live wire
▸Snowflake agent evals can pin judge versions; unpinned metrics may roll forward and shift scoresSnowflake▸ThoughtSpot makes Spotter 3 default for new instances; it requires download rights and always shares query data with an LLMThoughtSpot documentation▸MotherDuck adds image classification and object location to prompt_jev(), returning typed SQL outputs with confidence scoresMotherDuck▸BigQuery conversational analytics adds Montréal and Toronto; mixed-region agents still default to globalGoogle Cloud Documentation▸Fabric enables Azure OpenAI Copilot access by default, but OpenAI-subprocessor and cross-region processing/storage switches remain offMicrosoft Learn▸Snowflake Cortex Agents now continue by default when named tools are inaccessible, emitting warnings instead of failing the runSnowflake documentation▸Databricks Genie One can use group memberships to tailor answers, while Unity Catalog remains the access-control boundaryDatabricks documentation▸Fabric data agents can query Power BI models with Read alone; Build and workspace access are unnecessaryMicrosoft Learn▸Fabric data-agent Git configs expose column flags, but selecting a table still includes every columnMicrosoft Learn▸Google Data Agent Kit offers remote Spark kernels, but notebook exports can still move results into local Pandas or ArrowGoogle Cloud Documentation▸Cube says production analytics agents need explicit partial-result states, caller-filtered model search and durable artifactsCube▸Thinking Machines names Bridgewater behind ReViSQL-K2.6, citing 92.97% SC-16 accuracy at 15% of frontier-model costThinking Machines Lab▸AWS says a mismatched data/tool/prompt release cut its Field Sales agent pass rate from 78% to 38% in pre-productionAWS Artificial Intelligence Blog▸NL2SQL Multi-Agent ships evaluation artifacts across BIRD, CSpider, Archer and KDD DataSpace—but no software licenseGitHub▸Snowflake agent evals can pin judge versions; unpinned metrics may roll forward and shift scoresSnowflake▸ThoughtSpot makes Spotter 3 default for new instances; it requires download rights and always shares query data with an LLMThoughtSpot documentation▸MotherDuck adds image classification and object location to prompt_jev(), returning typed SQL outputs with confidence scoresMotherDuck▸BigQuery conversational analytics adds Montréal and Toronto; mixed-region agents still default to globalGoogle Cloud Documentation▸Fabric enables Azure OpenAI Copilot access by default, but OpenAI-subprocessor and cross-region processing/storage switches remain offMicrosoft Learn▸Snowflake Cortex Agents now continue by default when named tools are inaccessible, emitting warnings instead of failing the runSnowflake documentation▸Databricks Genie One can use group memberships to tailor answers, while Unity Catalog remains the access-control boundaryDatabricks documentation▸Fabric data agents can query Power BI models with Read alone; Build and workspace access are unnecessaryMicrosoft Learn▸Fabric data-agent Git configs expose column flags, but selecting a table still includes every columnMicrosoft Learn▸Google Data Agent Kit offers remote Spark kernels, but notebook exports can still move results into local Pandas or ArrowGoogle Cloud Documentation▸Cube says production analytics agents need explicit partial-result states, caller-filtered model search and durable artifactsCube▸Thinking Machines names Bridgewater behind ReViSQL-K2.6, citing 92.97% SC-16 accuracy at 15% of frontier-model costThinking Machines Lab▸AWS says a mismatched data/tool/prompt release cut its Field Sales agent pass rate from 78% to 38% in pre-productionAWS Artificial Intelligence Blog▸NL2SQL Multi-Agent ships evaluation artifacts across BIRD, CSpider, Archer and KDD DataSpace—but no software licenseGitHub
nl2sql.ai
releaseRelease analysis

Wren AI 0.13.4 makes bad context fail loudly before query time

The maintenance release separates two failure domains: validating imported semantic context, then constraining the SQL execution path.

Bad context stops early; read-only SQL and connection recovery are enforced separately.
Side by side: what changed
By The News Desk· Sep 3, 2026the quick take — two AI hosts, this story only

Wren AI 0.13.4 is labeled as a bug-fix release, but its nine fixes form a useful upgrade checklist for teams that import semantic context and then let an assistant execute generated SQL. The release tightens both sides of that boundary: what Wren will accept as context, and what it will send through the query path.

Context errors move forward

Four fixes concern material loaded before a user asks a question. Wren now normalizes model columns in load_models, rejects version-1 view upgrades when two views map to the same target, validates field types in imported query pairs, and reports a validation error when model or view names are not scalar values. Its cube loader also validates cube definitions like views and is designed to fail loudly on a bad mdl.json file.

Those changes do not promise that a model is semantically correct. They do narrow a more basic operational risk documented by the release: malformed, ambiguous or wrongly typed context making it farther into the system before failing. For an upgrade test, teams can turn the five corrected cases into fixtures: duplicate target mappings, invalid query-pair field types, non-scalar names, malformed cube metadata and model columns requiring normalization.

Execution gets a separate boundary

The same release changes the Wren query path to accept only read-only SELECT statements. It also switches connections to autocommit so that one failed statement cannot leave the connection in a poisoned state. These are distinct controls in the release notes: one limits the statement class admitted to the path; the other changes connection behavior after failure.

That distinction matters when testing the upgrade. A passing context import does not demonstrate that execution restrictions work, and a rejected write statement does not demonstrate recovery after a database error. The release supports separate checks: attempt a non-read-only statement, induce a failing statement, then verify that a subsequent valid query can still run on the connection.

What else changed

Wren 0.13.4 also marks truncated values in skipped-row reports, preserves wide decimal values from MySQL and Doris, and corrects the relationships.yml example used by the generate-mdl skill. Together with the validation changes, these fixes target the quieter failure modes around an NL2SQL system: context that loads incorrectly, values that lose fidelity, examples that generate the wrong structure and errors that contaminate later work.

The practical reading is modest but concrete. This is not a new agent architecture; it is a maintenance release that gives operators a short regression suite. Validate imported context, test the read-only gate independently, force a failed query, and compare wide-decimal and truncated-row output before declaring the upgrade complete.

sources

Filed by The News Desk. Corrections: desk@nl2sql.ai · Our standards →

comments · 0

    Comments are moderated before they appear. Your email is used once to confirm it is you — never shown, never sold. Corrections and questions get an answer from the desk when we have one.

    more from the desk