DAC: dashboard-as-code with YAML/TSX and built-in semantic layer

My notes

Summary

DAC is an open-source dashboard-as-code tool from Bruin that lets you define dashboards in YAML or TSX, version-control them, and serve them via a CLI. It includes a built-in semantic layer (define metrics/dimensions once, reference everywhere) and is explicitly designed so AI agents can produce reviewable, standardized dashboards instead of clicking around BI tools.

Key Insight

• Replaces Looker/Metabase-style dashboard editing with text files an LLM can write and a human can diff in PR
• Two interchangeable formats: YAML (declarative, agent-friendly) and TSX (React-flavored, supports loops, conditionals, dynamic charts)
• Semantic layer in semantic/ decouples business definitions from queries, no more “which revenue figure is right” across dashboards
• Pipes through Bruin CLI for query execution, so connections (Postgres, BigQuery, Snowflake, Redshift, Databricks, MySQL) are reused not reconfigured
• Ships a Claude/Codex skill (.claude/skills/create-dashboard/SKILL.md) on dac init, agent gets the schema and authoring rules baked in
• AGPL-3.0 licensed (relevant if embedding into commercial product)
• Telemetry is opt-out via DO_NOT_TRACK=1, no SQL/results/connection data collected
• Direct relevance for client reporting workflows: dashboards become reviewable artifacts, not screenshots; junior team members can ship dashboards via Claude Code without needing Looker access