# What is Holistics AI > AI analytics is only as trustworthy as the foundation it reasons from. Holistics AI reasons from your governed semantic layer using AQL, not from raw schema, which is why its answers stay reliable. :::info Don't see AI in your workspace? [Request access](https://www.jotform.com/230511857392457?betaFeature=Holistics%20AI) to get it enabled. ::: ## Why Holistics AI is structurally different Most AI analytics tools translate a natural-language question directly into SQL against the raw warehouse schema. The AI has to guess at table joins, canonical date fields, and what business metrics like "revenue" or "active customer" actually mean. The output looks plausible. Often it's silently wrong. Analysts end up verifying every answer. Holistics AI is built differently. Instead of generating SQL against raw tables, it generates [**AQL**](/as-code/aql/) against your **governed semantic layer**. AQL is a composable query language where metrics are first-class objects: your team's existing definitions of revenue, active customer, retention, and so on. The AI doesn't reinvent business logic; it reuses what you've already defined. The chain is: 1. **You define business logic once** in AML and AQL: composable, version-controlled, reviewable. 2. **AI generates AQL** (not SQL) from a natural-language question, reusing your governed metric definitions. 3. **Holistics compiles AQL to SQL** deterministically, runs it against your warehouse, and returns the result. This is why Holistics AI is reliable in places other AI tools aren't: period comparisons, cohort questions, ratios across grains, follow-ups that build on the previous answer. For the full mechanism, including why this approach is more verifiable, governed, and capable than direct text-to-SQL, see [Why Holistics AI is reliable](/docs/ai/architecture). ## What it does Holistics AI helps you work with data faster using natural language. You can ask questions, explore and visualize data, understand charts and dashboards, and build analytics assets, all through conversation. It works in both **Reporting** and **Development** environments. See the full list of capabilities in [What you can do with Holistics AI](/docs/ai/capabilities). ## How AI stays reliable Every AI answer is grounded in your semantic layer, not raw database tables. Every query goes through the same permission controls as normal Holistics usage. Generated AQL is compact and human-readable, which makes it easier to verify than the wrangled SQL that direct text-to-SQL systems produce. Read the full architecture in [Why Holistics AI is reliable](/docs/ai/architecture). ## Context and customization The quality of AI answers depends on the context the AI has access to. You can improve results by: - Enriching your [semantic and reporting layers](/docs/ai/context/semantic-and-reporting-layers) with descriptions, synonyms, and labels - Defining [custom context](/docs/ai/context/custom-context) for your organization's terminology and conventions - Adding [conversation context](/docs/ai/context/conversation-context) on the fly See [What AI uses for context](/docs/ai/context/overview) for the full picture. ## Data security Your data is protected by design. Holistics AI uses OpenAI with Zero Data Retention enabled: no data sharing and no API call logging. All queries go through your existing permission controls. You can also [bring your own LLM](/docs/ai/bring-your-own-llm) for full control. See [Data access & policy](/docs/ai/data-access-and-policy) for details. ## Setup Admins can enable AI, control user access, configure data sharing, and connect custom LLM providers from [AI Settings](/docs/ai/ai-settings). ## Extensibility Connect Holistics AI to your own tools using the [MCP Server](/docs/ai/connect-external-tools/mcp-server), or use [AI functions directly in your data warehouse](/docs/ai/run-ai-functions). ## Roadmap See what's coming next in our [roadmap](/roadmap).