Skip to content

As of November 2026, six of the 15 BI AI assistants we checked (from 14 vendors) can answer questions from raw warehouse or database tables with no semantic model built first: Sigma Assistant, Metabase Metabot, Hex Threads, Zoho Analytics Ask Zia, Fabric data agents, and Basedash. Six more answer only through a modeled layer. Power BI Copilot needs a semantic model, Looker needs LookML Explores, and ThoughtSpot needs a Spotter-enabled Model. Omni needs its model, Qlik needs a Qlik app, and Snowflake CoWork needs a semantic view. The plan matters as much as the data: Power BI Copilot needs paid Fabric capacity, and Tableau Agent needs Tableau Cloud+ or the Tableau+ bundle.

This is a reference table, not a ranking. It answers the setup questions that come before any accuracy test: what each assistant can query, what an admin has to build first, where the query runs, and which plan turns it on. For how the answers compare in practice, see our conversational analytics comparison.

How we compiled this

We read each vendor’s official documentation and pricing pages on November 22, 2026 and recorded what they state. We did not run the tools for this piece. Each cell is either stated in vendor docs or marked “not documented”. We did not infer anything from marketing pages or third-party reviews. Where two official pages disagree, the conflict is noted in that tool’s section.

We collected the same attributes for every assistant:

  1. What it answers from: raw tables, a semantic model, or a curated object such as a topic or space.
  2. Modeling prerequisite: what must exist before a business user can ask a first question.
  3. Where the query runs: live against your warehouse, or against an in-memory engine or extract.
  4. Admin setup and defaults: tenant settings, AI provider keys, permissions, and whether AI is on by default.
  5. Security: whether the assistant enforces the asking user’s row-level security.
  6. Plan and price: which edition unlocks the assistant, with list prices as of November 2026.

Which BI AI assistants can query raw tables without a semantic model?

Tool and assistant What the assistant answers from Model needed before the first question Where the query runs Can you see the query
Power BI Copilot Power BI semantic models Yes, a semantic model Semantic model (Import in memory, or DirectQuery to the source) Yes, DAX
Fabric data agent Up to 5 sources: lakehouse, warehouse, semantic model, KQL database, mirrored database, ontology No, you select tables Live, read-only SQL, DAX, or KQL Not documented
Tableau Agent The workbook’s connected data (extracts recommended, live supported, no cubes) No, but it works inside authoring Tableau extract or live connection Not documented
Looker Conversational Analytics LookML Explores, up to 5 per agent Yes, LookML Live, through the LookML model Yes, Looker query
Amazon Quick Sight Topics (up to 12 datasets), dashboards, datasets Yes for Topic Q&A, a topic SPICE, or Direct Query on Redshift, Athena, S3 Tables, Snowflake, Databricks Yes, SQL
Qlik Answers Qlik apps, indexed on every reload Yes, a Qlik app Qlik associative engine Reasoning trace
ThoughtSpot Spotter ThoughtSpot Models Yes, Spotter enabled per Model Live SQL in your warehouse Not documented
Sigma Assistant Warehouse tables or Sigma data models No Live on your warehouse Yes, View SQL
Omni Agent Topics or views in the Omni model Yes, an Omni model Live SQL Yes, inspectable query
Metabase Metabot Tables, metrics, questions, models, dashboards No Your connected database Yes, SQL
Zoho Analytics Ask Zia Workspace tables and data model No Not documented Yes, SQL in Query Tables
Databricks Genie Unity Catalog objects attached to a Genie Agent, up to 50 A Genie Agent (curated table set), not a semantic model Live, read-only SQL on a SQL warehouse Yes, Show code
Snowflake CoWork Agents with semantic views or Cortex Analyst models Yes, a semantic view and an agent Live SQL on your warehouse Traceability to queries
Hex Threads Semantic models and warehouse tables No (with Governed mode off) Live SQL Yes, backing notebook
Basedash chat Every connected source you can access, plus models and AI context No Live SQL on your database or warehouse Yes, Sources panel

Three assistants sit in between. Databricks Genie needs a curated Genie Agent of up to 50 tables, though not a semantic model. Tableau Agent works on whatever data source the workbook you’re authoring uses. Quick Sight chat can use datasets and dashboards, while Topic Q&A needs a topic.

Two patterns stand out. First, the assistants that need a model first are mostly the ones built on an existing semantic layer: LookML, Power BI semantic models, ThoughtSpot Models, Omni models, and Qlik apps. The assistant reuses definitions you already maintain. If you don’t have that layer, you have to build it before the assistant is useful. Second, query location varies. Power BI Import models and Qlik apps answer from an in-memory engine, while Looker, ThoughtSpot, Sigma, Genie, Snowflake, Hex, and Basedash send SQL to the warehouse at question time. That affects freshness and warehouse cost. For more on where metric definitions should live, see BI tools with built-in semantic layers.

What admins must set up before the first question

Tool Admin setup AI on by default Enforces user RLS Can build a dashboard Documented limits
Power BI Copilot Workspace on F2+ Fabric or P1+ Premium capacity; Q&A on for DirectQuery and Direct Lake models Yes, in the Fabric admin portal (US and EU) Yes Yes, report pages (not on live Analysis Services or streaming models) English only
Fabric data agent Same capacity; Copilot and Azure OpenAI tenant switches; pick tables, add instructions Follows Copilot tenant switch Yes, RLS and CLS Not documented 25 rows by 25 columns per answer; 5 sources
Tableau Agent AI turned on in site settings; Creator or Explorer site role No Not documented (Pulse: yes) No English plus a subset of languages
Looker CA Enable Gemini in Looker and Conversational Analytics; grant permissions No Yes, access grants and user attributes No, opens results as an Explore 5 Explores per agent; 50,000 rows per query
Quick Sight Enterprise authors build topics; add column descriptions, synonyms, semantic types Not documented Yes, dataset RLS Adds visuals to an analysis 12 datasets per topic; not in all regions
Qlik Answers Opt in to cross-region inference; make apps available App access on; cross-region off Yes, Section Access Yes, sheets and charts English only; 200 data points per chart; monthly question allowance
ThoughtSpot Spotter Enable AI in Admin portal; grant Spotter privilege; enable per Model Off for existing instances Yes, RLS and CLS Yes, Liveboards via SpotterViz Under 100 columns per Model recommended
Sigma Assistant Configure an AI provider (warehouse model or your own key); grant Use Assistant Yes, once a provider exists Yes, RLS and CLS Yes, build mode Indexes up to 2,000 data sources
Omni Agent Querier-level permission on a modeled connection Yes Yes, Omni permissions Yes 5 web searches per turn
Metabase Metabot Connect an AI provider (own key on self-hosted; managed AI on Cloud) Yes, but idle without a provider Yes, user permissions Saves charts to dashboards Checks first 100 tables if none is named
Zoho Ask Zia Account Admin configures Manage AI and access control Not documented Yes, role permissions Yes Not documented
Databricks Genie Create a Genie Agent on a pro or serverless SQL warehouse; grant SELECT; partner-powered AI on Not documented Yes, Unity Catalog row filters and masks Saves charts to dashboards 50 tables per agent
Snowflake CoWork Create a semantic view and an agent; grant USAGE and SELECT Yes, CORTEX_USER granted to PUBLIC Yes, row access policies Not documented 2,048-token SQL generation limit
Hex Threads Admin creates a data connection; Explorer role or higher Yes Only through warehouse OAuth (Enterprise) Yes, generative apps Not documented
Basedash chat Admin connects a data source; AI context generated from schema Yes, all plans Postgres and Supabase RLS; schema, table, column rules Yes 50 steps per question

Two requirements are easy to miss during a trial: Power BI’s capacity requirement and Snowflake’s trial-account rule. Microsoft says a Pro or Premium Per User license “alone isn’t sufficient” and that trial capacities aren’t supported. Snowflake turns AI features off on self-service trials until a credit card is added.

Which plan unlocks each assistant, and what it costs (November 2026)

Tool Plan that unlocks the assistant List price as of November 2026 Trial includes AI
Power BI Copilot Paid Fabric F2+ or Premium P1+ capacity F2 about $263/month pay-as-you-go in East US ($0.18 per CU-hour), plus Pro seats No, trial capacities excluded
Tableau Agent Tableau Cloud+ or Tableau+ bundle Contact sales; Standard ($15) and Enterprise ($35) per user per month don’t include it Yes, Tableau Cloud trial
Looker CA Looker (Google Cloud core or original) Licenses: call sales. CA unbilled through Jan 31, 2027, then $3 per 1M input and $20 per 1M output tokens above the allocation Not documented
Quick Sight Enterprise to author topics Professional $20, Enterprise $40 per user per month, plus $250 per account 30 days, up to 25 users, Enterprise features
Qlik Answers Every Qlik Cloud Analytics plan Starter $300/month (10 users), Standard $825/month, Premium $2,750/month Yes, 30 days
ThoughtSpot Spotter Pro (not Essentials) Pro $50/user/month with 25 Spotter queries; usage-based Pro from $0.10 per credit 14-day trial; AI not stated
Sigma Assistant All license tiers Quote only; Assistant uses no Sigma credits; agents 0.1 credit per message since Nov 1, 2026 7-day trial; AI not stated
Omni Agent Included AI credits No public pricing; $1.00 per credit over the allotment Not documented
Metabase Metabot Every plan, including open source Starter $100/month (5 users), Pro $575/month (10 users); managed AI $3.75 per 1M tokens, 1M included 14-day trial; AI not stated
Zoho Ask Zia Every plan, including Free Basic $24, Standard $48, Premium $115, Enterprise $455 per month billed yearly 15-day trial; AI not stated
Databricks Genie A Databricks workspace with a pro or serverless SQL warehouse User usage free through Jan 31, 2027; SQL warehouse compute billed Not documented
Snowflake CoWork A Snowflake account (no seat fees) AI credits $2.00 (global) or $2.20 (regional), plus warehouse compute Only after a card is added
Hex Threads Every plan, with credit grants Professional $36, Team $75 per editor per month; Community free with 5 credits 14-day Team trial; AI not stated
Basedash chat Every plan Startup $1,000/month plus AI usage, up to 25 users, $1,000 AI credits included; Enterprise custom Yes, 14 days, no card

For seat limits and AI costs at the team level, see our BI tool pricing breakdown.

Tool-by-tool notes

Power BI Copilot and Fabric data agents

Copilot answers from the open report’s semantic model and shows the DAX it ran. It can’t create report pages on models connected live to Analysis Services or on real-time streaming models. Fabric data agents are the raw-table option, but they are read-only and cap each answer at 25 rows and 25 columns. Both need F2+ or P1+ capacity.

  • Best for: organizations with governed Power BI semantic models already in Fabric.
  • Not ideal for: teams on Pro licenses without capacity, or non-English users.

Tableau Agent

Tableau Agent works on the data in the workbook you’re authoring and recommends extracts, but it also supports live connections. It builds vizzes and calculations, not dashboards. Pulse metrics need a single published data source. Tableau Next’s Conversational Analytics agent (formerly Concierge) needs a Tableau Semantics model and starts at $40 per user per month on the pricing page.

  • Best for: Tableau authors who want help building vizzes.
  • Not ideal for: business users who want a chat answer without opening authoring.

Looker Conversational Analytics

Looker’s assistant doesn’t query the database directly. It picks LookML fields, filters, and limits, then Looker runs the query. That keeps answers on governed definitions, but anything not modeled in LookML is out of reach. Token billing above the per-user allocation starts February 1, 2027, per Looker pricing.

  • Best for: companies with a mature LookML model.
  • Not ideal for: teams without LookML developers.

Amazon Quick Sight

Topics hold up to 12 datasets and can’t mix SPICE with Direct Query. Quick chat doesn’t yet support Direct Query datasets on sources outside the topic list. Authoring topics requires Enterprise, and the account fee is $250 per month on Quick pricing.

  • Best for: AWS-centric teams on Redshift, Athena, or S3 Tables.
  • Not ideal for: teams that want chat on any warehouse without curating topics.

Qlik Answers

Qlik Answers replaces Insight Advisor in tenants on the agentic experience and answers from Qlik apps. It respects Section Access, but it requires an opt-in to cross-region inference. Every plan on Qlik pricing includes a monthly question allowance that grows by edition. One admin page still describes Answers as an add-on, which conflicts with the pricing page.

  • Best for: existing Qlik shops with apps already built.
  • Not ideal for: teams that must keep inference in-region.

ThoughtSpot Spotter

Spotter works on Models. Admins enable AI, grant the Spotter privilege, and switch it on per Model. Queries run as SQL in your warehouse, and ThoughtSpot says the warehouse returns only results to ThoughtSpot. Pricing lists Spotter on Pro, and the Essentials feature list omits it.

  • Best for: governed self-service on Snowflake, Databricks, or BigQuery.
  • Not ideal for: teams that want to ask about raw tables before modeling.

Sigma Assistant

Sigma Assistant can use warehouse tables directly for users with the View connections permission, and View SQL shows the exact query. You must bring the model: a warehouse-hosted model (Cortex, Databricks, Vertex) or your own OpenAI, Anthropic, Gemini, or Bedrock key. Redshift, AlloyDB, and Azure SQL don’t support the warehouse-hosted option.

  • Best for: Snowflake, Databricks, or BigQuery teams that want chat on raw tables.
  • Not ideal for: teams that can’t procure their own LLM provider.

Omni Agent

The Omni Agent routes questions through topics or any view in your model and builds dashboards on request. AI settings are on by default. Omni has no public price list. The usage docs bill $1.00 per credit beyond the included allotment.

  • Best for: teams that want a shared model and AI in the same tool.
  • Not ideal for: buyers who need published list prices.

Metabase Metabot

Metabot searches tables, metrics, models, and dashboards, and it generates SQL that you run yourself. Self-hosted instances need their own AI provider key. Managed AI is Cloud-only at $3.75 per 1M tokens on Metabase pricing. Row and column permissions need Pro.

  • Best for: teams that want AI on raw tables with an open source option.
  • Not ideal for: large schemas, since Metabot checks only the first 100 tables unless you name one.

Zoho Analytics Ask Zia

Ask Zia is available on every plan, including Free. It runs on Zoho’s own LLM by default, or on OpenAI through Zoho credits or your own key. It creates reports and dashboards from your workspace’s data model, and it respects role permissions. Prices come from Zoho Analytics pricing.

  • Best for: small teams that want low-cost AI on imported data.
  • Not ideal for: teams that need documented live-warehouse query behavior.

Databricks Genie

A Genie Agent (formerly a Genie space) holds up to 50 Unity Catalog tables, views, or metric views. Genie won’t query anything not attached to the agent. Queries are read-only SQL on a pro or serverless warehouse, and you can save charts to dashboards. User usage is free through January 31, 2027, per the Genie docs.

  • Best for: Databricks lakehouse teams.
  • Not ideal for: data that lives outside Unity Catalog.

Snowflake CoWork

Snowflake CoWork (formerly Snowflake Intelligence) agents answer through semantic views, Cortex Search, and tools. An agent with no tools can only use the base model. Billing is in AI credits plus warehouse compute for every generated query.

  • Best for: Snowflake-only stacks with semantic views.
  • Not ideal for: multi-warehouse teams.

Hex Threads

Threads answers from semantic models and warehouse tables. With Governed mode off, it can use all data the user may access. Each thread is backed by a notebook, so analysts can audit the logic. Row-level security only comes through warehouse OAuth, which is Enterprise-only and limited to Snowflake, Databricks, and BigQuery.

  • Best for: analyst teams that want notebook-grade auditability.
  • Not ideal for: per-viewer RLS below Enterprise.

Basedash

Basedash chat writes SQL, runs it against your connected database or warehouse, and shows every query in a Sources panel. It works without a model: AI context is generated from your schema, and models are optional for verified metrics. It can build whole dashboards from a prompt. Direct connections cover Postgres, MySQL, SQL Server, Snowflake, BigQuery, Redshift, ClickHouse, and Databricks, among others, and SaaS data syncs through connectors. Row-level security is Postgres and Supabase only. Other sources use schema, table, and column rules that also hide objects from the AI. SSO and SCIM are Enterprise features. See the AI data analyst page and pricing.

  • Best for: teams that want chat and dashboards on raw tables without building a semantic layer first.
  • Not ideal for: per-viewer warehouse RLS on Snowflake or BigQuery, notebook workflows (Hex), or deep enterprise governance (Looker, ThoughtSpot).

Which assistant fits your data setup?

  • Your data is in Snowflake or BigQuery and you have no semantic layer yet: Sigma, Hex, Metabase, and Basedash answer from tables on day one. Sigma requires your own AI provider. Snowflake CoWork requires building semantic views first.
  • You already maintain LookML, Power BI models, or ThoughtSpot Models: start with the assistant that reads them (Looker CA, Copilot, Spotter), since it reuses definitions you already trust.
  • Your data is in an application database like Postgres or MySQL: Metabase and Basedash query these directly. Basedash also passes user groups to Postgres RLS policies.
  • You’re a Microsoft shop on Pro licenses: budget for an F2 capacity before planning around Copilot.
  • You need each viewer’s warehouse permissions enforced: Databricks Genie (Unity Catalog), Snowflake CoWork (row access policies), and Hex on Enterprise OAuth run queries as the user.
  • You want the lowest published entry price: Zoho Analytics puts Ask Zia on every plan, including Free. Metabase puts Metabot on every plan, including open source with your own AI key.

Frequently asked questions

Which BI software has a built-in generative AI assistant?

As of November 2026, every major BI tool ships one. Power BI has Copilot, Tableau has Tableau Agent, and Looker has Conversational Analytics. ThoughtSpot has Spotter, Sigma has Sigma Assistant, and Omni has the Omni Agent. Metabase has Metabot, Qlik has Qlik Answers, and Zoho Analytics has Ask Zia. Amazon Quick Sight has Quick chat and Topic Q&A. Databricks has Genie, Snowflake has CoWork, Hex has Threads, and Basedash has chat. They differ less in having AI than in prerequisites: some need a semantic model or a paid capacity before the first question works.

Which AI-enabled BI tools offer the smoothest plug-and-play connection to Snowflake and BigQuery?

Tools whose assistant can read raw warehouse tables have the shortest path. Sigma, Hex, Metabase, and Basedash all query Snowflake and BigQuery tables directly once a connection exists. Sigma also needs an AI provider configured. ThoughtSpot and Omni connect to both warehouses but answer through a Model or Omni model, so expect modeling work first. Snowflake CoWork is native to Snowflake but needs semantic views and an agent, and it answers from data in Snowflake rather than BigQuery.

Do I have to build a semantic model before a BI AI assistant can answer questions?

Yes, for Sigma Assistant, Metabase Metabot, Hex Threads (with Governed mode off), Zoho Ask Zia, Fabric data agents, and Basedash. These assistants read table and column metadata and write SQL directly. Accuracy on ambiguous metrics such as revenue or active users still improves when you add definitions. Most of these tools let you add models, glossaries, or AI context later. Looker, ThoughtSpot, Omni, Qlik, Snowflake CoWork, and Power BI Copilot answer only through their modeled layer. Databricks Genie sits in between: it needs a curated set of tables, not a semantic model.

What’s the typical setup time for a SaaS BI tool that translates natural language into SQL?

No vendor publishes a setup time, but the prerequisites tell you which tools are quick. If the assistant reads raw tables, setup is a data connection plus any AI provider key, which is usually same-day work. If it needs LookML, Power BI semantic models, ThoughtSpot Models, Snowflake semantic views, or Quick Sight topics, budget for modeling before users get reliable answers. That modeling time depends on how many metrics and tables you need covered.

Which BI AI assistants are turned on by default?

Per vendor documentation as of November 2026: Power BI Copilot is on by default in the Fabric admin portal for US and EU tenants. Snowflake grants CORTEX_USER to PUBLIC by default, and Omni, Hex, and Metabase enable AI by default. Metabot still needs an AI provider before it answers. Tableau Agent and Looker Conversational Analytics are off until an admin enables them. Spotter 3 is off by default for existing ThoughtSpot instances, and Qlik’s agentic Answers requires a cross-region inference opt-in.

Which BI tools create a chart automatically when you ask a question?

Most assistants in this list return a chart with the answer: Power BI Copilot, Looker CA, Quick Sight, Qlik Answers, Spotter, Sigma, Omni, Metabot, Ask Zia, Genie, CoWork, Hex, and Basedash. Fewer build a whole dashboard from one prompt. As of November 2026, the docs say Power BI Copilot (report pages), ThoughtSpot SpotterViz (Liveboards), Sigma build mode, Omni, Zoho, Hex generative apps, and Basedash can. Tableau’s docs say Tableau Agent can’t build dashboards.

Written by

Rachel van der Lugt avatar

Rachel van der Lugt

Founding Enterprise GTM at Basedash

Rachel van der Lugt is founding enterprise GTM at Basedash, where she leads enterprise go-to-market for an AI-native business intelligence platform. Her work focuses on helping larger teams evaluate, adopt, and roll out Basedash for governed analytics across the organization.

View full author profile →

Basedash lets you build charts, dashboards, and reports in seconds using all your data.