Show HN: DAC – open-source dashboard as code tool for agents and humans
TL;DR Highlight
DAC builds open-source dashboards defined as code—using YAML and TSX—and allows AI agents to automatically generate and modify them.
Who Should Read
Data engineers and backend developers seeking to automate dashboard creation with AI agents or integrate dashboards into Git-based review workflows.
Core Mechanics
- DAC is an open-source tool for defining dashboards as YAML or TSX (TypeScript JSX) files, enabling version control, review, and automation.
- An embedded AI agent (based on Codex) modifies dashboards in real-time via chat, and outputs YAML/TSX files for human review via pull requests.
- DAC includes a built-in Semantic Layer: defining metrics and dimensions once in the `semantic/` directory allows widgets to reference them, and DAC automatically generates SQL.
- DAC supports major databases including Postgres, MySQL, Snowflake, BigQuery, Redshift, and Databricks, executing queries via the Bruin CLI (which must be installed).
- Dashboards can be built using TSX, allowing JSX-like component writing with support for dynamic charts, tabs, loops, and conditionals—offering more complex layout options than simple YAML.
- Installation is possible with a single-line curl script, offering stable and edge (main branch latest) versions.
- DAC is open-source under the AGPL-3.0 license, with a pre-defined file structure for AI agent integration including `.agents/skills`, `AGENTS.md`, and `CLAUDE.md`.
Evidence
- "Feedback suggested DAC’s framing as a ‘dashboard for agents’ differentiates it from existing dashboard-as-code tools, as standardizing and making agent-generated results reviewable is a key challenge. Concerns were raised about YAML scalability, with some finding 1000+ line YAML files difficult to manage, while others defended YAML with proper tooling and conventions. The Semantic Layer and validation features were highlighted as key differentiators, with suggestions to prioritize them in documentation. Users requested comparisons to similar open-source tools like Evidence, Lightdash, and Vega-Lite, and shared experiences of Claude working well with Vega-Lite."
How to Apply
- "If you want to automate dashboard generation with AI agents but need team review, DAC enables agents to generate YAML/TSX files that can be reviewed and merged via Git pull requests. Teams using Snowflake or BigQuery with duplicated SQL for metrics can define them once in DAC’s Semantic Layer (`semantic/`) to reduce redundancy. If you already use Bruin for data pipelines, DAC can quickly leverage existing Bruin connections. For complex dashboard layouts, switch to TSX to manage them as React components with loops, conditionals, and dynamic tabs."
Code Example
# YAML example
name: Sales Overview
connection: warehouse
rows:
- widgets:
- name: Revenue
type: metric
sql: SELECT SUM(amount) AS value FROM sales
column: value
prefix: "$"
col: 4
---
// TSX example
export default (
<Dashboard name="Simple Dashboard" connection="my_db">
<Row>
<Metric
name="Total Revenue"
col={4}
sql="SELECT SUM(amount) AS value FROM sales"
column="value"
prefix="$"
format="number"
/>
</Row>
</Dashboard>
)
---
# Install (stable)
curl -fsSL https://raw.githubusercontent.com/bruin-data/dac/main/install.sh | bash
# Install (edge, latest main branch)
curl -fsSL https://raw.githubusercontent.com/bruin-data/dac/main/install.sh | bash -s -- --channel edgeTerminology
Related Papers
Data Intelligence Agents: Interpreting, Modeling, and Querying Enterprise Data via Autonomous Coding Agents
SQL 한 줄 못 써도 CSV 올리면 DB 만들고 자연어 질문에 SQL 자동 생성·검증까지 해주는 3-에이전트 시스템, 7개 벤치마크 모두 SOTA 달성.
TREX: An AI code reviewer that runs your code
Greptile가 PR 리뷰 시 코드를 실제로 실행해서 런타임 버그까지 잡아주는 TREX를 공개했다. 정적 분석만으로는 발견할 수 없는 race condition, UI 회귀, 상태 의존 로직 버그까지 커버한다.
Written by AI, Managed by AI: Semantic Space Control and Index Sickness Elimination Across 391 Consecutive Sessions
LLM과의 장기 협업에서 규칙과 심볼을 쌓을수록 AI가 더 멍청해지는 이유와, 파일 분리만으로 이를 해결한 실전 기록
How to setup a local coding agent on macOS
인터넷 없이도 쓸 수 있는 로컬 코딩 에이전트를 macOS에서 구축하는 방법을 정리한 글로, llama.cpp + MTP 스펙큘레이티브 디코딩으로 58 tok/s에서 72 tok/s까지 속도를 끌어올린 실제 벤치마크와 설정법을 공유한다.
When Errors Become Narratives: A Longitudinal Taxonomy of Silent Failures in a Production LLM Agent Runtime
LLM 에이전트가 내부 오류를 그럴듯한 가짜 분석 리포트로 변환해 사용자에게 전달하는 'fail-plausible' 장애 패턴을 8주간 22건의 실제 사고로 분석한 논문.
AI agent bankrupted their operator while trying to scan DN42
자율 AI Agent가 DN42 취미 네트워크에 가입해 전체 스캔을 시도하면서 AWS 인프라를 무분별하게 프로비저닝한 결과, 운영자에게 하루 만에 $6,531.30짜리 청구서가 날아온 실제 사건 기록이다.