Conversational BI in Brazil: a platform comparison for 2026

Disclosure first: Sozo Data publishes this page and also makes Entendo, one of the platforms listed. We applied the same criteria to every vendor, ours included, and recorded only what each vendor's official page says.

We reviewed the pages on October 7, 2026. For each vendor we noted six things: who is behind the product, who it says it is for, how it reaches your data, where the data stays when the vendor says so, how you buy it, and which languages it supports. When a page does not say, we wrote "not publicly stated". We did not fill gaps with guesses or hearsay.

The rules we followed:

• A competitor's price appears only if its official page publishes it, and then it carries the date we checked.
• We do not publish Entendo pricing. Our plans are sold on request.
• We do not rank anyone as "best". Each product fits a different kind of company, and the table shows how.
• Preview and pre-launch features are labeled as such.

If you spot an outdated or wrong detail, write to us through the contact page. We will fix it and update the review date.

The products here solve the same problem: someone without technical training asks a question in plain language and gets a number, a table or a chart back. What differs is where the vendor comes from and how it delivers. We sort them into three groups.

Global suites with a copilot

Microsoft (Power BI), Salesforce (Tableau) and Google (Looker) already have BI tools installed at many companies. Each added an AI assistant on top of the existing data model. The advantage is continuity: if you already maintain reports and semantic models there, that work carries over. The cost is dependence on the suite, its licensing and, in some cases, a specific cloud capacity.

Brazilian specialist platforms

These are Brazilian products built around asking questions in Portuguese about business data. In this comparison: Luria (PX Data, Rio de Janeiro), NexisBI and Entendo (Sozo Data). They differ a lot in scale. Luria says it targets large enterprises, NexisBI targets small and midsize businesses, and Entendo targets midsize companies and partners who resell under their own brand.

Consultancies and integrators with their own product

This group includes DataTalks (Somativa), Belake.ai (Dataside) and TecnoSpeed's conversational BI. The offer comes from a services firm or from a software vendor that sells to software houses, and implementation is usually part of the sale. TecnoSpeed is a special case because it describes the product as pre-launch with a waitlist.

Who is behind it: Microsoft. Official source: Copilot overview for Power BI (Portuguese), checked October 7, 2026.

The documentation describes Copilot as a generative AI feature inside Power BI, with chat experiences for business users and help for report authors who write DAX. The Copilot pane inside a report is generally available. The standalone full-screen Copilot and Copilot in apps are in preview.

Stated requirements: a paid Fabric capacity (F2 or higher) or Power BI Premium (P1 or higher). The page says a Pro or Premium Per User license alone is not enough, and that trial capacities and free SKUs are not supported. An administrator must keep the Azure OpenAI setting on. For tenants outside the United States and the European Union, Copilot is off by default until an admin allows processing outside the tenant's region.

How it reaches data: through the Power BI semantic model. The documentation advises preparing the model first and warns that, without preparation, Copilot may misread the data. When a question does not relate to the model, it answers from the language model's general knowledge.

Language: the page says prompts in languages other than English may sometimes return relevant answers, but multilingual use is not officially supported at this time. How you buy: capacity licensing. Usage is measured in capacity units, and Copilot billing is driven by token consumption.

Who is behind it: Salesforce, which owns Tableau. Official source: Tableau Agent page (Portuguese).

According to the official page, Tableau Agent is a generative AI assistant you talk to in natural language. It supports the analysis from data prep through exploration, visualization and consumption, and it runs on Salesforce's Agentforce Trust Layer. The page does not spell out data sources, languages, licensing requirements or pricing. If your company already runs Tableau, confirm edition and licensing with your Salesforce rep before planning a pilot.

Who is behind it: Google Cloud. Official source: Conversational Analytics in Looker, checked October 7, 2026.

The documentation defines Conversational Analytics as a chat-with-your-data feature powered by Gemini for Google Cloud and grounded in Looker's modeling layer (LookML). It is available in both Looker (Google Cloud core) and Looker (original).

How it reaches data: it queries Looker Explores. A data agent can connect to up to five distinct Explores, and a query returns at most 50,000 rows. Users need permission on the underlying Explores.

Maturity: core features are generally available. Dashboard agents, agentic workflows and query feedback monitoring are in preview. The page does not detail which languages Conversational Analytics supports. Google warns that output can look plausible and still be wrong, and recommends validating results. How you buy: not stated on that page.

Who is behind it: PX Data, based in Rio de Janeiro. Official source: luria.ai, checked October 7, 2026.

Luria says it serves large organizations in retail, shopping centers, energy, utilities, pharmaceuticals and financial services. Listed connectors include Google BigQuery, Databricks and Snowflake, the databases PostgreSQL, SQL Server, MySQL and Oracle, plus Power BI, Metabase, Google Sheets and Excel.

On data location, the page states the data stays in the customer's infrastructure and that Luria processes only the queries it needs. The platform calls itself model-agnostic and lists Google Vertex AI, Azure AI, AWS Bedrock, OpenAI and Anthropic Claude. Stated languages are Portuguese, English and Spanish.

How you buy: through Google Cloud Marketplace with unified billing, or by direct sale. The page publishes no price.

Who is behind it: Somativa. Official source: somativa.com.br/datatalks, checked October 7, 2026.

The page addresses decision-makers in finance, operations, data and IT who want answers without a reporting queue. Stated connectors are a native Omie connector, a consultative integration with Dremio, and data lake environments. For SAP or proprietary ERPs, technical feasibility is assessed in an initial diagnosis.

On data and permissions, the page says raw data stays in the company's infrastructure with no external transfer, and that the system filters information by the user's permissions before sending only the necessary context to AI models. Access control is role-based, and responses carry permission metadata for auditing. The architecture is described as model-agnostic, using GPT, Gemini and Claude, with business logic held in a semantic layer.

How you buy: it starts with a 30-minute consultative conversation, and rollout ranges from weeks (Omie) to custom timelines for data lakes. The page publishes no price.

Who is behind it: Dataside. Official source: dataside.com.br, checked October 7, 2026. A Microsoft Marketplace listing also exists, but we could not open it during the review. The belake.ai address did not respond on October 7, 2026.

Dataside's site describes Belake.ai as a generative AI solution that turns data into conversational interfaces. The featured cases cover document analysis in the legal sector and WhatsApp virtual assistants in sales. The page does not detail the underlying technologies, and the site places Belake.ai under Dataside's AI agents services.

How you buy: the customer implements a tailored solution and requests a quote from Dataside's specialists. The home page publishes no price. The official pages do not say which databases the product supports, so we record that as "not publicly stated".

Who is behind it: NexisBI, a Brazilian software-as-a-service company. Official source: nexisbi.com.br, checked October 7, 2026.

The site addresses entrepreneurs, finance directors and business owners who want to automate reporting without technical skills. It claims more than 120 integrations, including ERPs (TOTVS, Omie, Bling, SAP S/4HANA), e-commerce (Shopify, Mercado Livre, Nuvemshop), financial services, and data platforms such as BigQuery, PostgreSQL and MySQL. It also describes spreadsheet upload and querying and editing data over WhatsApp.

Price published on October 7, 2026: Starter at R$ 749 per year, Pro at R$ 1,190 per year and Scale at R$ 2,190 per year, with limits on projects, dashboards and users. The page does not say where data is stored. It cites only AES-256 encryption and LGPD compliance.

Who is behind it: TecnoSpeed. Official source: TecnoSpeed blog article (Portuguese), checked October 7, 2026.

The stated audience is software houses and their customers. The article describes integration through SQL mapping and use through familiar channels such as WhatsApp, with questions in Portuguese. The product is in pre-launch, with a waitlist and pre-sale terms. The text discusses benefits such as lower cost, differentiation and customer retention, and gives no price or plans.

Because it is not yet available, treat this entry as an announcement. The source says nothing about where data stays, which databases are supported, or how per-user permissions work.

Who is behind it: Sozo Data, a Brazilian company founded in 2024. Official source: entendo.sozodata.com.br. Since we make it, the details below come from our own documentation, and we invite you to verify them.

Entendo takes a question in Portuguese or English and returns a number, table or chart, with the generated SQL and the reasoning visible. The stated audience is midsize Brazilian companies with data in a database, usually the ERP's database, and software houses and consultancies that want to offer conversational BI to their own clients under their own brand. Our team assists with the rollout.

How it reaches data: it connects to PostgreSQL, MySQL, SQL Server, Oracle, Firebird and ClickHouse, plus Google Sheets, CSV and Excel. Today we connect to the database behind the ERP, with no official integration from ERP vendors. Connectors for other databases, data lakes and systems that expose their data through an API or equivalent format are built on demand and scoped in the project. Entendo Gateway is an agent installed inside the customer's network that opens an encrypted outbound connection and needs no inbound port. The database credential stays in the customer's network, and only the query result goes to the cloud.

Controls: read-only, with write commands blocked before execution, and every question and SQL statement logged. SSO through OIDC on the Business and Enterprise plans. AI models from OpenAI, Anthropic and Google, with bring-your-own-key on the Pro, Business and Enterprise plans. It includes a semantic layer with glossary, goals and KPIs, dashboards, exports, and partner white-label in production.

How you buy: monthly plans on request, assisted onboarding billed separately, and resale terms on request for partners. What we do not offer today: no WhatsApp channel in production and no ready-made cloud warehouse connectors such as BigQuery or Snowflake: connectors for warehouses and data lakes are built on demand.

Every vendor on this page, Entendo included, should answer these six questions clearly and in writing.

1. Where does the data stay? Ask whether result rows leave your network, which cloud they go to and how long they are kept. "Your data stays with you" can mean the credential, the raw data or the query result.
2. How do I see the SQL? Ask to see the generated query and the reasoning on a real question. Without that, nobody can audit a number headed to the board.
3. How do per-user permissions work? Test with two users of different roles asking the same question. Each should see only what they are allowed to see.
4. Which databases and systems? Confirm your database by name and version. Ask whether a connector exists for the system you use or whether the connection goes straight to the database.
5. Who does the rollout? Find out whether the vendor, a partner or your own team implements it, and what that costs on top.
6. How is usage billed? Ask about limits on questions, users or tokens, and who pays for the AI model.

Also ask for a demo using a hard question from your own routine, not the vendor's examples.

Not every company needs to switch tools. Staying with your current suite often makes sense when:

• your reports and semantic models are already organized and your team maintains them;
• your company already has the capacity the copilot requires (for Power BI, Fabric F2 or higher, or Premium P1 or higher);
• most questions revolve around existing reports, not data that sits in no model yet;
• IT prefers a single vendor and a single contract.

A specialist platform tends to weigh more when your data sits in ERP databases with no semantic model, when you do not want to maintain cloud capacity just for the copilot, or when a software house or consultancy must offer the feature to clients under its own brand. In many cases the two coexist: your current BI keeps the fixed reports and the conversational platform handles ad hoc questions.