What is the difference between oneAgent and the Fabric Data Agent?
The most important difference between oneAgent and the Microsoft Fabric Data Agent is how reliable answers come about. With the Fabric Data Agent, reliability comes from continuously maintained instructions, numerous example queries and constant testing. With oneAgent, it comes from a metric definition that experts approve once and that then applies to every question about that metric.
The other differences at a glance:
- Language: According to Microsoft's documentation (as of May 2026), the Fabric Data Agent supports English only. oneAgent answers questions in German or English.
- Prerequisites: The Fabric Data Agent requires a paid Microsoft Fabric capacity, and the data must reside in Fabric. oneAgent works on an existing data warehouse, including Microsoft Fabric, or connects source systems directly through more than 550 connectors.
- Working together: oneAgent complements the Fabric Data Agent where business teams need reliable figures. With the Fabric Data Agent, differently worded questions about the same metric can lead to different results, and business users cannot check the generated query themselves. oneAgent returns the same figure for every approved metric, no matter how the question is asked. oneAgent takes over metrics from the semantic model in Power BI or Microsoft Fabric, or they are defined directly in oneAgent.
- From deviation to task: In the Microsoft ecosystem, this path is spread across several components that the data team sets up. In oneAgent, it is one continuous workflow with a goal, alerts and a task with an owner and a deadline.
- Data processing: oneAgent runs in Frankfurt or in the customer's own data center. With the Fabric Data Agent used directly in Fabric on a capacity in the EU, AI processing stays within the EU Data Boundary. When it is used through Microsoft 365 Copilot or other services, processing can take place outside the Fabric compliance boundary, according to Microsoft.
- Billing: oneAgent is licensed per user and can be rolled out one business team at a time. The Fabric Data Agent runs on a Fabric capacity, with per-user licenses added depending on usage.
The Fabric Data Agent is the obvious choice for companies that already run Fabric, work in English and have a data team with spare capacity for ongoing maintenance. oneAgent is built for companies where the approved definition should simply apply and business teams carry on by themselves.
What is the Microsoft Fabric Data Agent?
The Microsoft Fabric Data Agent is a feature in Microsoft Fabric that data teams use to build their own AI agents for questions in natural language. A Fabric Data Agent uses a language model to translate a question into a query in SQL, DAX or KQL, runs it against the connected data and returns the result.
A Fabric Data Agent can connect up to five data sources. Supported sources are lakehouses, warehouses, SQL databases, mirrored databases, Power BI semantic models, KQL databases and ontologies in Fabric, as well as Microsoft Graph. The data must reside in Microsoft Fabric or be mirrored there.
How well a Fabric Data Agent answers depends heavily on how it is set up. Microsoft recommends carefully preparing the semantic model, giving the agent instructions, adding up to 100 example queries per data source for SQL and KQL sources, and checking the answers with evaluations. Microsoft also advises regularly reviewing and updating instructions and example queries whenever data sources or requirements change. This setup is usually done by the data team; business users then ask the questions.
The Fabric Data Agent requires a paid Fabric capacity of size F2 or higher, or a Power BI Premium capacity of P1 or higher.
Where can the Fabric Data Agent be used?
The Fabric Data Agent can be used directly in Microsoft Fabric and embedded in other Microsoft services. It has been generally available in Microsoft Copilot Studio since August 2026. Microsoft announced general availability of the Microsoft 365 Copilot integration in June 2026; the corresponding documentation, as of May 2026, still lists it as a preview. Through Microsoft 365 Copilot, the Fabric Data Agent is also available in Microsoft Teams. It can additionally be connected to Microsoft Foundry through the Model Context Protocol (MCP).
For companies that already work with Microsoft 365 and Fabric, this is a real advantage. The Fabric Data Agent uses the existing permissions and can be governed with Microsoft Purview.
What new capabilities did the Fabric Data Agent gain in 2026?
Since August 2026, the Fabric Data Agent returns charts in addition to text and tables, including line, bar and stacked bar charts as well as pie, scatter and area charts. According to Microsoft architect James Serra, this capability is still in preview. According to the documentation, a single answer is limited to 25 rows and 25 columns.
With the updates through August 2026, Microsoft moved the Fabric Data Agent to the GPT-5.1 language model. For the model upgrades, Microsoft cites roughly 20 percent higher accuracy based on its own internal benchmarks. Also in preview are a Python code interpreter and an AI-assisted setup that suggests instructions and example queries. Documents can be included in preview through an Azure AI Search index. Users can also inspect which query the agent generated and which steps it ran.
What is oneAgent?
oneAgent is an AI BI assistant from oneLake GmbH for mid-sized companies. With oneAgent, employees ask questions about business figures in natural language, in German or English, and receive answers based on metrics that experts in the company have defined and approved. Beyond that, oneAgent keeps goals in view, reports deviations and makes sure a responsible person takes care of them within a set deadline.
How does oneAgent answer questions?
oneAgent separates understanding the question from calculating the figure. A language model identifies what is being asked, such as which metric, which time period and which filters are meant. If a question is ambiguous, oneAgent asks for clarification. Before answering, oneAgent summarizes how it understood the question. This lets users see right away whether the right metric is being calculated with the right time period and the right filters.
The calculation itself follows a fixed definition for approved metrics, with source, calculation logic, filters and owner. From this, oneAgent generates the database query according to fixed rules. The same question about an approved metric therefore always leads to the same result, regardless of how it is worded.
Every answer shows which source the figure comes from and which filters apply. oneAgent presents results as text, tables, charts or dashboards.
Which data can oneAgent connect to?
oneAgent works on an existing data warehouse or connects source systems directly. In Microsoft Fabric, oneAgent accesses both the data warehouse and Power BI semantic models. Other supported data warehouses and databases include SQL Server, PostgreSQL, Snowflake, Databricks and Oracle. More than 550 connectors also connect ERP, CRM and e-commerce systems such as SAP, Microsoft Dynamics 365, Salesforce or Shopware. Data can be queried live or loaded into the oneAgent Lake.
Metrics that are already defined in the semantic model in Power BI or Microsoft Fabric are taken over by oneAgent. The BI team reviews and approves the imported definitions instead of creating them from scratch.
What does oneAgent do beyond answering questions?
On request, oneAgent turns a query into a goal. Users define which value should be reached by when, and oneAgent suggests realistic target values. If a metric deviates from its goal, oneAgent reports it on its own and shows where the deviation comes from, for example which region, product group or customer group. oneAgent suggests who should take care of it and by when, creates a task after confirmation and tracks it until it is done.
Where does oneAgent run?
oneAgent runs in a Microsoft Azure data center in Frankfurt. Alternatively, it can be operated in a German data center or on premises. Access is governed by a role and permission concept. A free trial with demo data is available.
How do reliable answers come about?
AI assistants for business data give reliable answers when the language model knows what the metrics mean. Microsoft and oneAgent agree on this point. The approach differs: With the Microsoft Fabric Data Agent, reliability comes from ongoing maintenance of instructions, example queries and tests. With oneAgent, it comes from a metric definition that is approved once and then applies.
How does the Fabric Data Agent ensure reliable answers?
The Fabric Data Agent uses a language model to generate a new query in SQL, DAX or KQL for every question. To make these queries correct, Microsoft recommends several steps: preparing the semantic model for AI, giving the agent written instructions, adding up to 100 example queries per data source for SQL and KQL sources, checking the answers with evaluations, and republishing the adjusted agent when problems occur.
These steps improve accuracy, but they do not make the translation rule-based. The language model decides anew for every question which query to generate. Differently worded questions about the same metric can therefore lead to different queries. In September 2026, Power BI expert Reza Rad publicly showed how a Fabric Data Agent, used through Microsoft 365 Copilot, returned different figures than the underlying Power BI report, and how this can be corrected with additional instructions.
For Copilot in Power BI, Microsoft additionally offers "verified answers." Here, the data team approves selected visuals and links them to defined phrases, with up to 250 per semantic model and 15 phrases per answer.
How does oneAgent ensure reliable answers?
oneAgent calculates approved metrics according to a fixed definition. Experts from business teams, controlling or BI define the source, calculation logic, filters and owner for each metric and approve it. Metrics that are already defined in the semantic model in Power BI or Microsoft Fabric are taken over by oneAgent, so the BI team only needs to review and approve them.
Once a metric is approved, oneAgent generates the query for it according to fixed rules. The language model understands the question but does not calculate the figure itself. However a question about an approved metric is worded, the result is the same. Before each answer, oneAgent summarizes how it understood the question. Every answer shows its source and filters, so users can verify it.
Approved metrics need no example queries, no instructions for a language model and no recurring evaluations. A definition only changes when the responsible person changes it.
What does the maintenance effort mean for a small BI team?
For a BI team at a mid-sized company, the maintenance effort decides whether an AI assistant is practical at all. In many mid-sized companies, the BI team consists of just a few people who also build reports and dashboards, look after data sources and answer ad hoc requests.
With the Fabric Data Agent, maintaining the agent comes on top of that: every new type of question, every change to the data model and every new data source can require new example queries, adjusted instructions and fresh tests. This work does not end with the setup; it continues for as long as the agent is in use. The move to GPT-5.1 in August 2026 shows how concrete this effort is: Microsoft pointed out that the agent's behavior may change as a result and recommended re-running evaluations and updating instructions.
With oneAgent, the effort lies in defining and approving the metrics. It occurs once per metric. If the metrics already exist in the semantic model, the definition step disappears entirely and only the approval remains. From then on, business teams work independently with the approved metrics, and the BI team gains time for data quality, analysis and advisory work instead of correcting the agent after the fact.
How do oneAgent and the Fabric Data Agent complement each other?
oneAgent and the Microsoft Fabric Data Agent can be used together. In that setup, oneAgent provides business teams with reliable access to the company's metrics.
People in sales, controlling or management ask the same question in different ways and cannot check the query behind it themselves. With the Fabric Data Agent, the result can depend on the wording, because a language model generates the query anew for every question. oneAgent always returns the same figure for every approved metric, no matter how the question is asked. Metrics that are already defined in the semantic model in Power BI or Microsoft Fabric are taken over by oneAgent. All other metrics are defined by experts directly in oneAgent. oneAgent queries the underlying data in the data warehouse in Microsoft Fabric and in Power BI.
oneAgent therefore replaces neither Microsoft Fabric nor the Fabric Data Agent. The data team keeps working with Fabric, and business teams get a tool whose figures they can rely on, complemented by goal tracking and tasks that a responsible person takes care of within a set deadline.
Which tasks does the Fabric Data Agent handle and which does oneAgent handle?
The Fabric Data Agent suits the data team and exploratory questions that combine several data sources. A Fabric Data Agent can combine up to five sources and is embedded directly in Microsoft 365 Copilot, Teams and Copilot Studio. James Serra, solution architect at Microsoft, also recommends data agents mainly for exploratory questions and points to reports and dashboards for reliable, repeatable answers to known questions.
oneAgent suits business teams that repeatedly ask about the same business metrics and need to rely on those figures: sales, controlling, division management and executive management. oneAgent answers questions, always calculates approved metrics the same way and shows the source and filters with every answer. oneAgent also tracks goals, reports deviations and creates tasks that a responsible person takes care of within a set deadline.
With this division of work, the data team keeps working in Fabric, and business teams get a tool they can operate themselves, without the data team having to maintain an agent.
Can oneAgent take over metrics from Power BI and semantic models?
Yes. oneAgent takes over metrics that are defined in the semantic model in Power BI or Microsoft Fabric. The BI team therefore does not have to define existing metrics a second time. It reviews the imported definitions and approves them in oneAgent. From that moment, oneAgent returns the same value for every approved metric in every question. Metrics that are still missing from the semantic model are defined and approved by experts directly in oneAgent.
For the company, this means that the metric shown in the Power BI report is the same one business teams query in oneAgent. The definition work that went into the semantic model pays off a second time.
Do oneAgent and the Fabric Data Agent return the same figures?
When oneAgent takes over a metric from the same semantic model that the Fabric Data Agent uses, both answers are based on the same definition. The difference lies in how the question is translated: The Fabric Data Agent generates the query anew for every question with a language model, so the result can depend on the wording. oneAgent generates the query for approved metrics according to fixed rules and therefore returns the same value for every question.
How does a deviation become a completed task?
Turning a deviation into a completed task takes several steps: setting a goal, monitoring the metric, notifying people when it deviates, assigning a responsible person and following up until the task is done. In the Microsoft ecosystem, these steps are spread across several components that the data team sets up and connects. In oneAgent, they form one continuous workflow that business teams operate themselves.
How does this work in the Microsoft ecosystem?
The Microsoft ecosystem provides separate components for the individual steps:
- Power BI scorecards represent goals with owners, target values, start and due dates, and a rule-based status.
- Fabric Activator triggers emails, Teams messages or Power Automate flows when defined thresholds are reached, including on the basis of Power BI reports.
- The Operations Agent in Fabric Real-Time Intelligence has been generally available since the end of June 2026. It monitors data based on natural-language instructions, investigates anomalies, suggests actions and carries them out via Teams after approval.
- Power Automate connects the components and can create tasks in a task tool such as Microsoft Planner.
These building blocks can be combined into a complete chain. The data team, however, has to assemble and maintain it. A task with an owner and a deadline that is tracked until completion is not built into any of the components; it only comes about through Power Automate and an additional task tool.
The Operations Agent is also designed for technical and real-time scenarios. According to Microsoft's documentation, it works with one Eventhouse or ontology source, cannot calculate averages, minimums or maximums in ontology rules, and cannot combine AND conditions. It understands instructions and business goals in English only. A recommendation that is not approved expires after three days, and actions run with the permissions of the person who created the agent.
How does this work in oneAgent?
In oneAgent, the whole chain grows out of the question, in one tool. On request, oneAgent turns a query into a goal. The user defines which value should be reached by when, and oneAgent suggests realistic target values, including an estimate of what is achievable at the current pace.
oneAgent then monitors the metric on its own. If it deviates from the goal, oneAgent reports it proactively and shows where the deviation comes from, for example which region, product group or customer group. oneAgent suggests who should take care of it and by when, and drafts a task description. After confirmation, oneAgent creates the task and tracks it until it is done.
Because oneAgent works with approved metrics, every alert is based on the same figure that appears in every answer. This matters: monitoring is only as reliable as the figure being monitored. An alert based on an incorrectly calculated metric would assign a problem that does not exist to the wrong person and create nothing but extra work. Or worse: a real problem goes unnoticed because the metric is calculated incorrectly.
Business teams set up goals and alerts themselves, in German or English. The data team does not need to connect any components.
What are the prerequisites for oneAgent and the Fabric Data Agent?
The Microsoft Fabric Data Agent requires a paid Microsoft Fabric capacity of size F2 or higher, or a Power BI Premium capacity of P1 or higher. The data must reside in Microsoft Fabric or be mirrored there. Using it through Microsoft 365 Copilot may require additional per-user licenses.
oneAgent does not need a Fabric capacity. oneAgent requires access to a data warehouse or to the source systems, plus approved metrics. oneAgent takes over the metrics from a semantic model in Power BI or Microsoft Fabric, or experts define them directly in oneAgent. This requires a person or function in the company that is responsible for the metrics and approves them, for example in controlling or BI.
What applies if a data warehouse in Microsoft Fabric already exists?
If a company already runs a data warehouse in Microsoft Fabric, both solutions can be used without additional infrastructure. The Fabric Data Agent runs on the existing Fabric capacity. oneAgent accesses the data warehouse in Fabric and the semantic models in Power BI and takes over the metrics defined there.
What applies if the data warehouse is outside Microsoft Fabric?
If the data warehouse is outside Microsoft Fabric, for example on SQL Server, PostgreSQL, Snowflake, Databricks or Oracle, the data first has to be made available in Fabric for the Fabric Data Agent. Microsoft offers database mirroring and data pipelines for this. Both require a Fabric capacity.
oneAgent connects directly to the existing data warehouse. There is no need to transfer the data to another platform.
What applies if there is no data warehouse yet?
If the company's data still sits in individual systems such as ERP, CRM, accounting or an online store, a data foundation first has to be built in Microsoft Fabric for the Fabric Data Agent: loading the data, preparing it in a lakehouse or warehouse and creating a semantic model. This is a data project in its own right that requires Fabric expertise.
oneAgent connects these systems directly through more than 550 connectors, including SAP, Microsoft Dynamics 365, Salesforce, DATEV and Shopware. Data is queried live or loaded into the oneAgent Lake. Connecting the systems usually takes a few days.
Which languages do oneAgent and the Fabric Data Agent support?
Officially, the Microsoft Fabric Data Agent supports English only. As of May 2026, Microsoft's documentation states: "The Fabric data agent doesn't currently support non-English languages." Questions, instructions and example queries should therefore be written in English. In practice, the Fabric Data Agent partly answers questions in other languages as well, but not reliably. The Operations Agent also understands instructions and business goals in English only.
oneAgent answers questions in German or English. For each metric, the terms actually used in the company can be stored, such as "Umsatz," "Erlös" and "Revenue" for the same metric. This way, oneAgent understands business teams' questions in their own language.
For mid-sized companies in the DACH region, language is often decisive. Sales, purchasing, controlling and management there usually work in German, and an AI assistant for business teams has to understand their questions the way they are asked.
Where is the data processed?
With the Microsoft Fabric Data Agent, where processing takes place depends on how the agent is used. oneAgent runs in Frankfurt, in a German data center or on premises.
Where does the Fabric Data Agent process data?
When the Fabric Data Agent is used directly in Microsoft Fabric, it runs on the company's Fabric capacity, and questions are processed through Azure OpenAI. If the capacity is within the EU Data Boundary, this processing stays in the EU, according to Microsoft. The EU Data Boundary covers the EU as a whole, not a specific country.
When the Fabric Data Agent is embedded in Microsoft 365 Copilot, Copilot Studio, Microsoft Foundry or other applications, answers can be processed and stored outside Fabric's compliance boundary and region, according to Microsoft. The data policies of the respective service then apply.
For companies with high data protection requirements, this is not a reason to rule it out, but it is a point to check before embedding the agent, especially in regulated industries.
Where does oneAgent process data?
oneAgent is operated in compliance with the GDPR in a Microsoft Azure data center in Frankfurt. Alternatively, it can run in a German data center or entirely on premises. Access to metrics and data is governed by a role and permission concept.
How are oneAgent and the Fabric Data Agent billed?
oneAgent is licensed per user. Companies can therefore start with a single business team and gradually extend oneAgent to other areas, without having to expand a platform or capacity.
The Microsoft Fabric Data Agent has no price of its own. It runs on a Fabric capacity that is billed by size and shared by all of a company's Fabric workloads. Depending on usage, per-user licenses are added, for example for Power BI or Microsoft 365 Copilot.
How high the costs are in an individual case depends, for both solutions, on the existing infrastructure and the number of users.
How do oneAgent and the Fabric Data Agent differ in detail?
oneAgent and the Microsoft Fabric Data Agent differ mainly in how reliable answers come about, who sets up and maintains the solution, and whether the path from deviation to completed task is built in. The following tables compare both solutions by topic, as of September 2026.
How do the answers from oneAgent and the Fabric Data Agent differ?
| Feature | oneAgent | Microsoft Fabric Data Agent |
|---|---|---|
| Type of solution | Ready-to-use AI BI assistant for business teams | Feature in Microsoft Fabric that data teams use to build their own AI agents |
| Natural-language questions | Yes | Yes |
| Language | German and English | Officially English only (Microsoft documentation, as of May 2026) |
| How reliability comes about | Metric definition approved once, with source, calculation, filters and owner | Instructions, up to 100 example queries per SQL or KQL source and evaluations, continuously maintained |
| Translating the question | Language model understands the question; the query for approved metrics is generated by fixed rules | Language model generates a new query in SQL, DAX or KQL for every question |
| Differently worded questions about the same metric | Same result for approved metrics | Result can depend on the wording |
| Human approval | On every single metric, with an owner | No per-metric approval in the Data Agent; in Copilot for Power BI, "verified answers" for selected visuals (up to 250 per semantic model) |
| Answer transparency | Summary of how the question was understood before the answer; source and filters with every answer | Generated query and run steps can be inspected; clarifying questions for ambiguous requests in preview |
| Output | Text, tables, charts and dashboards | Text, tables and, since August 2026, charts (preview); up to 25 rows and 25 columns per answer |
What data and prerequisites do oneAgent and the Fabric Data Agent require?
| Feature | oneAgent | Microsoft Fabric Data Agent |
|---|---|---|
| Supported data sources | Data warehouses, including Microsoft Fabric (warehouse and Power BI semantic models), SQL Server, PostgreSQL, Snowflake, Databricks and Oracle; more than 550 connectors for ERP, CRM and e-commerce systems | Lakehouse, warehouse, SQL database, mirrored database, Power BI semantic model, KQL database and ontology in Fabric; Microsoft Graph |
| Number of data sources | No limit on connections, through more than 550 connectors and all common data warehouses | Up to five sources per agent; a company can create multiple agents |
| Metrics from the semantic model | Taken over, reviewed and approved; additional metrics can be defined directly in oneAgent | Power BI semantic model usable as a data source |
| Documents and text | Public web pages can be read into the chat by URL | Documents through an Azure AI Search index (preview) |
| Technical prerequisites | Access to a data warehouse or to source systems | Paid Fabric capacity of F2 or higher, or Power BI Premium capacity of P1 or higher; data in Microsoft Fabric |
| Setup and maintenance | Definition and approval per metric, once | Agent set up by the data team; instructions, example queries and tests on an ongoing basis |
| Day-to-day use | Business teams | Business teams ask questions; the data team sets up and maintains |
How do oneAgent and the Fabric Data Agent support the path from deviation to task?
| Feature | oneAgent | Fabric Data Agent and other Microsoft components |
|---|---|---|
| Tracking goals | Create a goal from a query on request, with suggested realistic target values | Power BI scorecards with owners, target value and due date, set up separately |
| Reporting deviations | Reports proactively and shows where the deviation comes from, for example by region, product or customer | Fabric Activator for thresholds; Operations Agent based on natural-language instructions (English only, one Eventhouse or ontology source) |
| Task with owner and deadline | Responsible person and deadline suggested, task created after confirmation and tracked until completion | Not built in; has to be added through Power Automate and a task tool such as Microsoft Planner |
| Who sets up the chain | Business teams, in one tool | Data team, from several components |
How do operation, integration and billing differ?
| Feature | oneAgent | Microsoft Fabric Data Agent |
|---|---|---|
| Operation | Microsoft Azure in Frankfurt, a German data center or on premises | On the company's Fabric capacity; AI processing for capacities in the EU within the EU Data Boundary |
| Integration with Microsoft 365 | Standalone tool that accesses Microsoft Fabric and Power BI | Native in Microsoft 365 Copilot, Teams, Copilot Studio and Microsoft Foundry; governed through Microsoft Purview |
| Billing | Per user, rolled out one business team at a time | Fabric capacity billed by size, plus per-user licenses depending on usage |
When is the Fabric Data Agent the better choice?
The Microsoft Fabric Data Agent is the better choice for companies that already use Microsoft Fabric in production, work in English and have a data team with spare capacity for ongoing agent maintenance.
When Fabric is in production use and the data team has time for maintenance
If the data already resides in Microsoft Fabric and the data team can maintain instructions, example queries and tests on an ongoing basis, the Fabric Data Agent can be introduced without additional infrastructure. The capacity is in place and the permissions are set up.
When English is the working language
The Fabric Data Agent officially supports English only. In international teams or companies with English as their working language, this limitation does not matter.
When questions should be asked directly in Microsoft 365 Copilot and Teams
The Fabric Data Agent is natively embedded in Microsoft 365 Copilot, Teams and Copilot Studio and governed through Microsoft Purview. Organizations that want to keep data questions entirely within Microsoft interfaces benefit from this integration.
When it comes to real-time and operational data
For technical scenarios such as sensor data, production equipment, logistics or data pipelines, Microsoft offers a specialized solution in the Operations Agent in Fabric Real-Time Intelligence, which monitors data every few minutes.
When is oneAgent the better choice?
oneAgent is the better choice for companies where business teams should work independently with reliable metrics, without the BI team having to maintain an AI agent on an ongoing basis. This applies especially to companies that already use Microsoft Fabric and Power BI and, as the next step, are looking for an agentic solution that makes metrics, goals and tasks accessible to business teams.
When business teams work in German
oneAgent answers questions in German or English and understands the terms that are actually used in the company. Sales, controlling, purchasing and management ask in their own language.
When metrics must be binding
oneAgent returns the same figure for every approved metric, no matter how the question is worded. This is crucial when figures have to hold up in management meetings and nobody should be arguing about which value is correct.
When the BI team is small
With oneAgent, the effort occurs once per metric: defining and approving it, or just approving it when a semantic model already exists. Ongoing maintenance of instructions, example queries and tests is not needed. A BI team of just a few people gains time instead of taking on new maintenance work.
When metrics are already defined in Power BI
oneAgent takes over metrics from the semantic model in Power BI or Microsoft Fabric. The work that went into the model becomes usable for business teams, and the Power BI report shows the same figure as oneAgent's answer.
When deviations should turn into completed tasks
oneAgent leads from the question to the goal, the alert and the task with an owner and a deadline, in one tool and operated by business teams. In the Microsoft ecosystem, the data team would have to assemble this chain from several components.
When data does not reside only in Microsoft Fabric
In addition to Microsoft Fabric, oneAgent connects other data warehouses and, through more than 550 connectors, systems such as SAP, DATEV, Microsoft Dynamics 365, Salesforce or Shopware directly, without first transferring the data to another platform.
When strict data protection requirements apply
oneAgent runs in Frankfurt, in a German data center or on premises. For companies in regulated industries or with strict internal policies, this is often a prerequisite.
The two solutions are not mutually exclusive. How oneAgent and the Fabric Data Agent work together is described in the section "How do oneAgent and the Fabric Data Agent complement each other?"
Frequently asked questions about oneAgent and the Fabric Data Agent
Can oneAgent access Microsoft Fabric?
Yes. In Microsoft Fabric, oneAgent accesses both the data warehouse and Power BI semantic models. The data stays in Fabric; there is no need to transfer it to another platform.
Does oneAgent replace Microsoft Fabric or the Fabric Data Agent?
No. oneAgent complements Microsoft Fabric and can be used alongside the Fabric Data Agent. The data team keeps working in Fabric, and oneAgent makes the metrics defined there usable for business teams, with goal tracking, alerts and tasks.
Does the BI team have to redefine metrics for oneAgent?
No. oneAgent takes over metrics that are already defined in the semantic model in Power BI or Microsoft Fabric. The BI team reviews and approves the imported definitions. Metrics that are still missing from the semantic model are defined directly in oneAgent.
Why can the Fabric Data Agent return different figures for similar questions?
The Fabric Data Agent uses a language model to generate a new query for every question. Differently worded questions about the same metric can therefore lead to different queries and results. Microsoft recommends narrowing this down with instructions, example queries and evaluations. oneAgent generates the query for approved metrics according to fixed rules and returns the same result for any wording.
Do oneAgent and Power BI return the same figures?
Yes, when oneAgent has taken over the metric from the semantic model in Power BI. Both are then based on the same definition, and the Power BI report shows the same figure as oneAgent's answer.
Does the Fabric Data Agent understand German?
Not officially. According to Microsoft's documentation (as of May 2026), the Fabric Data Agent supports English only, and questions, instructions and example queries should be written in English. In practice, other languages partly work, but not reliably. oneAgent answers questions in German or English.
What is the difference between the Fabric Data Agent and Copilot in Power BI?
Copilot in Power BI works inside Power BI and answers questions about reports and semantic models. The Fabric Data Agent is a separate item in Microsoft Fabric that data teams set up for specific tasks, with up to five data sources and its own instructions, and that can be embedded in Microsoft 365 Copilot, Teams and Copilot Studio. Both require a paid Fabric capacity of F2 or higher or a Power BI Premium capacity of P1 or higher. How oneAgent differs from Copilot in Power BI is covered in a separate comparison.
What is the Operations Agent in Microsoft Fabric?
The Operations Agent is a component of Fabric Real-Time Intelligence and has been generally available since the end of June 2026. It monitors data based on natural-language instructions, suggests actions and carries them out after approval, for example through Teams or Power Automate. It is designed for technical and real-time scenarios, works with one Eventhouse or ontology source and understands instructions in English only. oneAgent is designed for business teams monitoring business metrics.
Can oneAgent assign tasks to employees?
Yes. If a metric deviates from its goal, oneAgent suggests who should take care of it and by when, and drafts a task description. After confirmation, oneAgent creates the task and tracks it until it is done.
Is oneAgent GDPR compliant?
Yes. oneAgent is operated in compliance with the GDPR in a Microsoft Azure data center in Frankfurt. Alternatively, it can run in a German data center or on premises.
How quickly can oneAgent be introduced?
Connecting to Microsoft Fabric or other data sources usually takes a few days. After that, the first metrics are taken over or defined and approved. A typical start focuses on the few metrics that are actually discussed in management meetings, followed by a gradual expansion.
Can oneAgent be tested for free?
Yes. A free trial with demo data is available. The "Get started" page describes which capabilities the trial includes.
Sources and status
This comparison is based on Microsoft's official documentation, the Microsoft Fabric update blogs, articles by Microsoft experts and product information from oneLake GmbH. Last reviewed: September 25, 2026. Microsoft is continuously developing the Fabric Data Agent, so individual details may change. Where the documentation is older than an announcement, this is noted in the text.
Microsoft documentation
- Fabric data agent concepts, Microsoft Learn, as of May 11, 2026: how it works, data sources, prerequisites, language support, example queries, output limits, recommendation for regular review
- Consume a data agent from Microsoft 365 Copilot, Microsoft Learn, May 2026: licenses for use through Microsoft 365 Copilot
- Consume Fabric data agent from Microsoft Foundry via Fabric IQ, Microsoft Learn: data processing outside the Fabric compliance boundary when embedded in other services
- Copilot tenant settings, Microsoft Learn: processing within the EU Data Boundary
- Verified answers in Power BI, Microsoft Learn, as of September 10, 2026: verified answers in Copilot for Power BI
- Copilot for Power BI overview, Microsoft Learn: capacity requirements for Copilot in Power BI
- Operations agent and Operations agent best practices and limitations, Microsoft Learn, as of August 6, 2026: capabilities and limitations of the Operations Agent
- Fabric Activator, Microsoft Learn: alerts on thresholds
- Goals and scorecards in Power BI, Microsoft Learn: goals with owners and due dates
- Licence clarification for Fabric Data Agent, Microsoft Q&A, answer from a moderator (Microsoft External Staff), June 15, 2026: Copilot license per user
Microsoft announcements
- Building the agentic analytics stack: Fabric Analytics at Build 2026, Microsoft Fabric Blog, June 2, 2026: general availability in Microsoft 365 Copilot, roughly 20 percent higher accuracy based on internal benchmarks, code interpreter and AI-assisted setup in preview
- Operations agent generally available, Microsoft Fabric Blog, June 3, 2026: general availability of the Operations Agent
- Fabric August 2026 Feature Summary, Microsoft Fabric Blog, August 27, 2026: charts, move to GPT-5.1 with the recommendation to re-run evaluations, general availability in Copilot Studio
Expert articles
- Microsoft Fabric Data Agents: bringing structured and unstructured data together, James Serra, solution architect at Microsoft, September 3, 2026: supported sources, inspecting generated queries, preview features, use cases for data agents and reports
- Fabric Data Agents are English-first (for now), Luca Zavarella, Microsoft MVP, March 2026: behavior with questions in languages other than English
- Fabric Data Agent giving wrong answers? Fix it with AI agent instructions, Reza Rad, Microsoft MVP, video from September 2026, summary on HubSite365: diverging figures between the Data Agent and the Power BI report, and a fix through instructions
Information about oneAgent
All information about oneAgent comes from oneLake GmbH, as of September 2026. The "Get started" page describes the scope of the free trial.