What Is ClickHouse AI SQL Generation?

ClickHouse ships a built-in text-to-SQL helper in its command-line client (v25.7+). It turns a plain-text question into a one-shot SQL query against the ClickHouse instance you're connected to. It's a useful developer convenience for exploring data — but it is a feature, not a governed enterprise search layer. The contrast below is where Tursio's bet on context as a first-class data structure shows up.

Where ClickHouse AI SQL Falls Short

• No persistent context. Every question re-discovers the schema from scratch — listing databases, tables, and reading CREATE TABLE statements. There is no memory between queries, no learned business meaning, no reusable semantic model.
• Non-deterministic. One-shot generation with no conversation history means the same question can yield different SQL and different results on different runs. For enterprise reporting, this is a dealbreaker.
• ClickHouse only. Queries run against a single ClickHouse instance. If your data also lives in SQL Server, Oracle, PostgreSQL, or flat files, this feature cannot reach it.
• Schema sent to external APIs. The feature sends your schema to an external LLM (Anthropic or OpenAI) using your API key. For organizations with sensitive data, this means schema metadata leaves your environment.
• No governance. No policy enforcement, no access control, no PII handling. It's a raw text-to-SQL pipe with no guardrails.
• Developer-only interface. Available only at the CLI (clickhouse client / clickhouse-local). Business users — the people who actually need to ask data questions — cannot use it.

How Tursio Is Different

• Persistent context graph. Tursio builds the context once — join paths, column semantics, dimensions/measures, ontologies, PII tags — and reuses it on every query. No re-deriving schema each time. This is context pushdown: the expensive semantic work happens once, not on every question.
• Deterministic answers. The same question returns the same answer for every user and every agent. The context graph constrains generation so results are reproducible.
• Platform-agnostic. Tursio spans SQL Server, Oracle, PostgreSQL, Snowflake, Databricks, MySQL, ClickHouse, and flat files. One search layer across your entire data estate.
• On-premises, private. Runs inside your environment as a Docker image. Data never leaves your perimeter. Bring your own LLM keys.
• Governed search. Per-user data policies, RBAC, PII enforcement, integration with tools like Microsoft Purview.
• For everyone, not just developers. Business users search through Teams, Excel, Copilot, and the Tursio UI. Agents query via MCP and search API.

Side-by-Side Comparison

Bottom Line

ClickHouse's feature answers "can a developer get SQL from a question at the CLI?" Tursio answers a fundamentally different question: "can every business user and agent get the same, trustworthy answer across all our data — governed, on-prem, and without re-deriving context every time?" Re-discovering the schema on each query is exactly the tedious, costly, non-deterministic work Tursio removes by capturing context once and pushing it down.