DeepSQL 1.1 makes AI dashboard generation survive the browser session

DeepSQL has released version 1.1.0, two days after its first public open-source release, with an architectural change aimed at a familiar failure mode in agentic analytics: a browser disconnect no longer has to terminate dashboard generation. The release makes dashboard turns server-owned and persistent, while safely rejecting concurrent turns. Its release notes say the client receives a dashboard shell first and then individual widgets over server-sent events, allowing the preview to mount before the entire artifact is complete. DeepSQL’s changelog and v1.1.0 release notes document the change.

That combination separates two concerns that are often conflated in AI interfaces: the transport used to show progress and the ownership of the work itself. SSE still provides incremental updates, but the generation job now lives beyond a particular SSE connection. DeepSQL records a generation_status and uses optimistic locking on saved dashboards; the same release adds dashboard version history and restore support. Together, those mechanics make a long-running generation turn less dependent on a healthy browser tab and give users a path back after an unwanted edit. The release changelog lists the persistence and versioning changes.

The dashboard surface also gains cloning, folders, favorites, search, refresh and auto-refresh, a public TV kiosk mode, and AI-evaluated alerts. DeepSQL describes generated dashboards as self-contained HTML rendered in a sandboxed iframe without network access; data is supplied through a read-only query bridge to the backend. That design gives generated presentation code room to vary without handing it direct database or internet access. The project README explains the dashboard sandbox and query bridge, while the release notes enumerate the new dashboard controls.

Beyond dashboards, 1.1.0 makes non-public PostgreSQL schemas first-class across the Editor, Brain and Advisor interfaces, and combines Slow Queries with Workload Analysis in a single Performance area. The MCP package remains at 0.27.0 because this cut does not require MCP API changes. DeepSQL’s changelog records both changes.

The upgrade path carries an important operational caveat. DeepSQL does not run Flyway migrations at runtime; operators not relying on Hibernate ddl-auto=update must manually apply the V111–V114 SQL changes for generation state, dashboard versions and alerts. The release ships a backend JAR, frontend archive, source bundles, CycloneDX SBOMs and checksums for air-gapped deployment, but the project’s primary install path still builds the stack locally with Docker Compose. The v1.1.0 release notes provide the migration and artifact details, and the README documents the Compose-based installation.