What Is PromptQL?

PromptQL is Hasura's AI data platform, the company's successor product to its GraphQL engine. Rather than one-shot text-to-SQL, PromptQL has the LLM write a deterministic, multi-step query plan (a “PromptQL program”) that runs against a semantic layer Hasura calls the supergraph, built from Hasura connectors to sources like PostgreSQL, SQL Server, MySQL, MongoDB, Snowflake, and ClickHouse. To give the LLM business context, PromptQL maintains an internal wiki of domain knowledge and retrieves relevant pieces at query time. Hasura has also marketed PromptQL to credit unions.

Where PromptQL Falls Short

• Context depends on a wiki and retrieval. PromptQL focuses the LLM by retrieving wiki entries and metadata. Retrieval is exactly where focus breaks down on large, messy schemas: if the right join, metric, or alias is not retrieved, the plan is wrong, however deterministic its execution.
• Setup is a modeling project. The supergraph must be configured through Hasura's connectors and metadata, and the wiki must be written and kept current. That is engineering work before business users or agents get value.
• Consumption pricing that compounds. Annual contracts are prepaid credits drawn down by usage — about 4.2¢ per PromptQL program, plus LLM input and output tokens at provider rates when using Hasura's LLM. Multi-step plans mean more tokens per question; cost rises with adoption.
• Developer-first, cloud-first. PromptQL grew out of a developer API platform and is delivered primarily as a cloud service. Regulated institutions that need a packaged, on-prem product with governance out of the box have more to build themselves.
• Limited functionality for the price. Investor and customer conversations consistently put PromptQL among newer entrants that combine high cost with a narrower feature set than an enterprise search deployment needs.

How Tursio Is Different

• Self-building context graph. Tursio infers joins, dimensions, measures, aliases, and PII automatically from schema, query history, and BI definitions (Power BI, Tableau) — no hand-written wiki — and continuously tunes it from feedback.
• Operator trees grounded to real values. Tursio's planner decomposes each question into a logical operator tree and compiles it to the native SQL dialect. “Small models” (hierarchical statistics and hash indexes) constrain every operand to values that exist in the database, so entities cannot be hallucinated.
• Packaged and on-prem. One Docker image with automated training, RBAC, audit logs, Entra ID/OAuth, and BYO LLM keys. Nothing leaves your perimeter.
• Lower, flat cost. A flat license with about 3¢ per query in tokens, billed to your own LLM subscription. In MCP evaluations, Tursio used up to 5x fewer tokens than querying databases directly.

Query Planning: Overlap and Differences

PromptQL and Tursio share a core insight: separate planning from the LLM and execute a deterministic plan, rather than trusting one-shot SQL. That makes PromptQL Tursio's closest methodological cousin.

The difference is what grounds the plan. PromptQL's LLM writes the program using whatever context retrieval surfaces from a wiki. Tursio's plan is built from a structured context graph and constrained to real database values at every step — so the same question returns the same correct answer for every user and every agent, without depending on what retrieval happens to find.

Side-by-Side Comparison

Bottom Line

PromptQL gets the architecture direction right, but leaves enterprises to model and maintain the context it depends on, and charges per program on top of tokens. Tursio automates that context, grounds every plan in real data, and ships as a governed on-prem product on a flat license.