Ask your data anything In plain language, from WhatsApp or Slack

When someone types "how's the month going in the south region" on WhatsApp, the question does not reach a language model with free access to the database. First, the identity of the person asking and their permission scope are resolved. Then Metrix interprets the intent and translates it into objects in the semantic layer: "the month" is the period finance defined, "south region" is a real territory hierarchy and "how's it going" resolves to the net sales metric agreed on in discovery. Only once that translation is done is the query generated, already scoped by that person's permissions, and run against the source system or its read replica. The result comes back with the figure, the period, the filter applied and the source, and in parallel the audit log is written: who asked, when, which query ran and how many rows it returned. The answer is then formatted to fit the shape of the data (a single figure, a ranking, a time series) and delivered to the channel where the question started: WhatsApp, Slack, Microsoft Teams or the web interface. The whole thing takes seconds, not because the model is fast, but because the heavy lifting was done before anyone asked.

Metrix is not a repository: it does not copy your information into a database of its own or ask you to migrate anything; it connects to the sources you already have and queries them where they live. In practice that includes relational databases (SQL Server, PostgreSQL, MySQL, Oracle Database), analytical warehouses (BigQuery, Snowflake, Amazon Redshift, Databricks), your ERP (Tango Gestión, Bejerman, SAP, NetSuite), your CRM (Salesforce) and any system with an API or a SQL view available. Each source is connected with a read-only user and, when the production database cannot handle analytical queries, Metrix points to a replica. What defines how fresh the data is is not Metrix but the source: if your ERP consolidates billing every hour, the answer reflects that hour, and this is documented metric by metric so nobody confuses "real time" with "whatever the system loaded last". If you already have Power BI or Tableau, they are not replaced: Metrix reads the same model and coexists with the dashboards your team maintains, which are still better for exploring a long hypothesis across many dimensions.

The scene repeats itself at every company: sales brings one number to the meeting, finance brings another, and the first half hour goes into arguing about which one is right. Almost always both are right and what changes is the criterion, because one counts the sale when the order is placed and the other when it is invoiced, one includes returns and the other does not. No query tool solves that on its own, and an agent that answers fast without solving it only gets the argument started sooner. That is why the heart of the project is not the chat: it is the semantic layer, the place where each metric is defined once, with its formula, its period, its exclusions and a named owner who can change it. That is also where the hierarchies live (which branches make up a region, which SKUs make up a product family) and the synonyms people actually use when they ask. Building it is the uncomfortable part of the project and it is where the outcome is decided: it takes conversations between departments, not configuration. The payoff is that afterwards every answer, whoever asks it and through whatever channel, comes from the same criterion.

Democratizing data access without governance means opening the entire database to the whole company, so control is applied in three layers. Permissions are resolved before the query runs: the query is generated already scoped to the reach of the person asking, so a branch manager who asks about sales gets their own and does not even see that others exist. It is not a cosmetic filter on top of a full result. Sensitive columns, such as salaries, customers' personal data or margins by account, are masked by role even for someone who is allowed to see the row. And every interaction leaves a trail: user, timestamp, natural-language question, query run and volume returned, with a retention period agreed with your security team and detailed in Data Transparency. That log also works as product input, because it shows what the organization asks and what the agent still cannot answer. On infrastructure, Vantegrate does not run its own clouds: your implementation runs on Salesforce or on Oracle Cloud Infrastructure, and auditable certifications such as SOC 2 Type II and ISO 27001 belong to those platforms. In Argentina, the Personal Data Protection Law (Law 25,326) is met with configurable controls for consent, access, rectification and retention within your own implementation.

Metrix does not clean dirty data, and it is worth saying loudly: a conversational agent running on inconsistent information does not warn you that something is wrong; it answers with the same confidence as if everything were fine, and that confidence is worse than the silence of a report nobody read. If your customer master data is duplicated, if half the company works off desktop spreadsheets or if nobody can decide what an active customer is, data quality is a project in its own right: that work comes first, and no interface replaces it. Nor is it a data warehouse or an integration tool: it does not transform, consolidate or orchestrate ETL processes. It does not write to your systems; it only reads. It does not build predictive models or statistical forecasts: it answers about what happened and what is happening. And it does not replace the BI team; it takes them out of the queue of repetitive requests, and someone still has to keep modeling and looking after the data. If your use case is exploring a hypothesis for two hours across six dimensions, a dashboard is still the best tool. Metrix shines on the specific, frequent question, the one that usually gets answered by interrupting an analyst.