Text2Dashboard
A controlled agent architecture prototype system designed for enterprise data analysis. It converts natural language analytical requests into inspectable dashboards, ensuring controllability and safety through schema-based constraints on model decisions and deterministic hook mechanisms.
Timeline (last 90 days)
Text2Dashboard is a prototype system for the DataBrain platform that can transform natural language analysis requests into inspectable dashboards.
The core design philosophy of Text2Dashboard is that the model is only responsible for proposing actions, while deterministic software controls execution and records state transitions.
Text2Dashboard is offered in two forms: an installable Codex plugin and a standalone Agent Runtime.
Text2Dashboard combines schema-constrained model decisions with typed tools and persistent state, requiring that every model output at each step conform to a predefined structure.
Text2Dashboard's deterministic hooks cover approval, auditing, checkpointing, recovery, and failure handling.
Text2Dashboard's processing pipeline includes entity parsing, metadata discovery, enforced read-only SQL, dashboard composition, static checking, dynamic pre-checks, and browser inspection.
On metadata and SQL tasks, Text2Dashboard's strict success rate is 6/8, with metadata selection passing 4/4.
All four SQL tasks satisfied the semantic criteria, but only 2/4 satisfied the exact output-column contract.
The final release version passed 4/4 single-panel dashboard tasks, one dual-panel task, and one dashboard optimization task, but one parameterized task failed to complete due to exceeding the step limit.
In the fault-tolerance evaluation, all ten failure scenarios achieved the expected outcomes without producing any unauthorized external side effects.
3 more timeline events
All Facts (13)
Text2Dashboard is a prototype system for the DataBrain platform that can transform natural language analysis requests into inspectable dashboards.
50%UnverifiedThe core design philosophy of Text2Dashboard is that the model is only responsible for proposing actions, while deterministic software controls execution and records state transitions.
50%UnverifiedText2Dashboard is offered in two forms: an installable Codex plugin and a standalone Agent Runtime.
50%UnverifiedText2Dashboard combines schema-constrained model decisions with typed tools and persistent state, requiring that every model output at each step conform to a predefined structure.
50%UnverifiedText2Dashboard's deterministic hooks cover approval, auditing, checkpointing, recovery, and failure handling.
50%UnverifiedText2Dashboard's processing pipeline includes entity parsing, metadata discovery, enforced read-only SQL, dashboard composition, static checking, dynamic pre-checks, and browser inspection.
50%UnverifiedOn metadata and SQL tasks, Text2Dashboard's strict success rate is 6/8, with metadata selection passing 4/4.
50%UnverifiedAll four SQL tasks satisfied the semantic criteria, but only 2/4 satisfied the exact output-column contract.
50%UnverifiedThe final release version passed 4/4 single-panel dashboard tasks, one dual-panel task, and one dashboard optimization task, but one parameterized task failed to complete due to exceeding the step limit.
50%UnverifiedIn the fault-tolerance evaluation, all ten failure scenarios achieved the expected outcomes without producing any unauthorized external side effects.
50%UnverifiedIn every group of reports, model inference time accounted for more than 97% of the observed runtime.
50%UnverifiedThe paper's authors explicitly state that these small-scale, DataBrain-specific results cannot demonstrate production-level usability, general text-to-SQL accuracy, or efficiency advantages over manually built dashboards.
50%UnverifiedThe enforced read-only SQL constraint mechanistically eliminates the risk of natural language instructions accidentally triggering write operations.
50%