Building a fast path for Lightdash Agents with Jev

Lightdash developed Fast mode using Jev, TypeSafe’s decision model, to handle common chart-edit follow-ups without requiring a full LLM agent run, allowing requests such as changing a chart to a bar chart to be interpreted as fixed actions and applied by existing code. In tests across 25 scripted follow-ups, Fast mode reduced median response time from 17.5 to 1.7 seconds, avoided calling the agent LLM for 72% of requests, and substantially reduced token use while producing the same final charts. The system relies on constrained choices, confidence thresholds, and a completeness check to ensure Jev’s planned edits match the request; uncertain or incomplete requests are routed to the normal agent. Testing also revealed issues involving missed existing filter values and response-handling bugs, leading to lessons about testing real user paths, accounting for model variability, validating experiment design, and monitoring feature usage as well as failures. Lightdash customers can request access to Fast mode and provide feedback to help improve it.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.