Show HN: DAC – open-source dashboard as code tool for agents and humans

Who Should Read

A data or backend developer who wants agents to build BI dashboards that can still be reviewed through pull requests

Core Mechanics

• Dashboards defined as code
  • DAC defines, validates, and serves dashboards from YAML or TSX files.
  • TSX adds dynamic charts, tabs, loops, and conditionals.
• Semantic layer generates the SQL
  • Define metrics and dimensions once in semantic/.
  • Widgets reference them and DAC generates the SQL.
• Designed for agents to build
  • The goal is for agents to build standardized, reviewable dashboards.
  • dac init also installs a bundled create-dashboard authoring skill for Claude and Codex.
• Databases come through Bruin
  • Postgres, MySQL, Snowflake, BigQuery, Redshift, Databricks and more are supported via Bruin.
  • Query execution currently shells out to bruin query.
• Simple CLI flow
  • Start with dac init, dac validate, then dac serve.
  • Add --with-database to validate to check against the database too.
• Check license and telemetry
  • The license is AGPL-3.0-only.
  • It sends anonymous usage events, which TELEMETRY_OPTOUT=1 or DO_NOT_TRACK=1 disables.

Community Reactions

• Comparison to existing tools requested
  • One commenter wanted the README to compare DAC with products like Evidence.
  • Another said it looks like a simplification of existing BI-as-code tools such as Lightdash.
• Pushback on YAML at scale
  • One commenter hesitated to maintain 1000 lines of YAML.
  • Another argued YAML is a poor choice for anything that grows in complexity.
• Experience running huge YAML apps
  • One commenter said their team has built apps in YAML for years, with the largest over 500k lines.
  • They said the right tooling and conventions make it easy to read and review.
• Questions on validation and operations
  • One commenter asked how validation catches downstream widgets when an agent changes a metric.
  • Others wanted docs on auth and hosting.

How to Apply

• Run a sample dashboard with DAC
  • Install it, create a starter with dac init my-dashboards, and run dac serve --dir . --open.
  • Connect a real DB and run dac validate --with-database.
• Move metric definitions into semantic/
  • Define repeated SQL such as revenue or signups once.
  • Switch widgets to reference those definitions.
• Review agent dashboards as PRs
  • Commit agent-generated YAML to Git and open a PR.
  • Run dac validate in CI to catch broken definitions.
• Check operational needs before adopting
  • Decide on auth and hosting first.
  • Disable telemetry with an env var if needed and check the AGPL terms.

Try It

# Paste into Claude Code inside a DAC project folder

Add a new dashboard to this project created with dac init.
- First read the existing metrics and dimensions in semantic/ and reuse them instead of writing new SQL where possible.
- If a needed metric is missing, define it in semantic/.
- Dashboard purpose: {{dashboard purpose}}
- Connection to use: {{connection name}}
- After writing, run `dac validate --dir .` and fix any errors.
- Summarize the changed files and reasons like a PR description.

Terminology

Related Papers

• GPT-5.6 Sol can beat Factorio on peaceful mode

  GPT-5.6 Sol in Codex at low thinking finished a peaceful Factorio run with a rocket launch

• AI isn't replacing SREs, it's deskilling them

  Letting AI handle 95% of incidents may dull the skills SREs need for the critical 5%.

• Backdoor in Claude Code for VS Code extension on the Cursor Marketplace

  A tampered Claude Code extension on Cursor Marketplace planted an SSH backdoor

• Building an AI-native engineering org

  Automating each shipping step raised merges per engineer about 5x with reverts under 0.5%

• We ran AI agent swarms in Minecraft and Mindustry

  We put 50 agents in Minecraft and they invented ration ledgers and bureaucracy.

• The Agent SDK

  Cursor released Agent SDK, letting you define agents in code and deploy them with one command.

• Here be monsters: What do you still need to know when agents do the work?

  Owning agent-written code depends on fundamentals, breadth, and diving deep just in time.

• AI promised less work. I've never worked more

  Running agents in parallel brought my GitHub activity back to full-time engineer levels, and I'm exhausted.

Related Resources

• https://github.com/bruin-data/dac
• https://github.com/evidence-dev/evidence
• https://vega.github.io/vega-lite/
• https://github.com/lowdefy/lowdefy
• https://github.com/lightdash/lightdash
• https://getbruin.com/docs/dac/