Platform Documentation

Welcome to dataface.ai. Follow these steps to set up your account and begin analyzing data.

Sign Up & Login

Navigate to the Registration page to create an account. You can sign up using your email and password, or use Google OAuth for single sign-on (SSO). Once registered, you can Log In at any time from the top right of the landing page.

Subscription & Pricing

dataface.ai is packaged for enterprise evaluation and rollout. Upon signing up, you can request an evaluation workspace for sandbox validation and scoped datasource testing.
To upgrade your limits or access advanced features:

• Navigate to the Settings or Credits page from your Dashboard sidebar.
• Review the available packages: Evaluation for sandbox validation, Business for governed team rollouts, and Enterprise for custom security, infrastructure, and support requirements. See full details on our Pricing Page.
• Contact sales to scope contract terms, deployment architecture, credit allocation, support needs, and procurement requirements.

dataface.ai supports zero-data-movement analytics by connecting directly to your existing databases and APIs.

Connection Status & Deletion

Your connected sources appear as cards on the Dashboard.

• Status Indicators: A green dot indicates a healthy connection. If credentials expire or the host becomes unreachable, the status will turn red.
• Delete Database: Click the trash/delete icon on any datasource card to instantly remove it. This deletes all cached metadata, schema information, and connection credentials from our systems permanently.

When you connect a datasource, dataface.ai runs an automated background extraction to read your Schema (table names, column names, data types). It does not extract the underlying row data.

Viewing Extracted Metadata

You can review what the AI "knows" about your database to ensure accuracy:

1. Locate the datasource card on your Dashboard.
2. Click the Metadata / Schema button (often represented by an info or table icon).
3. A popover or modal will display the synchronized tables and columns.

If you add new tables to your database, simply click the Refresh Metadata button on the card to sync the latest structure to the AI agent.

The Chat interface is the core of dataface.ai. You can access it by clicking the floating chat button on the bottom right of the screen, or by directly entering the /chat route.

Context Locking

In the chat input box, you will notice a Database Icon. Clicking this opens the Context Linker.

• Locking Context: When the lock icon is engaged (amber/locked state), the AI is strictly confined to querying the selected databases. It will not hallucinate or attempt to query external sources.
• Adding Context: You can link multiple databases into a single chat session by clicking "Link More" in the context menu.

You have granular control over how your data interacts with Large Language Models (LLMs). Next to the send button in the chat input, you will find the Analysis Toggle.

Execution Trace Transparency

To maintain complete transparency, you can view exactly what operations the AI performed on your database. Under any data response, click the </> SQL Query button to reveal the exact native query that was executed on your system.

What actually runs between a question and an answer. Six stages, each one a place the system is allowed to refuse, which is why it can be pointed at production data. Every guarantee below is enforced in code, not policy.

1. 01classify

   Intent is resolved against the schema graph before anything is generated.

   guarantee Ambiguous intent stops here and asks, rather than guessing a column.

2. 02plan

   An engine is selected per question: a single source, or a federated plan across several.

   guarantee A cross-source answer never guesses at a join. It uses a mapping you confirmed.

3. 03generate

   SQL is written from the catalog: table names, column names, types, and a few example values per column.

   guarantee Your tables are never in the prompt. Query results reach the model only if a user turns on AI analysis.

4. 04validate

   Every table and column is checked against the live schema before execution.

   guarantee SELECT only. DDL and DML are rejected, not sanitised and run.

5. 05execute

   The query runs on your source, under credentials you issued and can revoke.

   guarantee Read-only, on your infrastructure. Nothing is staged or copied.

6. 06answer

   The result set is shaped into prose and a chart, with the SQL kept alongside it.

   guarantee The query is always visible. Nothing is asserted that you cannot re-run.