Multi-source natural-language query routing** — Combines up to 5 sources per agent (lakehouse, warehouse, Power BI semantic model, KQL/Eventhouse, ontology, Microsoft Graph) and routes each question to the right generator — NL2SQL, NL2DAX, NL2KQL, or a Graph query.
- **Inherited governance, not bolted on** — Every query runs using the requesting user's own credentials, respects Power BI Row-/Column-Level Security, and is subject to Microsoft Purview DLP and access-restriction policies automatically.
- **Configurable via instructions + example queries** — Up to 100 example question/query pairs per source, plus free-text instructions, steer the agent toward an organization's own vocabulary and preferred source for a topic.
- **Public REST API, Python SDK and MCP-style runtime endpoint** — A public management-plane API and Python SDK (made generally available in 2026) let developers create, configure and publish data agents in code; a runtime OpenAI-compatible endpoint and MCP-based invocation from Copilot Studio/Azure AI Foundry open it to external orchestrators.
- **Git-backed ALM** — Data agent configurations can be version-controlled via Git integration and promoted dev-to-production through Fabric deployment pipelines, with built-in diagnostics for troubleshooting query generation.

## Use cases

- **Self-serve BI Q&A** — A regional sales manager asks "which products underperformed in EMEA last quarter?" in plain English and gets a governed answer pulled live from the warehouse, without filing a ticket with the data team.
- **Embedding governed data access into a larger agent** — A platform team building a multi-agent system in Azure AI Foundry adds the company's Fabric data agent as a tool, so any orchestrated agent can pull governed financial or operational data without a separate, ungoverned integration.
- **Programmatic querying via the public API** — A developer uses the Fabric data agent Python SDK to create and publish a new agent as part of a CI/CD pipeline, rather than clicking through the Fabric portal by hand.

## Integrations

Lakehouses, Warehouses, Power BI semantic models, KQL/Eventhouse databases, ontologies, Microsoft Graph (up to 5 combined per agent) · Microsoft 365 Copilot, Copilot Studio, Azure AI Foundry, Microsoft Teams · Microsoft Purview (DLP, access-restriction policies, audit/eDiscovery) · Public REST API + Python SDK (management plane); OpenAI-compatible runtime endpoint

## Sources

- https://learn.microsoft.com/en-us/fabric/data-science/concept-data-agent
- https://community.fabric.microsoft.com/t5/Fabric-Updates-Blog/Fabric-data-agent-API-is-now-public-Build-Fabric-data-agents/ba-p/5230588
- https://learn.microsoft.com/en-us/fabric/data-science/fabric-data-agent-sdk
- https://azure.microsoft.com/en-us/pricing/details/microsoft-fabric/
- https://dataroots.io/blog/a-closer-look-at-microsoft-fabric-pricing-billing-and-autoscaling
- https://medium.com/data-science-at-microsoft/one-agent-many-channels-how-to-query-fabric-data-agents-everywhere-7e85f453f6e6