M↓ View as Markdown ↗

How AI Analyst handles your data¶

This page explains exactly what happens to your data when you use AI Analyst — from the moment a user asks a question to the answer that appears on screen. It is written for IT, security, and governance teams evaluating the platform.

Info

In short: AI Analyst reads from your database using encrypted, read-only connections. Your data is never copied to our systems. Nothing is ever used to train AI models.

Core guarantees¶

🔒 Encrypted connections — All communication between AI Analyst and your data source uses TLS encryption in transit. Data at rest is encrypted using AES.

👁️ Read-only access — AI Analyst only ever reads from your data source. It cannot write, update, or delete any records.

🚫 No data is copied — Your data stays in your own database; we never take or store a copy of your database, tables, or their contents. We only retrieve the specific rows needed to answer a question. Those results are shown to the user and, as described under "Traces for debugging" below, may also be captured in short-lived engineering diagnostic traces — they are not otherwise stored or persisted by AI Analyst.

🤖 Not used for AI training — Your data, your questions, and your results are never used to train or fine-tune any AI model.

How your questions are answered¶

When a user asks a question, AI Analyst does not send your data to an AI provider and ask it to search through everything. Instead, it follows a precise, traceable process grounded in your semantic layer.

sequenceDiagram
    participant U as User
    participant AA as AI Analyst
    participant SL as Semantic Layer
    participant DB as Your Database
    participant LLM as AI Provider

    U->>AA: Asks a question in natural language
    AA->>SL: Looks up relevant metrics, models & business definitions
    AA->>LLM: Sends question + semantic layer metadata (no raw data)
    LLM-->>AA: Writes a SemQL query
    AA->>DB: Executes compiled SQL (read-only)
    DB-->>AA: Returns query result (specific rows only)
    AA->>LLM: Sends result rows for summarisation
    LLM-->>AA: Produces a natural language answer
    AA-->>U: Displays answer with full source traceability

What this means in practice:

The AI provider never has direct access to your database
Your database credentials never leave your infrastructure
The AI provider only receives: the user's question, the relevant business definitions from your semantic layer (metric names, descriptions, relationships), and the specific rows returned by the query
All queries are compiled from SemQL — a structured semantic query language — and executed read-only against your data source

How Steward works with your data¶

Steward is the AI agent that helps data teams build and maintain the semantic layer. It has a different, and broader, interaction with your data than the analyst agents, because building a trustworthy semantic layer requires understanding the real data, not just its structure.

sequenceDiagram
    participant DT as Data Team
    participant ST as Steward
    participant SL as Semantic Layer
    participant DB as Your Database
    participant LLM as AI Provider

    DT->>ST: Asks Steward to build or improve a model
    ST->>SL: Reads existing model & metric definitions
    ST->>DB: Reads table structure (column names & types only)
    ST->>DB: May run read-only ad hoc SQL queries to inspect or validate real data
    ST->>LLM: Sends schema metadata, query results, and semantic layer context
    LLM-->>ST: Proposes changes to the semantic layer
    ST-->>DT: Shows plan for review
    DT->>ST: Approves
    ST->>SL: Applies changes to the semantic layer

What Steward accesses:

The structure of your tables — column names, data types, and relationships — to understand what can be modelled
Your existing semantic layer definitions, to ensure new suggestions are consistent
Actual row-level data, when it runs a read-only, ad hoc SQL query to explore a table or validate a proposed model or measure. This can include real values from your tables, including PII if present, and those values may be sent to the AI provider as part of that exploration/validation step

What Steward does not do:

Write, update, or delete any data — all database access is strictly read-only
Persist query results beyond the session — ad hoc query results are not stored by AI Analyst once the session ends

Steward is a tool for your data team, used under the same access controls (roles/permissions) your organization grants that team to the underlying data source — it is not restricted to schema-only access. If you need to guarantee Steward (or any data-team user) never sees certain sensitive columns at all, exclude or mask them upstream (e.g. a scrubbed database view) rather than relying on Steward not looking.

Info

Unlike the analyst agents (which only ever see the specific rows returned by a compiled, semantic-layer-governed query), Steward can run its own exploratory SQL against your data source. Treat Steward access the same way you'd treat direct, read-only database access for a data team member.

What is stored and where¶

All data stored by AI Analyst lives in a PostgreSQL database hosted on Google Cloud Platform in the European Union.

What we store:

Conversation history — the questions users ask and the answers AI Analyst provides
Your semantic layer — model definitions, metrics, dimensions, business logic, and glossary terms
AI Analyst configuration — instructions, access settings, saved prompts
Audit logs — a record of who did what and when, for your compliance needs

What we do not store:

Copies of your database, tables, or their contents
Any raw data from your connected data sources, outside of the diagnostic traces described below

Traces for debugging:

We keep diagnostic traces of AI interactions in a secure logging service (Langfuse), including the queries run and their result rows, so engineering can debug issues. PII in these traces (emails, phone numbers, SSNs, credit card numbers, IBANs) is automatically redacted before being sent; other query data is not, since it's needed to debug the query itself. Traces are used solely by engineering, never used to train or fine-tune any AI model, and are covered by the same data processing agreements as the rest of the platform.

AI providers¶

AI Analyst uses large language models from leading AI providers — including Microsoft Azure, Anthropic, and Google Cloud — to process questions and generate answers. The provider used at any given time depends on the nature of the task.

All providers operate under standard data processing agreements. None of them use your data for model training.

Summary¶

	AI Analyst agents	Steward
Accesses your database	✅ Read-only queries, scoped to your semantic layer	✅ Read-only, including table structure and ad hoc exploratory queries
Sends data to AI provider	Query results (specific rows returned by a compiled query)	Schema metadata, and row-level data when exploratory queries are run
Stores your data	❌ Not in the product database¹	❌ Not in the product database¹
Used for AI training	❌ Never	❌ Never

¹ Query results and tool activity are captured in short-lived, PII-scrubbed engineering diagnostic traces (see "Traces for debugging" above) — they are not stored in AI Analyst's own product database.

For information about our broader security posture, certifications, and contractual commitments, see How Actian AI Analyst protects your data.

For a full record of actions taken in your workspace, see Audit Logs.