Build on your dbt project and query history. Let AI surface ambiguous business definitions, resolve them with your analyst, and turn each decision into shared context and evaluation cases. Free, Apache-2.0, and on your schedule.
Open source · Apache-2.0 · read the docs ↗
A quick introduction to the open-source repo, from the user's perspective.
AI-generated explainer · 1:47
Your analytics context builds on the sources you already maintain. Keep model logic, lineage, and column metadata in dbt. Give your agent access to that project and add the dashboard-level meaning it needs to answer real business questions.
Read the dbt project and authorized query history. Find recurring questions, competing definitions, and missing assumptions. Existing SQL is evidence to review, not automatically the right answer.
Does “active customer” mean a paying account or a product user? Which dashboard filters and time window apply? AI presents the ambiguity; your analyst confirms the interpretation. Reference the technical definitions where they already live.
Each confirmed decision becomes a business-context entry and a test case. Keep both in a reviewable repository your team owns, and share the maintained context with supported agents.
Compare answers with context off and on. Check live results against an analyst-approved reference when warehouse access is available. See what improved, what stayed the same, and what still needs work.
The local evaluation runner checks generated SQL requirements and entity choices with a model judge. It does not execute warehouse queries or grade numerical snapshots. Live answer verification is a separate guided workflow.
Read about evaluation ↗Use Claude Code, Codex, or Cursor with approved read-only data access and a domain expert. Your team maintains the references, runs comparisons, and reviews failures. The tools are Apache-2.0; model and warehouse usage are yours.
Get the repo ↗A scoped evaluation sprint with Nodal. Build context and tests, establish a baseline, and leave with a prioritized plan.
See the white-glove sprint →