Your AI is only as good as what it can find. 

Part One of How to prepare your Qlik App for Qlik Answers laid out a six-step checklist:

• clarify field names,
• streamline the data model,
• check date formats,
• build master items with descriptions,
• configure vocabulary,
• and validate with real queries.

It ended with a simple line: you do not need a perfect data model, you need a clear one. This is what clear means underneath the surface. Keep the shortlist clean, and the model's fluency has good material to work with.

Part Two details what Qlik Answers is doing when it reads your apps and expands your understanding of why those six steps move the needle.

Content

• The mental model: retrieval rather than omniscience
• Lever 1: Visibility
• Lever 2: Names
• Lever 3: Master items
• Lever 4: Descriptions
• Lever 5: Vocabulary
• In Summary

The mental model: retrieval rather than omniscience

It is easy to assume the AI "sees" your entire app and simply needs to be clever enough to choose well. In practice it works a little differently, and for good reason. An app with 500 fields would present far more noise than signal to any model.

Instead, the system builds a search index of your app's metadata. Every visible field and master item gets a compact semantic profile, and that profile is what the system searches against.

For a master measure, that profile is built from several signals together:

• the name of the master item
• the description you have written for it
• the actual expression behind it
• any business glossary term you have linked to it

The system reads these as a whole and forms one understanding of what the asset means. When a question arrives, it retrieves the handful of assets that best match its meaning and passes them to the AI.

That single idea reframes the work. The useful question is rarely "is the AI clever enough to understand my field?" More often it is "will my field be found, and will it be the natural choice when it is?"

We'll tackle this in order, using five different levers. These levers act at different stages, which is why the sequence in which we outline them is important. Visibility decides what exists. Names influence what gets found first. Master items shape what the AI builds your numbers with and whether they carry your definitions. Descriptions help the right things get found at all. Vocabulary closes the last gap between your words and the model's understanding.

Working the later levers on an app still full of visible technical fields means refining assets that are competing with a great deal of noise. It tends to go more smoothly to start with the gate, then the definitions, then the language.

Lever 1: Visibility

The one truly decisive setting.

Hiding a field in your business logic does more than lower its priority. It takes the field entirely out of consideration. A hidden field is never profiled, never indexed, and never retrievable, so it cannot appear in an answer.

That is why visibility is worth looking at first. Every technical field left visible, every surrogate key, ETL timestamp, or FIELD_CODE_2_FINAL, competes with your real business fields. Working through a wide app and hiding the fields that should never be an answer is usually a few hours of work, sometimes less, and it improves every question asked afterwards. It is one of the most approachable, high-value things you can do to an app.

Hide with intent. If people do ask about a field, hiding it takes away their ability to get answers about it. Think of this as curation rather than cleanup.

Lever 2: Names

Names carry weight in retrieval.

When a question contains a term that closely matches an asset's name, that asset is favored strongly in the results.

So it pays to name assets the way your users actually speak. If everyone in the business says "Net Revenue", a master measure called "Net Revenue" tends to be found quickly and reliably. One called "NR_calc_v3" leaves the system to infer what you meant. It often will, and it is worth giving it a clear signal where you can do so easily.

Lever 3: Master items

Define your metrics once, with care.

Master items sit high on this list for a straightforward reason: when a well-defined master item and a raw field could both answer a question, the system leans toward the master item. It is your sanctioned, promoted answer.

The stronger argument, though, is trust rather than mechanics. The model does understand what "net sales" means in general terms. The question is whether you would prefer it to work from the general idea or from your definition.

Does your net sales exclude intercompany transfers? Is it net of returns or gross of them? Which currency, at which exchange rate, over which fiscal calendar? Answered from raw fields, each of those becomes a reasonable assumption. Answered through a master measure, it follows your business logic, your set analysis, and your exceptions, consistently. When the number reaches a report your leadership is reading, a defined metric gives everyone confidence in where it came from. Define the metric once, as a master item, and every answer built on it inherits the definition you stand behind.

It is worth building a master item for every metric that carries a real business definition, while keeping the library tidy. A collection of "Sales", "Sales 2", and "Sales New" tends to reintroduce the very ambiguity master items are meant to resolve, so clear naming here matters as much as the definitions themselves.

Lever 4: Descriptions

Think of them as configuration rather than documentation.

Descriptions are often treated as a nice-to-have for the data catalog. In an AI-ready app, they do more.

The description you write on a field or master item feeds directly into its semantic profile, which is what the system searches. A clear note on what "Net Sales" includes and excludes helps that item surface for the many ways people might phrase the question, and sets it apart from the other sales measures nearby.

There is no need to describe everything. The ones worth your time are the ambiguous ones: lookalike measures, fields whose names are abbreviations, anything a new analyst would have to ask a colleague about. If a person would want the explanation, the search index benefits from it too.

Lever 5: Vocabulary

Focus on your language, not on educating the model.

This is the lever people tend to overthink, so it is worth being clear about its purpose. You are not here to teach the AI general knowledge. There is a genuinely capable model underneath, and it already understands a great deal of the world.

It knows what revenue, profit, and margin are. It understands churn, retention, and attrition. It is comfortable with YoY, MoM, and QoQ, with fiscal quarters, gross versus net, currencies, countries and regions, standard units, and the everyday language of sales, finance, and operations. There is little value in a glossary entry explaining that "revenue" means money coming in.

Where vocabulary earns its keep is the language that is specific to you, which the model has no way to know:

• Your internal product codenames.
• An acronym only your company uses.
• The name your business gave to a segment it defined itself.
• A regional grouping that exists only on your org chart.
• If your business says "GMV" and the field is named "Gross_Merch_Value"

Those are examples the model cannot bridge on its own, and this is precisely where a vocabulary entry helps.

So a small, proprietary vocabulary tends to serve you best. A handful of terms that are truly yours will often do more than a long list restating things the model already knew. Vocabulary is a useful tool with a focused job: teaching the model your dialect rather than the dictionary.

Two items from that checklist do not appear as levers above: date formats and validation. Neither is a retrieval signal. A Date or Timestamp tag is a data type requirement your time-based questions depend on, not something the search index weighs when it ranks assets. Validating with real queries is not a lever at all. It is how you find out whether the five above actually worked.

In Summary

When people ask your app questions in plain language, the AI does not read your whole data model. It searches a curated index of your fields and master items, retrieves a small shortlist, and reasons over that. Most of what helps an app answer well comes down to making sure the right assets land on that shortlist. A sensible order to work through: hide what should never be an answer, name assets the way your users talk, define your real metrics as master items, describe the ambiguous ones, and add vocabulary for the terms that are genuinely yours.

Environment

• Qlik Cloud
• Qlik Answers

This article answers the most frequently asked questions about Qlik Discovery Agent. It is split into five sub-sections:

• General FAQ
• Availability and Integration
• How does Discovery Agent work?
• Technical FAQ
• Insight Generation Logic

If you are looking for information on how to get started, check out the Discovery Agent Interactive Walkthrough and our Discovery Agent Documentation.

General FAQ

What is Qlik Discovery Agent?

Discovery Agent is an AI-powered, always-on monitoring capability in Qlik Cloud that automatically detects meaningful changes, anomalies, and trends in your data. It requires no rules, thresholds, or manual setup. Discovery Agent identifies spikes, drops, trend shifts, baseline changes, and data quality issues, then delivers clear, plain-language insights in a prioritized feed.

What makes Discovery Agent different from traditional alerting?

Traditional BI alerts rely on predefined thresholds or manual logic. Discovery Agent uses the Qlik Analytics Engine and its associative capabilities to evaluate wide combinations of data relationships automatically and proactively surface only those insights that matter. It is context aware, adaptive, and far more scalable than rules driven systems.

Who is Discovery Agent for?

• Executives who want high-level visibility into what changed
• Business users who need easy-to-read insights without deep analytics skills
• Non-technical users who do not want to build dashboards or rules
• Analysts who want automated detection to replace repetitive KPI monitoring

Availability and Integration

Does Discovery Agent work with Qlik Cloud Analytics?

Yes. Discovery Agent is built directly into Qlik Cloud Analytics and leverages the Qlik Analytics Engine for associative, large scale anomaly detection.

A Premium or Enterprise subscription is required. See Qlik Pricing for details or contact your Qlik account representative. 

Does Discovery Agent integrate with Qlik Answers?

Yes. You can ask questions directly from an insight card, and context from the insight will be transferred into Qlik Answers.

Is Discovery Agent available on Qlik Sense Enterprise Client-Managed?

No. Discovery Agent is built exclusively for Qlik Cloud.

Does Discovery Agent impact app performance?

No. Monitoring runs outside active dashboards, ensuring no performance impact on live analytics experiences.

Does Discovery Agent follow Qlik Cloud security and governance rules?

Yes. Insight delivery respects user permissions, governed access, and security boundaries.

How does Discovery Agent work?

How does Discovery Agent detect insights?

Discovery Agent analyzes updated app data models using associative evaluation to identify:

• Spikes or dips
• Trend shifts
• New or emerging baselines
• Record highs and lows
• Values above or below expected patterns
• Missing or inconsistent data

No rules or thresholds are required.

Is Discovery Agent real-time?

Discovery Agent is always on, but processes changes when the application’s data model updates. Insights refresh after reload and appear in the feed once the system evaluates new data. Updated are currently capped at one reload per day.

How often does the feed refresh?

The feed automatically refreshes upon reload. For most apps, this occurs once per day or whenever new data is introduced.

Can I customize the insights I see?

Yes. You can follow specific apps or insight categories once the Following tab is released. Filtering options are also planned to help tailor results.

Technical FAQ

What are Insight Triggers?

Insight Triggers are structured metric definitions that serve as the foundation for generating analytical insights within the application. Each trigger is composed of a measure or expression,  such as a calculated field or KPI, along with a set of additional configuration parameters. These parameters include the frequency at which the trigger evaluates data and the type of calculation to be applied (example: sum, average, count).

Together, these elements define the conditions under which an insight is surfaced to the user.

Do I Have to Set a Date Period for Each Trigger?

Yes, a date period is required for every trigger you configure.

All insights generated by the system are trend-based, meaning they analyze data over time to identify patterns, changes, or anomalies. This requires a date period to be added to the trigger's associated group. Without a defined time range, the system cannot perform the temporal comparisons necessary to produce meaningful insights.

How Do I Refresh the Insight Feed?

The Insight Feed refreshes automatically each time the page is reloaded. No manual refresh action is required. The feed itself is regenerated once per day, and this regeneration is triggered by the introduction of new data into the application or applications that contain active triggers. As a result, the feed will always reflect the most recent data available as of the last daily reload cycle.

Can I Filter Individual Values in a Dimension?

Filtering functionality is available in the Feed. A Filter button is currently visible at the top of the feed during the preview phase of the application. Users can use this to find specific insights in the feed.

How Many Triggers Can I Create?

• Each application supports a maximum of 50 triggers. For each trigger that has been configured, the system will calculate and present up to 5 insights per reload cycle.
• A tenant-wide guardrail exists to help control the volume of insights in the feed, providing broad analytical coverage across your defined metrics and expressions.

Where Are Triggers Stored?

Triggers are stored directly within the application in which they are created. They are not stored externally or in a centralized repository. That means each application manages its own set of triggers independently, and triggers defined in one application will not carry over to or affect another application.

Can I Ask Questions Directly in the Feed?

Direct question-and-answer functionality within the feed is available.

The Insight Feed is integrated with Qlik Answers, enabling users to ask natural language questions without leaving the feed interface. Because each card displayed in the feed is tied to a specific application, context from the relevant card will be automatically transferred to Qlik Answers to ensure accurate, contextually appropriate responses.

Why Am I Seeing a Large Number of Old Insights?

This behavior is expected and occurs specifically after the first reload following the creation of new triggers.

During this initial reload, the system performs a comprehensive scan of all available historical data, rather than only the most recent data. This allows it to identify any and all qualifying insights across the full dataset. This is a one-time process. All subsequent reloads after this initial one will only evaluate and surface insights based on newly introduced data, so the volume of older insights will not continue to grow with each reload.

Does This Feature Require the Cross-Region Inference Toggle?

Yes. The Insight Feed and its associated trigger functionality require the cross-region inference toggle to be enabled. Please ensure this setting is activated in your environment before attempting to configure triggers or access the feed. If you are unsure how to enable the cross-region inference toggle, contact your system administrator or refer to the relevant configuration documentation.

How Do I Remove Insights from the Feed?

To remove specific insights from the Insight Feed, you must delete the trigger that is generating those insights. Because the feed is dynamically generated based on active triggers, removing a trigger will prevent its associated insights from appearing in future feed reloads.

Deleting a trigger is a permanent action.

If you wish to stop surfacing certain insights temporarily, consider whether disabling or modifying the trigger may be a more appropriate course of action, depending on your platform's available options.

Does This Feature Support Section Access?

Section Access is not currently supported for applications used with the Insight Feed.

Any application that has Section Access enabled is incompatible with this feature at this time. As a result, all users who have been granted access to a given application will be able to see the insights generated from that application's triggers, regardless of any Section Access restrictions that may otherwise apply within that application.

This is an important consideration when deciding which applications to configure with triggers, particularly for datasets that contain sensitive or role-restricted data. Support for Section Access may be introduced in a future release.

Insight Generation Logic

How many data points are required to generate an insight?

Below is the minimum data requirement:

Weekly/Monthly/Quarterly/Yearly aggregation

• Spike detection, record high/low, model deviation: 4–50 data points
• Trend/baseline change: 20–50 data points

Daily aggregation

• Spike detection, record high/low, model deviation: 7–365 data points
• Trend/baseline change: 20–365 data points

Why am I not getting any insights?

Missing dates in the date field may prevent calculations. Creating a master calendar in the
load script can resolve this. Qlik is exploring options for date imputation.

Capability

While there is a lot of hype around #AI in the world, there is no denying it's power ... for the RIGHT USE CASES.

Imagine the ability to ask Natural Language questions of the data you have in ServiceNow without needing to pre-load the data in Qlik. Pretty cool in any situation. But downright required in situations where there are reasons for the data to not egress out of ServiceNow. 

ServiceNow provides something called AI Agents, that are very similar to Qlik Answers, in that users are able to simply ask natural language questions about their data. You gotta LIKE that.

Also like Qlik Answers, ServiceNow AI Agents are able to be called via a well documented REST API. Which means you can consume those endpoints from within Qlik. You gotta LOVE that. 

Scenario

That capability provides tremendous value for our joint customers for so many scenarios. The one I will focus on in this article is regarding a Customer Success Manager working in a Qlik.

The Qlik application they utilize has lots and lots of data that they can do analytics with using traditional dashboarding and visualization. Of course they simply open Qlik Answers to ask questions of any data that is already loaded in the application. Assume for this scenario that the CSM is focusing on customer renewals for whatever product they might be selling. 

They come across an account that is due for renewals soon, they can see that Qlik Predict has predicted that the account is likely to churn, and has ranked their health very low. They can see the NPS scores for the account have dropped significantly. Oh boy, they need to turn that frown, upside down. 

Before they call the customer, they would also like to know if they have any outstanding trouble incidents/problems. 

Not hard to imagine this use case. Having access to that information provides vital insights that the CSM needs to know how to effectively communicate with the customer when they call. How they get access also matters. They could always stop their day job, working in Qlik, and go login to ServiceNow, become a master in filtering and find the incidents that are open for this specific customer. But with the handy dandy Qlik Sense extension called ServiceNow Agent Dream shared in this post they don't have to. They simply ask "Do they have any open incidents?" 

Scenario Demonstration

I've created a simple Next - Next - Finish type demonstration that allows you to see this scenario in action. You simply start the tour, and press Next when you are ready to advance through it. 

Walk through the CSM Demonstration

ServiceNow AI Agent Dream Setup

If your goal was simply understand the capability and the scenario so you know "what's possible" then you can stop reading. But if your goal is to actually try and get this dream capability up and running in your environment then please continue. The next link is a very similar demo where you simply press Next to advance through, that walks you through how to create the same ServiceNow AI Agent that I used in my scenario. 

Walk through creating a ServiceNow AI Agent

Now that you at least understand the capability it's time to dive into the 3 attachments for this post:

1. Incident_Intelligence_Agent_Guide.docx - This document will walk you through how to build the ServiceNow AI agent since you will need 1 in your environment to call. Please refer to ServiceNow help on the topic or your BFF Claude/ChatGPT for help should anything in your environment differ from mine. 

2. ServiceNowAgentDream_Setup_Guide.docx - This document will walk you through how to install the ServiceNow Agent Dream extension and then how to configure it. OAuth authentication is required for ServiceNow AI Agents, so I walk you through how to do that as well. 

3. ServiceNowAgentDream.zip - This is the Qlik Sense extension that you need to download and install in your Qlik Sense environment/tenant. 

The Model Context Protocol (MCP) is an open standard that lets AI assistants connect to external data sources and tools.

This article provides step-by-step instructions for connecting Mistral Vibe to the Qlik MCP server in any tenant. Once connected, you can ask Mistral to check something in Qlik for you, live, instead of you tabbing over to do it yourself.

Content

• Before you start
• What you'll need
• Mistral Vibe web app (chat.mistral.ai)
• Vibe CLI
• Troubleshooting

Before you start

Mistral gives you two separate places to do this, and they don't share their setup: the Mistral Vibe web app (formerly known as "Le Chat":chat.mistral.ai) and the Vibe CLI (the terminal tool). Pick the article section for the one you actually use. If you use both, you'll need to do both; no shortcut covers the two at once.

What you'll need

• An account with a Qlik Cloud Tentant Admin role, with the appropriate user permission for MCP (see Qlik MCP server access and permissions | help.qlik.com).
• The MCP Qlik Cloud tenant URL: https://<name>.<region>.qlikcloud.com/api/ai/mcp

  Recommendation: Copy the URL; a typo leads to a silent failure during setup.

Mistral Vibe web app (chat.mistral.ai)

This path creates what Mistral calls a Connector: a connection to the Qlik MCP server that Mistral hosts and manages for you, usable from chat.mistral.ai (and Work, if your org has it). It's registered once on console.mistral.ai, Mistral's Studio, and from then on it just shows up as something you connect to like any other integration.

Step 1: Add a Connector in Studio

1. Log into console.mistral.ai
2. Go to Build > Connectors (or head straight to console.mistral.ai/build/connectors). 
3. Click + Add Connectors on the top right corner
   
   
   
   

Step 2: Create the connector and authenticate

1. Go to Custom MCP Connector 
2. Fill out the form:
   • Name: Connector name in Mistral (this is the name you will reference later)
   • Description: Connector Description
   • Server URL: https://<name>.<region>.qlikcloud.com/api/ai/mcp
   • Visibility: private if the connection is only intended for you, shared_workspace if your whole team should get it
     
     
     
     
3. Continue by filling out Authentication information, keeping the Authentication Method to Auto-detect. Mistral will select OAuth 2.1.

   Mistral OAuth requires Dynamic Client Registration; see Configuring OAuth dynamic client registration | help.qlik.com.

4. Leave the OAuth 2.1 Advanced settings (Redirect URI, Client ID, Client Secret) as they are
5. Click Create
   
   
   
   
6. You are redirected to the login page as defined in Qlik Cloud (your identity provider or Qlik Account if there is no identity provider specified in the tenant).
7. Click Approve 

   You must be logged in as a Tenant Admin.

   

Step 4: Use the connection in chat.mistral.ai

1. Open chat.mistral.ai
2. Go to Connectors in the + icon at the bottom left of the chat bubble, and select Qlik MCP Server (the actual name of your MCP connector in Mistral Studio).
   
   
   
   Qlik MCP Server will show up in the bubble chat, ready to use (in our example, the connectors are both MCP presales-iberia and Qlik MCP Server; in your case, it will show the name previously provided).
   
   From here, ask Mistral something that needs Qlik, and it'll reach for it on its own.

Vibe CLI

The Vibe CLI doesn't read Studio Connectors, at least not yet, so even if you already set up a Studio Connector, you will still need to complete the CLI setup separately to define the MCP server inside Vibe. 

To obtain and install Vibe CLI, follow the steps in Install the Vibe CLI and send your first prompt | docs.mistral.ai.

Step 1: Add the Qlik Server

1. Open Vibe from the terminal with the command:

   vibe

2. Once you see Vibe's prompt, add the Qlik server with the following command, including your tenant name and region, as well as the name you want to give your MCP Server.
   
   In our example, we use qlik_mcp_server_cli as the name; yours can be any descriptive text that helps you identify the Qlik MCP Server connection in the Vibe CLI.
   

   /mcp add https://<name>.<region>.qlikcloud.com/api/ai/mcp --name qlik_mcp_server_cli --scope user_default --scope mcp:execute --transport streamable-http

   
   
   
   
3. You will see a line confirming the server was added, then "Starting OAuth login...".

   At the time of writing, Vibe will experience a defect at this exact step. It cannot finish logging in on the first try, failing with the error:
   
   Error: No MCP servers configured.
   
   
   
   This is a known defect in Vibe CLI 0.21.0. Vibe CLI 0.23.2 addressed part of it, and if your version of Vibe CLI is current, this issue shouldn't happen anymore.

    

Step 2: Exit and reopen Vibe

1. Exit Vibe using the command:

   /exit

2. Open Vibe with the command:

   vibe

Step 3: Complete signing in to Qlik

1. Begin the login process with the following command, replacing qlik_mcp_server_cli with the name you set previously.

   /mcp login qlik_mcp_server_cli

   Vibe prints a web address and tries to open your browser for you. If nothing opens, copy that URL and paste it in a browser window yourself.
   
   
   
   
2. Log in with your Qlik Cloud tenant admin and click Approve. 
   
   
   
   Once completed, close that tab and return to the terminal.
   
   
3. Confirm the setup was successful by running the command:

   /mcp status

4. Look for your Qlik MCP server name in the list and verify it is listed as authenticated, not as needs auth.
   
   It will show 60 tools at the time of this writing; as Qlik continuously evolves the MCP server, this number may vary in the future.
   
   
   
   Qlik MCP Server is now live in Vibe.
   
   

Where the configuration is saved

All of this writes to one file: ~/.vibe/config.toml (by default /Users/<you>/.vibe/config.toml on a Mac and C:\Users\<you>\.vibe\config.toml). You don't need to touch it by hand, /mcp add does it for you, but it helps to know what "it worked" looks like on disk.

You will be able to find the following near the bottom of the file:

[[mcp_servers]]
name = "MCP Server Name"
transport = "streamable-http"
url = "https://<tenant>.<region>.qlikcloud.com/api/ai/mcp"

[mcp_servers.auth]
type = "oauth"
scopes = [
    "user_default",
    "mcp:execute",
]

If you ever want to double-check the setup, or remove it and start over, this is the block you're looking for.

Troubleshooting

"Error: No MCP servers configured." right after /mcp add (Vibe CLI). The known bug in Mistral Vibe. Exit Vibe, reopen it, run /mcp login MCP-Server-Name again. This will be resolved once Mistral ships the fix.

502 error when trying to log in (Vibe CLI). This means the CLI has been pointed at the Studio Connector instead of a server defined inside Vibe itself, usually by reusing its OAuth URL or trying to log in through it directly. It won't work, and it can't: the CLI doesn't know how to talk to a Studio Connector. The fix isn't a retry; it's going back to the setup and completing the Vibe CLI connection.

Browser doesn't open on its own during login (either Studio or Vibe CLI). Copy the URL that was printed and paste it in yourself.

/mcp status still says "needs auth" after you logged in (Vibe CLI). Make sure the browser actually showed you a success screen before you closed it. If it did and the status still looks wrong, just run /mcp login MCP-Server-Name again.

Not sure which section you need? If you're chatting with Mistral in a browser, use Studio. If you're typing into the Vibe CLI in a terminal, use the Vibe CLI section. Doing one doesn't undo or interfere with the other.

Environment

• Qlik MCP
• Qlik Cloud

We all learned the rule in grade school math: getting the right answer doesn't count if you can't show your work. Yet much of enterprise AI asks data teams to bet the quarter on a black box. Qlik Answers takes a different approach. This post goes under the hood on the architecture of the reasoning trace, how we handle schema linking, and how you can audit the engine yourself.

After asking your question, click Show Reasoning.

The Architecture: Semantic Routing and Multi-Agent Orchestration

Every natural language input runs through a multi-agent pipeline. The Answers Agent acts as the orchestrator. Rather than attempting to solve the query in a single zero-shot prompt, it uses semantic routing to decompose the question, map the required context, and dispatch work to specialized execution agents:

• Data Analyst Agent: Handles structured data and text-to-query translation. It navigates your data model, performs schema linking to identify the correct dimensions and measures, constructs the deterministic calculation expressions, and generates the resulting visualizations. For multi-part queries, this agent executes iteratively.
  
  An example of the Data Analyst Agent's reasoning
• Knowledge Base Agent: Manages unstructured sources via a RAG (Retrieval-Augmented Generation) pipeline. It retrieves, chunks, and ranks passages from connected documents and transcripts, handling vector search retrieval rather than structured data queries.

The value of the trace is observability. If a response falls short, you aren't left guessing whether it was a routing failure, a vector retrieval miss, or a bad query formulation. The orchestration view isolates the exact node where the gap occurred, putting control back in the hands of the data architect.

Field Discovery: Beyond Keyword Matching

The hardest problem in text-to-analytics is schema linking; accurately mapping a user's messy natural language to rigid database columns. When the Data Analyst Agent interprets a question, it doesn't rely on basic string matching. It uses semantic indexing.

The engine generates embeddings for the user's intent and maps them against the vector representations of your data model's metadata. The trace exposes this ranked vector retrieval list, showing the confidence score for each candidate field.

Strongly matched fields, where the vector proximity is tight, are selected directly. When two fields occupy a similar semantic space, the trace shows you which one the agent selected and why the competing field was discarded. You are observing the actual selection logic, not a post-hoc summary generated by an LLM.

Grounding: Separating Math from Generation

The highest risk in generative BI is a narrative summary that hallucinates a trend contradicting the chart right below it.

Qlik Answers prevents this by separating deterministic calculation from probabilistic text generation. The narrative is heavily grounded. It is constrained by the payload returned by the underlying associative engine. The trace carries this evidence, which references directly to the UI. Every quantitative claim in the generated text is hard-linked to the specific chart object or calculation it was derived from. You can trace any generated sentence back to the exact measure, dimension, and query formulation that produced it.

Tuning the Semantic Layer

Enterprise data models are inherently messy. No one has perfect metadata. The system is designed to work against the model you have today, and the trace makes it highly tunable.
Because field selection relies on semantic indexing, you can diagnose when the engine misinterprets a proprietary acronym or an obscure technical column name like REV_Q3_ACTL_V2. By viewing the trace, you know which fields would benefit from enriched plain-language descriptions in your semantic layer. You are optimizing the metadata that actually matters to user queries, rather than attempting to boil the ocean and document your entire warehouse.

Why the Trace is a Feature, Not a Log

Plenty of systems write backend debug logs. The difference here is that the reasoning is structured as a first-class UI component: orchestration flow, semantic schema linking, and deterministic evidence grounding, all in one view. It serves the business user who simply wants the answer, the analyst who wants to validate the logic, and the data owner who wants to tune the index. One trace, three jobs, no one forced into another's role. That is what reasoning transparency looks like when it is engineered into the product rather than bolted on after the fact.

A Practical Workflow: Trace, Diagnose, Tune

Here is how this plays out day to day.

A user asks a question, and the answer looks off. Instead of filing a vague "the AI got it wrong" ticket, open the reasoning panel on that response and walk it top to bottom:

1. Check the routing. Did the question go to the Data Analyst Agent or the Knowledge Base Agent? If a numbers question got routed to document retrieval, the fix is usually in how the question was phrased or how your assistant's scope is configured, not in the data model.
2. Check the field selection. Look at the ranked candidate fields and their confidence scores. If the agent picked Gross_Rev when the user meant Net_Revenue, the two fields are sitting too close together in semantic space. Add a plain-language description to the field that was lost, or to both.
3. Check the expression. The trace shows the calculation the engine actually constructed. This is where an analyst confirms the aggregation, filter context, and time grain match the user's intent.
4. Check the grounding. Every claim in the narrative links back to a chart object or calculation. If a sentence doesn't hold up, follow the reference, and you will find exactly which evidence produced it.

Each pass through this loop makes the next answer better, because the fixes land in the semantic layer, where every future query benefits. Teams that treat the trace as a tuning instrument rather than a curiosity see their answer quality compound over weeks, not quarters.

For more information, see Troubleshooting Qlik Answers responses with the reasoning trace .

The Bigger Point

Trust in enterprise AI isn't won with accuracy benchmarks on a slide. It is won the first time a skeptical analyst clicks into a trace, follows the logic, and finds nothing hidden. After that, the conversation changes from "can we trust it" to "how do we make it better." That shift, from auditing to tuning, is the whole reason the reasoning trace exists.

If you want to see it in action, open any answer in Qlik Answers and expand the reasoning view. And if you've used the trace to debug a field mapping or tighten up a semantic layer, share what you found in the comments. The best tuning patterns we know came from customers doing exactly that.

Environment

• Qilk Cloud Analytics
• Qlik Answers

The Model Context Protocol (MCP) is an open standard that lets AI assistants connect to external data sources and tools.

This article provides step-by-step instructions for connecting Antigravity to the Qlik MCP server in any tenant.

Step 1: Create the OAuth client to be used by Antigravity

1. In the Qlik Cloud Administration activity center, go to OAuth.
2. Click Create new and use the following values:
   • Client type: Web
   • Name and Description: Name and describe your client
   • Add the required Scopes: user_default and mcp:execute
   • Redirect url: https://antigravity.google/oauth-callback
3. Click Create and copy the ClientID and Secret 

   Do not forget to save the client secret. Qlik will not show it again.

Step 2: Create the MCP connection in Antigravity

1. In Antigravity, open Antigravity IDE by clicking Open IDE
2. Click the ellipses (...) menu in the top right corner
3. Click MCP Servers
   
   
   
   
4. Click Manage MCP servers (A) in the top right corner and View raw config (B) to open your mcp_config.json
   
   
   
   
   
   
5. Enter your configuration in the following format:
   
   

   {
    "mcpServers" : {
      "qlik-mcp": {
        "oauth": {
          "clientId": "00000000",
          "clientSecret": "00000000" 
        },
        "serverUrl": "https://YOURTENANT.qlikcloud.com/api/ai/mcp" 
      }
    }
   }

   
   
   
   
6. Save your configuration
7. Return to Antigravity, go to Settings > Customization, navigate to installed MCP servers, and click Refresh
8. Click Authenticate 
   
   
   
   
9. A redirect takes you to the Antigravity URL; copy the Authorization code
   
   
   
   
10. Paste the code and click Submit
    
    
    
    
    
    

The connection is now available in a conversation.

Environment

• Qlik MCP
• Qlik Cloud

This article provides answers to the most frequent questions asked about Qlik MCP.

For the more general Qlik Answers FAQ, see Qlik Answers Agentic Analytics FAQ.

How do I troubleshoot my connection to the LLM I’m using?

Qlik Model Context Protocol (MCP) server integrates Qlik Cloud into your LLM workflow, allowing you to work with Qlik Cloud using your LLM without having to leave your LLM. Connection issues will often be tied to misconfiguration.

• Before you begin, review Deploying and administering Qlik MCP server and Connecting to the Qlik MCP server
• If you are using ChatGPT, verify that you are on the correct ChatGPT plan (see Connecting to ChatGPT)
• Typically, OAuth authentication issues with your client will be caused by:
  • Incorrect Remote MCP server URL (<your tenant URL>/api/ai/mcp)
  • Wrong Client ID
    
• If an error code is returned, reconnect; if reconnecting does not resolve the issue, submit the code returned by MCP to your LLM vendor for further troubleshooting

How do I troubleshoot my app if I am not getting the answers I expect?  

In a case where you do not get the response you expect based on the sources, or you receive an error:

• Ask again in case of an error (see the appendix for possible error codes)
• Rewrite the prompt (see Best practices for chatting with Qlik Answers)

Has your app been prepared for Qlik Answers?

• Best practices for preparing applications for Qlik Answers
• Writing master item descriptions for Qlik Answers

How much does the Qlik MCP cost?

For now, Qlik MCP will continue to be priced based on current models for the number of questions asked. You get capacity at corresponding levels in Standard, Premium, and Enterprise editions, as well as Qlik Sense Enterprise SaaS. There is currently no additional cost for structured data questions or task automation requests; a question is a question.

Use of the MCP server consumes questions when Qlik is accessed using Tool Calls. A Tool Call is a request made by the LLM to interact with Qlik's capabilities, such as, but not limited to, querying databases, calling APIs, or performing computations. These are typically visible in the LLM's log.

For Qlik's MCP server, 5 Tool calls consume 1 question. More questions may be purchased for expanded use cases.

See Pricing and the Qlik MCP server product description for details.

Qlik’s pricing does not include your chosen LLM subscription or usage, which will need to be paid separately.

Does Qlik MCP work with my existing Qlik apps?

Yes. Qlik MCP works on top of existing Qlik Sense applications and uses the same data, logic, and security model.

But to get the best experience, apps should be prepared beforehand:

• Best practices for preparing applications for Qlik Answers
• Writing master item descriptions for Qlik Answers

What capacity limitations do we need to consider?

Your Qlik Cloud subscription determines the quota of questions asked by users. If you are licensed for Qlik Answers, both MCP and Qlik Answers will use your monthly question capacity. See Administering Qlik MCP server. 

What happens when we reach a capacity limit?

Question capacity quotes are per month and reset every month. When you hit your limit, users can no longer ask questions until the next month. Overage is only allowed, depending on your subscription. For more information, see Qlik MCP server product description.

For more information on overage, see Overage. 

How do we turn off features?

Features can be turned off for individual users through user scopes.

1. Go to your Administration Center
2. Go to Manage Users
3. Switch to the Permissions tab
4. Click User Defaults and change the scope as required

See Control access to AI features.

The Model Context Protocol (MCP) is an open standard that lets AI assistants connect to external data sources and tools.

This article provides step-by-step instructions for connecting Gemini Enterprise to the Qlik MCP server in any tenant.

Step 1: Create the OAuth client to be used by Gemini Enterprise

1. In the Qlik Cloud Administration activity center, go to OAuth.
2. Click Create new and use the following values:
   • Client type: Web
   • Name and Description: Name and describe your client
   • Add the required Scopes: user_default and mcp:execute
   • Redirect URL: https://vertexaisearch.cloud.google.com/oauth-redirect
3. Click Create and copy the ClientID and Secret 

   Do not forget to save the client secret. Qlik will not show it again.

Step 2: Create the MCP connection in Gemini Enterprise

1. Create an application in your Gemini Enterprise console
   
   
   
   
2. Create a datastore
   
   
   
   
3. Select Custom MCP server and configure it as follows:
   • MCP server URL: https://{tenant url}/api/ai/mcp
   • Authorization URL: https://{tenant url}/oauth/authorize
   • Access Token URL: https://{tenant url}/oauth/token
   • ClientId: As copied from your Qlik Cloud OAuth application
   • Client Secret:  As copied from your Qlik Cloud OAuth application
   • Scopes: user_default mcp:execute 

     Scopes are separated by a space.

     
     
     
4. Once all details are filled, click Login and approve the consent dialog
5. Click Continue and optionally enter an MCP Server Description, as well as MCP Agent Instructions
6. Click Continue and provide a Data connector name
   
   
   
   
7. Click Create
8. Select the Datastore name
   
   
   
   
9. Click Actions in the menu to the left
   
   
   
   
10. Click Refresh custom actions
    
    
    
    
11. Select all tools to enable them and click Enable actions
    
    
    
    

Step 3: Using the connection

1. Return to your application and click Configure web application
   
   
   
   
2. Click the database icon
   
   
   
   
3. In the pop-up, click Authorize
   
   
   
   
4. You are now ready to test and use the application.

Environment

• Qlik MCP
• Qlik Cloud

Enabling Data Analysis as a user who was assigned a custom role designed for Qlik Answers (Legacy) fails with:

Scope insight-advisor:experience does not exist or is not available on your tenant
Error code: IDENTITIES-10504

Resolution

Create a new custom role for Agentic Qlik Answers.

In Permissions:

1. Expand Features and Actions
2. Set Data analysis and Insights feed to Allowed 
   
   

Assign the new role to the users or groups who need to use Qlik Answers. Delete the old role if it is no longer needed. 

Cause

Custom Roles created for legacy Qlik Answers rather than the new Agentic Qlik Answers have different role permissions settings on the back end. Newly created custom roles will not run into the same permission issues as Legacy Qlik Answers.

Environment

• Qlik Answers
• Qlik Cloud

After submitting a question in Qlik Answers, the real-time reasoning output is not displayed while the question is being processed. The reasoning stepper, which normally shows the model’s step-by-step thinking in real time, is either missing or fails to load. The issue may occur intermittently and does not consistently reproduce across sessions or browsers.

Example:

Environment

• Qlik Cloud Analytics
• Qlik Answers

Cause

This is a client-side issue, rather than a defect in Qlik Answers, and is commonly caused by a browser extension.

Qlik Answers delivers its real-time reasoning output via a streaming HTTP response (Server-Sent Events / chunked transfer encoding). Browser extensions can interfere with this mechanism by intercepting, buffering, or suppressing streaming network traffic at the browser level. When any such extension disrupts the stream, the reasoning stepper never receives the data it needs to render.

Extension types known to cause this class of conflict include, but are not limited to:

• DRM and Media Proxy Extensions: Intercept low-level browser network requests to proxy DRM-protected media streams; inadvertently buffers non-media streaming responses.
• VPN and Proxy Extensions: Route all browser traffic through a proxy layer, introducing buffering that breaks Server-Sent Event connections.
• Advanced Ad and Content Blockers: Their network-level request filtering can strip or delay streaming response headers.
• Privacy and Anti-Tracking Extensions: May block or throttle requests classified as tracking-like, including persistent HTTP connections.
• Script Injectors and Page Modifiers: User scripts that intercept fetch or XMLHttpRequest can interfere with streaming response handling.
• Corporate Security and Endpoint Protection: Perform SSL inspection or traffic analysis that can break SSE streams.

Resolution

This article documents a solution for Google Chrome.

1. Verify whether an extension is the cause:
   
   
   1. Open Chrome and press Ctrl+Shift+N (Windows) or Cmd+Shift+N (Mac) to open an Incognito window.
   2. Navigate to Qlik Cloud and submit a question in Qlik Answers. As extensions are disabled by default in Incognito, the reasoning view appears. This confirms a browser extension as the root cause. 
      
      
2. Identify the conflicting extension:
   
   
   1. Navigate to the Chrome extensions page: chrome://extensions/
   2. Disable all extensions.
   3. Re-enable them one at a time, testing Qlik Answers after each, until the conflict is reproduced. Pay particular attention to extensions listed in the Cause section.
      
      
3. Once the conflicting extension is identified, choose one of the following:
   
   
   • Disable or remove the conflicting extension entirely, or
   • Allowlist Qlik Cloud in the extension’s settings.
     
     Most VPN, ad blockers, and security extensions support per-site exceptions that will resolve the conflict without requiring the extension to be removed.

If the issue persists after disabling all extensions:

The same class of interference can also be caused by corporate network proxies, SSL inspection appliances, or endpoint security agents operating outside the browser. If Incognito also fails to display the reasoning view, engage your IT or network security team to verify whether outbound Server-Sent Event connections to Qlik Cloud are being intercepted or buffered at the network level.

This conflict is not unique to Qlik Answers. Any web application that relies on Server-Sent Events or streaming HTTP responses may be affected by extensions or network components that proxy or intercept traffic at the browser level.

If you continue to experience issues after following the steps above, please open a support case and include the following:

• Browser version and operating system
• A full list of installed browser extensions
• Whether the issue reproduces in Incognito mode
• A screen recording showing the behavior

Related Content

• Answering questions with Qlik Answers | help.qlik.com
• Supported browsers for Qlik Cloud Analytics | help.qlik.com
• How to contact Qlik Support | Qlik Community

When attempting to add an application to Qlik Answers, the error invalid logical model or other indexing errors appears, followed by the process hanging without further processing or simply erroring out. 

The behavior is the same in both user-based tenants and capacity-based tenants.

The application reloads normally when scheduled or when reloaded manually. This indicates there is no issue with the application itself, which is not exhibiting any errors performing typical analytics activity.

This resolution only applies to the agentic Qlik Answers solution.

Resolution

If your application has existing business logic, your custom business logic may interfere with the expected Qlik Answers indexing preparation.

To resolve this:

1. Navigate to the Logical Model in your application; click the navigation dropdown (A), scroll down, and choose AI optimization (B)
   
   
   
   
2. Switch to the Fields & master items (C) tab, then click Reset to default (D)
   
   
   
   
3. Go to the Analytics home page and find your application
4. Reload the application
5. Enable the application for Qlik Answers; this will re-index it
6. Ask a question

If changes are made to the logical model after performing the above steps, those changes will be reindexed automatically. Keep in mind that this re-indexing may take up to 24 hours.

Cause

In some instances, business logic may need to be reset or may not be consistent with best practices, preventing Qlik Answers from indexing the Qlik application. 

See Best practices for preparing applications for Qlik Answers for details.

Internal Investigation ID(s)

SUPPORT-8526

Environment

• Qlik Answers
• Qlik Cloud Analytics

Related Content

• How to prepare your Qlik App for Qlik Answers - Qlik Community - 2542848
• Best practices for preparing applications for Qlik Answers

Modern large language models that power advanced AI capabilities, such as Qlik Answers, require processing infrastructure that may not be available in every Qlik region. To provide our AI capabilities in as many regions as possible, Qlik Cloud now offers cross-region inference as an opt-in feature. You determine if available inference locations meet your data residency and compliance requirements.

Latest Update
June 30th, 2026

Qlik Predict released in our Brazil, adding a new inference location. See Inference processing locations for Qlik Predict for details. 

Key highlights

• Regional compliance: Inference locations are tailored based on the underlying availability of AWS infrastructure.
• Transparency: Qlik provides at least 30 days' notice before any processing location changes

For complete details about inference locations for your region and data protection measures, see the full documentation available in Enabling cross-region inference.

Please subscribe to this article to receive updates to any inference location.

Environment

• Qlik Cloud

This article provides answers to the most frequent questions asked about Qlik Answers.

For the Qlik MCP FAQ, see Qlik Model Context Protocol (MCP) FAQ. 

What is Qlik Answers Agentic Analytics?

In February 2026, we launched our new agentic experience, which will enhance decision-making and improve productivity through a combination of assistants and agents running on a cutting-edge architecture. This initial release includes out-of-the-box agents for structured data analytics, unstructured knowledge, discovery of anomalies, and help and assistance. These agents take advantage of our foundational capabilities, including our data products and unique analytics engine, to execute complex, multi-step tasks in a trusted, scalable, and secure manner.

How will Qlik Answers evolve?

Qlik Answers is the primary AI assistant for people to interface with agentic AI. It will understand the intent of natural language questions and engage the underlying agentic framework to execute tasks, build responses, and take actions.

Qlik Answers now combines structured data analytics with unstructured content and general knowledge and reasoning from LLMs to deliver the most complete and relevant answers and insights, helping our customers improve decisions, productivity, and business outcomes in ways not possible before.

Looking ahead, as we build additional agents, such as prediction agents and pipeline agents, they will all be invoked through Qlik Answers. A broader set of agents is planned, all aimed at helping users get more value from their data and become more productive as Qlik continues to evolve.

What is the value to me as a customer?

With Qlik Answers now able to handle both structured and unstructured data, you can drive hundreds more informed decisions and actions each day. You can drive productivity through automation of a broad range of data and analytics tasks and workflows. And with plug-and-play simplicity, you can quickly deploy assistants in a matter of hours, reducing risk, speeding time-to-value, and future-proofing their investments in AI.

How much does the Qlik Answers cost?

For now, Qlik Answers will continue to be priced based on current models for the number of questions asked. You get capacity at corresponding levels in Standard, Premium, and Enterprise editions, as well as Qlik Sense Enterprise SaaS, with additional capacity available for purchase as needed. 

There is currently no additional cost for structured data questions or task automation requests; a question is a question.

For additional details, refer to Pricing.

What if I have a valid subscription, but cannot use Qlik Answers yet?

Since launch, Qlik Answers has been rolled out across regions. If you have Standard, Premium, and Enterprise editions, check if your region supports it (see Supported regions).

If it is not yet available to you, then:

• MCP is already available to you regardless; see Deploying and administering Qlik MCP server 
• Be on the lookout for an in-product message alerting you once your tenant has access

Do I have to be a Cloud customer to use it?

Yes, you must be a Qlik Cloud customer to use Qlik Answers. Qlik Answers is built on cloud-native technologies, specifically large language models (LLMs) that require significant compute resources and specialized infrastructure, and there is no mechanism to deploy these technologies in an on-premises environment.

However, you don’t have to fully migrate their analytics environment or documents to the cloud to take advantage of Qlik Answers. Analytics apps can be pushed to the cloud as needed to support Qlik Answers. See Qlik Answers and applications distributed from Qlik Sense Enterprise on Windows for details.

Will I get both Qlik Answers and Insight Advisor?

No. You will use either Qlik Answers or Insight Advisor, not both at the same time.

Qlik Answers represents the AI-first experience going forward. When a tenant chooses Qlik Answers, that becomes the primary way users interact with analytics. Insight Advisor is not available in parallel within the same tenant.

This is a deliberate choice to avoid duplicated experiences, inconsistent results, and user confusion.

Will Qlik Answers become available on premises?

No. Qlik Answers is cloud only.

There are no plans to bring Qlik Answers to on-premises environments. The product relies on cloud native AI services, managed infrastructure, and continuous model evolution.

What happens to Insight Advisor? Is it going away?

Insight Advisor is not being discontinued.

If you remain on Insight Advisor, you can continue using it. However, within a tenant, you must choose between Insight Advisor and Qlik Answers. You cannot run both experiences side by side.

What happens to the business logic we already built in Insight Advisor?

The most important and relevant business logic is preserved when moving to Qlik Answers.

That said, Qlik Answers is built for a newer generation of AI-driven analytics. In many cases, customers will find they no longer need to manually build or maintain the same level of logic, because the system handles more of that automatically.

The value is not in recreating everything exactly as it was, but in moving to a simpler, more capable experience.

What is the difference between Qlik Answers and Qlik MCP?

This is essentially a buy vs build decision:

• Qlik Answers is a fully managed, out-of-the-box solution. It features AI agents specifically
  designed to take advantage of our unique capabilities, including our analytics engine.
  
  Everything is preconfigured, secured, monitored, and continuously improved. Customers can start using it immediately without worrying about infrastructure, model selection, or orchestration.
  
  It’s an easy, plug-and-play solution.
  
  
• Qlik MCP is for customers who want to build and customize themselves. It is aimed at broader AI initiatives and assistants, allowing organizations to build on top of Qlik’s trusted data and analytics foundation. It’s designed for technically advanced teams that want to integrate their own models, customize deeply, and manage their own AI stack.
  
  Essentially, it’s “APIs for AI.” 
  
  
• If you want speed, simplicity, and support, Qlik Answers is the right choice. If you want full
  control and customization, and to build and operate it yourself, MCP is the right path.

What AI models does Qlik Answers use?

Qlik Answers is built on AWS Bedrock and currently utilizes Anthropic Claude models. The specific model versions vary by agent function and are continuously evaluated and updated based on performance, accuracy, latency, and cost optimization.

Our Model Selection Philosophy:

Qlik maintains flexibility in model selection to continuously improve the user experience as AI technology evolves. Different agents within the Qlik Answers architecture may use different models optimized for their specific tasks (e.g., semantic understanding, code generation, reasoning).

Can I bring my own AI model to Qlik Answers?

No. Not at this stage.

Qlik Answers is a managed experience with curated models and configurations. Customers who want to use their own models or bring custom AI stacks should use MCP instead.

Does Qlik Answers work with my existing Qlik apps?

Yes. Qlik Answers works on top of existing Qlik Sense applications and uses the same data, logic, and security model.

But to get the best experience, apps should be prepared beforehand:

• Best practices for preparing applications for Qlik Answers
• Writing master item descriptions for Qlik Answers

Does it respect master items and business logic?

Yes. Master measures and dimensions are always prioritized. If business logic exists, Qlik Answers uses it rather than creating new calculations.

Does it generate charts automatically?

Yes. Qlik Answers generates appropriate visualizations such as KPIs, bar charts, or time-based charts depending on the question.

How does security work in Qlik Answers?

Qlik Answers inherits and enforces Qlik's established security model without exception. All existing security rules, section access configurations, and row-level security policies apply automatically.

Key security principles:

• Users only see data they are authorized to access
• Section access restrictions are enforced
• Application-level permissions control which apps users can query

Field-level security (if implemented) is respected in all analyses.

No additional security configuration is required. Organizations with complex security requirements can continue using their existing Qlik security implementations with confidence.

Can different users get different answers to the same question?

Yes, if their access rights differ. Answers are always scoped to the user’s permissions.

Do I need to prepare my data differently?

While no special data preparation is required beyond standard Qlik Sense data modelling best practices, the apps themselves should be prepared beforehand to give you the best experience possible:

• Best practices for preparing applications for Qlik Answers
• Writing master item descriptions for Qlik Answers

Does Qlik Answers support follow-up questions?

Yes. Qlik Answers understands conversational context, allowing users to refine or continue their analysis.

Does Qlik Answers support multiple languages?

At its initial GA release, Qlik Answers is optimized and fully supported for English language queries and responses.

While the underlying large language models have multilingual capabilities and may be able to process queries in other languages with varying degrees of accuracy, non-English language support is not officially validated, documented, or supported by Qlik at this time.

Additional language support is planned for future releases based on demand and regional priorities.

Does it replace analysts or data teams?

No. It accelerates analysis and reduces repetitive work but does not replace human expertise or decision-making.

Can I limit which apps are available to Qlik Answers?

Yes. Only enabled and indexed applications are available.

Does it work across multiple apps?

Not in the current GA release. Qlik Answers operates within the context of a single Qlik Sense application per query. Multi-application query capabilities are planned for a future release.

What scopes do I need to be able to ask questions?

If you want to ask questions in an app, you just need the ‘Data analysis’ scope. If you plan on asking questions to an assistant, you need the ‘Data analysis’ and ‘Search knowledge base’ scopes.

What is the risk of Cross-Region Inference?

Cross-region inference has minimal risks as the data still stays within the AWS Virtual Private Network. The only difference here is that the LLM call gets processed in a different region due to GPU availability.

What about the speed of responses?

We have made a deliberate design decision to prioritize the quality of answers and insights over the speed of responses. In general, Qlik provides a far richer reasoning process and answer than competing products, and this results in a longer response time. We are planning to improve and optimize this, as well as introduce a faster mode for simpler questions in the future.

How do I troubleshoot my app if I am not getting the answers I expect?  

Qlik Answers always references its sources in detail. To begin troubleshooting, check the citations, which will show:

• The app used
• The fields and expressions used to build the visualization
• The knowledge base used
• The text chunks being referenced

In a case where you do not get the response you expect based on the sources, or you receive an error:

• Ask again in case of an error (see the appendix for possible error codes)
• Rewrite the prompt (see Best practices for chatting with Qlik Answers)

Has your app been prepared for Qlik Answers?

• Best practices for preparing applications for Qlik Answers
• Writing master item descriptions for Qlik Answers

What capacity limitations do we need to consider?

Your Qlik Cloud subscription determines the quota of questions asked by users. If you are licensed for Qlik Answers, both MCP and Qlik Answers will use your monthly question capacity. See Administering Qlik MCP server. 

What happens when we reach a capacity limit?

Question capacity quotes are per month and reset every month. When you hit your limit, users can no longer ask questions until the next month. Overage is only allowed, depending on your subscription. For more information, see the Qlik Answers product description.

For more information on overage, see Qlik Answers capacity | Question capacity. 

What functional limitations do we need to consider?

• See Qlik Answers Limitations.
• Qlik Answers will work with one app or assistant at a time. It does not search through all apps. 

How do we turn off features?

Features can be turned off for individual users through user scopes.

1. Go to your Administration Center
2. Go to Manage Users
3. Switch to the Permissions tab
4. Click User Defaults and change the scope as required

See Control access to AI features.

If you have previously enabled the feature, the entirety of Qlik’s Agentic Analytics can easily be turned off again by configuring AI features in Qlik:

1. Go to your Administration Center
2. Go to Settings
3. Go to the Tenant tab
4. Toggle Enable cross-region inference in AI features in Qlik

See Enable cross-region inference.

Appendix 

Error codes

These error codes should only be used to reference what is an expected error. Retry if you receive any of these errors.

Retry and Processing Errors

• DAA-001: Retry limit exceeded.
• DAA-002: Maximum duplicate processing attempts exceeded.

App and Document Errors

• DAA-007: App ID not found in swarm.
• DAA-008: Failed to retrieve the document.
• DAA-009: Invalid App ID.
• DAA-010: App not indexed for answers.

Chart and Sheet Errors

• DAA-016: Failed to create chart.
• DAA-017: Unexpected error while adding chart.
• DAA-018: Sheet creation failed.

Expression and Hypercube Errors

• DAA-024: Expression builder (v2) error.
• DAA-025: Hypercube length exceeded.

Semantic Search Errors

• DAA-030: No related data found in semantic search.
• DAA-031: Error while fetching semantic search results.
• DAA-032: Not enough relevant assets found in search.
• DAA-033: Validation error during semantic search.

Access Verification Errors

• DAA-050: Missing access control context.
• DAA-051: HTTP error occurred during access control request.
• DAA-052: Access control request timed out.
• DAA-053: Connection error during access control request.
• DAA-054: Access control privilege denied.
• DAA-055: Unexpected error in access control process.

When asking the Qlik Answers Documentation Assistant a question and checking the source, it throws the following error:

Cannot access the source

Resolution

This issue occurs when the Assistant user accessing the Documentation assistant does not have permission to the source. For reference, when setting up a documentation assistant, there should be two spaces:

• Assistant Data: Managed space
• Assistants: Managed space

As the Knowledge base lives in the Assistant Data space, confirm that Can consume data and Can view permissions are set:

For more information, see Qlik Answers use case: Documentation assistant.

In addition, if the Documentation assistant is consuming data from a Direct Access Gateway, confirm that the Assistant users have the permission Can Consume Data for the space where the Direct Access Gateway is installed.

Environment

• Qlik Answers
• Qlik Cloud

OpenAI/ChatGPT now sends a request redirect_uri during the OAuth authorization flow (visible as a new “Redirect” field in ChatGPT’s UI). Qlik Cloud validates that redirect_uri against the Redirect URLs registered on the associated OAuth client. If the exact URL isn’t registered, Qlik rejects the login with errors like Invalid redirect_uri, redirect_uri is not registered, code OAUTH-1 (HTTP 400). This is standard OAuth behavior and is enforced by Qlik’s OAuth endpoints. 

Qlik’s MCP administration guide explicitly instructs tenant admins to add the LLM client’s callback URL under Add redirect URLs, and it gives explicit examples (including ChatGPT):

• ChatGPT: https://chatgpt.com/connector_platform_oauth_redirect
• Claude: https://claude.ai/api/mcp/auth_callback
  
  These are the values that must be registered in the OAuth client used for the MCP connection.

There are several possibilities why this may have worked previously in your setup:

• ChatGPT had a static callback URL that we pre-programmed as a DCR workaround into a global static client ID for all Qlik tenants. ChatGPT now generates a unique callback URL on new Add (MCP Server). 
• ChatGPT changed, or now surfaces a redirect URI you must register. Qlik notes these callback URLs “come from your LLM client” (for ChatGPT, the example is https://chatgpt.com/connector_platform_oauth_redirect). If the client introduces or changes its callback, you must update the OAuth client in Qlik Cloud. [help.qlik.com]
• You previously authenticated with a different callback URL than ChatGPT uses now, so the registered value no longer matches the redirect_uri sent in the current flow. 

Resolution

1. In the Qlik Cloud Administration Center, go to your OAuth configuration and open the OAuth client ChatGPT/MCP uses (or create a new Native client for an LLM if needed).
   
   
2. Under Redirect URLs, add exactly the redirect and callback URL your LLM shows. This is retrieved from ChatGPT’s UI.
   
   Example: https://chatgpt.com/connector_platform_oauth_redirect
   
   Tip: Match it character‑for‑character (protocol https, domain, path, and any trailing slash). OAuth validation requires an exact match to what is registered. 
   
   
3. Save, then retry the connection from ChatGPT.
   
   If the OAuth client is correct and the URL matches, the browser flow should proceed to the Qlik consent and approval page and complete successfully. 

Environment

• Qlik MCP
• Qlik Cloud

Insight Advisor is not available when a tenant is using the Qlik Answers agentic experience. Use Qlik Answers instead to explore your application's data and create sheets and charts. For more information, see Qlik Answers.

You may wish to switch back to Insight Advisor if your deployment relies on the Microsoft Teams integration.

Resolution

Disable Qlik Answers.

1. Go to your Administration Center
2. Go to Manage Users
3. Switch to the Permissions tab
4. Click User Defaults 
5. Locate Features and actions, and change the Agentic AI settings to Not allowed 
   
   
   

See Control access to AI features.

Environment

• Qlik Cloud
• Qlik Answers

This article provides a practical guide for data modelers, BI admins, and analytics engineers.

Qlik Answers is a powerful solution - it lets your business users ask questions in plain language and get accurate, contextual answers directly from your data model. No dashboard navigation, no waiting on report requests. Just ask, and get an answer.

Out of the box, Qlik Answers already understands a remarkable amount of business language. But like any intelligent tool, the quality of its answers depends on the quality of what it has to work with. A data model with ambiguous field names or undocumented metrics might work fine when a developer manually hand-picks the right fields for a chart - but when an AI resolves a natural language question against that same model, those small inconsistencies start to matter.

Here’s a quick example. When someone asks “What’s our discount rate?”, Qlik Answers intelligently maps that question to fields in your semantic layer. If your model exposes Discount_Amount, Discount_Amount_Final_V1, Discount_Amount_Final_Sep24, Discount_Value, Discount1, and Discount2, the engine has to make a choice, and without clear naming, even the smartest AI can’t be sure which one you intended. It’s a signal that the model could use a little attention.

The great news is that with some straightforward preparation, you can unlock the full potential of Qlik Answers and give your users an experience that feels almost magical. This guide walks you through exactly how to get there.

What Qlik Answers Already Handles

If you’ve configured Business Logic for Insight Advisor before, you might be wondering: “Do I need to do all of that again?”

No - and that’s one of the best things about Qlik Answers. It uses an LLM-based approach that already understands common business language out of the box. Terms like “sales,” “revenue,” “customer,” “average,” and “quarter” just work. Standard aggregations, temporal concepts, and general business vocabulary are understood without any configuration on your part.

Where Qlik Answers benefits from your help is with your organization’s specific context. It doesn’t yet know that Discount1 is actually a coupon discount and Discount2 is a loyalty discount. And it can’t tell which of your three revenue fields is the current authoritative version. That is the context only you can provide.

With a few focused preparation steps, you’ll set Qlik Answers up to deliver accurate, trustworthy results from day one.

Before You Start

Three things worth doing before diving into your data model:

1. Understand your complete field list. Scan for ambiguous names (Revenue_v1, Revenue_Final), technical fields (IDs, keys, ETL flags), and fields with similar names that represent different concepts.
2. Identify your top 10–20 business metrics. For each, document the authoritative calculation, the plain-language definition, and common synonyms.
3. Talk to your business users. Find out what terms they actually use. “Churn” might mean “Customer Attrition Rate” in your model. “CAC” might not map to anything. This tells you where vocabulary mappings and renaming are needed.

Step 1: Clarify Field Names

This tends to be the highest-impact change you can make. Ambiguous field names are the most common cause of incorrect field selection.

For every group of similarly named fields, ask: do these represent different business concepts, or are they redundant versions of the same thing?

If they’re different concepts, give them distinct, business-aligned names:

If they’re redundant versions, pick the authoritative one, create a master measure if the calculation is complex, and hide the rest using Business Logic visibility controls.

Naming principles:

• Use full wording: Customer Name instead of CUST_NM_V1_Final
• Add contextual qualifiers: Customer City and Store City instead of two fields both named City1_V1_Final and City_ST
• Indicate the field’s role: use words like count, total, amount, or percentage to clarify the aggregative nature
• For Booleans, use proposition prefixes: is_active, has_churned
• Try to steer clear of cryptic abbreviations, bare adjectives, and generic nouns without context. A field called Amount doesn’t communicate what it’s an amount of.

Step 2: Streamline the Data Model

Every visible field is a candidate answer to a user’s question, so fewer irrelevant fields means fewer wrong answers. A streamlined model is also faster to index.

Hide technical fields. In Business Logic → Logical Model → Visibility, set these to Hidden:

• Primary and foreign keys (CustomerID, OrderID, ProductKey)
• ETL metadata (load timestamps, row numbers, processing flags)
• Staging and intermediate calculation fields
• Internal-use fields (prefixed with % or _ for expression use). You can also prefix field names with % in the load script to automatically hide them.

Consolidate redundant fields. If your model has Revenue_Old, Revenue_New, and Revenue_Current, users asking about “revenue” will get inconsistent results. It’s worth picking the authoritative version and hiding the rest.

Hidden fields remain fully functional for calculations, expressions, and existing charts. You’re only removing them from the Qlik Answers query scope, so nothing breaks.

Step 3: Check Date Field Formats

Time-based queries are among the most common in natural language analytics (“revenue by month,” “trends over time,” “compare this quarter to last”). If your date fields are loaded as plain text, Qlik Answers won’t recognize them as dates. That means no auto-calendar, no chronological sorting, and no correct time-based analysis.

In Data Manager or Model Viewer, check the tags on every date-related field. You want Date or Timestamp tags. If you see $ascii or Text, fix it in the load script:

Date(Date#([SourceDateField], 'MM/DD/YYYY')) as [Order Date]
Timestamp(Timestamp#([SourceTimestamp], 'MM/DD/YYYY hh:mm:ss')) as [Order Timestamp]

After fixing, test with queries like “Show me trends over time” and “Sales by month” to confirm the engine applies chronological logic correctly.

Step 4: Create Master Items with Descriptions

Master items are one of your strongest levers for improving Qlik Answers accuracy - and this is where the platform really shines. When processing questions, Qlik Answers intelligently gives greater weight to master items than to raw fields in the data model, because it recognizes that master items represent curated business intent. It’s a great example of how the engine is designed to work with you.

For each of your top metrics, create a master measure with a validated expression and a clear description. The description matters - Qlik Answers uses it to understand context and match user intent. A good description explains what the metric measures, how it’s calculated, and when to use it.

For detailed guidance on writing effective master item descriptions, see the help documentation: Writing master item descriptions for Qlik Answers.

Step 5: Configure Vocabulary (Selectively)

Qlik’s Business Logic vocabulary feature lets you define synonyms and map business terms to fields. It’s a useful tool, though you may need less of it than you’d expect. Because Qlik Answers is powered by an LLM, it already has a strong grasp of standard business terms: “sales,” “revenue,” “customer,” “average,” and “quarter” all work right out of the box. You only need to step in for the terminology that’s unique to your organization.

Where vocabulary adds value:

• Internal product codes or KPI nicknames that users reference by shorthand
• Legacy terminology that doesn’t match current field names
• Domain-specific language that is not widely used outside your industry

What to watch out for:

• Synonyms that duplicate existing field values; this adds ambiguity, not clarity
• Assigning the same synonym to multiple fields (such as “sales” mapped to two different measures)
• Vague terms like “top” or “bottom” that can be interpreted multiple and conflicting ways

Configure in Business Logic → Vocabulary. Map each synonym to a specific field or master item, and test with queries using those terms to confirm the mapping resolves correctly.

Step 6: Validate with Real Queries

It’s helpful to run representative queries across these categories and verify the results:

Use the reasoning panel. In the Source tab, click View Reasoning to see exactly which fields the engine selected and why. This is the fastest way to diagnose incorrect results and trace them back to a semantic layer issue.

For each test query, check:

• Did the engine select the correct field?
• Do the numbers match expected values?
• Are hidden fields excluded from results?
• Are master items being used where defined?
• Do time-based queries return chronologically correct results?

If a query doesn’t resolve correctly:

1. Trace it back to the model
2. Rename a field, add a vocabulary entry, hide a competing field
3. Then re-test until your top metrics consistently resolve as expected

The Bottom Line

You don’t need a perfect data model to get great results from Qlik Answers. You just need a clear one.

There’s no need to define what “revenue” or “quarter” means. By making sure your model is unambiguous, your dates are properly typed, your key metrics are defined, and your field list is clean, you’re giving Qlik Answers everything it needs to deliver the kind of instant, accurate insights your business users have been waiting for.

These are established data modeling best practices that have always mattered — Qlik Answers just makes the payoff more immediate and visible. Invest a little time in preparation, and you’ll be amazed at what your users can accomplish.

Are you ready to learn more?

• Continue with Part Two, which details what Qlik Answers is doing when it reads your apps and expands your understanding of why those six steps move the needle.
• For the complete technical reference, including detailed guidance on field naming conventions, master item descriptions, and synonym configuration, see the official documentation: Best practices for preparing applications for Qlik Answers.

Environment

• Qlik Answers
• Qlik Cloud

After upgrading to Qlik Talend Cloud Enterprise Edition R2025-08, some users reported that the tClaudeAIClient component was missing from the Talend Studio Palette.
Despite attempts to search for the component in the Palette or import it manually, they were unsuccessful.

Resolution

To restore the tClaudeAIClient component in Talend Studio, follow the steps below:

1. Open Talend Studio.
2. Navigate to Feature Manager by selecting "Help" > "Install Additional Packages" from the top menu or by using the shortcut icon for the Feature Manager on the Studio toolbar.
3. In the Feature Manager window, go to Components.
4. Locate EmbeddingAI under the available component families.
5. Select EmbeddingAI, then click Download or Install.
6. Once installation completes, restart Talend Studio.

After restarting, verify that the tClaudeAIClient component is available in the Palette under the AI family.

Cause 

The tClaudeAIClient component is a member of the AI family, which is offered through the EmbeddingAI optional feature. However, this feature may not be automatically installed or enabled by default, as it depends on the user's Studio configuration and feature synchronization settings.

Additional Notes

The EmbeddingAI package includes additional AI-related components beyond tClaudeAIClient.
If the component still does not appear after installation, ensure your Studio is synchronized with your Qlik Talend Cloud license and feature repositories.
For enterprise environments with restricted update policies, check with your Talend administrator to confirm access to optional feature downloads.

Environment

• Qlik Talend Cloud Enterprise Edition R2025-08 and later

• Talend Studio (Cloud or Local Installation)