dashdown-md 0.1.3

Markdown-driven analytics pages, served by Python.

Turn Markdown into interactive analytics dashboards.

_{▶ Watch the full 60-second tour — Markdown + SQL → a live, filterable dashboard}

Write a .md file with some SQL and a few component tags. Dashdown serves it as a live, interactive dashboard — no build step, no JavaScript framework, no npm.

:::query name=monthly_sales connector=main
SELECT month, SUM(amount) AS revenue
FROM sales GROUP BY month ORDER BY month
:::

# Sales

<Counter data={monthly_sales} column="revenue" label="Revenue" format="currency" />
<LineChart data={monthly_sales} x="month" y="revenue" title="Monthly Revenue" />

That's a complete dashboard page — a KPI card and a chart, backed by a real query. Point the CLI at the folder and it's live in your browser.

---

Quickstart

pip install dashdown-md          # Python 3.10+
# or, with uv:
uv tool install dashdown-md      # install the CLI globally
# or run it without installing:  uvx --from dashdown-md dashdown new my-dashboard

dashdown new my-dashboard     # scaffold a project
dashdown serve my-dashboard   # → http://localhost:8000

Drop a CSV in data/, write .md files in pages/, and edit live — the dev server hot-reloads as you save.

---

Why Dashdown

• It's just Markdown. Author dashboards in plain .md — prose, headings, callouts, and Mermaid diagrams sit right next to your charts. Version them in git, review them in a PR — or hand the folder to a coding agent and let it write them (see Built for coding agents).
• SQL is the API. Embed queries inline or share them from a queries/ library. ${param} placeholders wire filters to SQL with context-aware, injection-safe substitution — and the SQL never ships to the browser.
• No frontend to build. No webpack, no Node, no framework to learn. The frontend is hand-written ES modules served as static files; you write Python and Markdown.
• Fast by design. Pages render instantly with no data, then the browser fetches each query asynchronously. Filters update the URL and re-fetch only what changed.
• Runs offline. ECharts, Alpine, Tailwind/DaisyUI, fonts — all self-hosted. Zero external requests, works air-gapped.

---

What you can build

• Charts — line, bar, pie/donut, scatter, treemap, funnel, radar, gauge, heatmap, sankey, candlestick, box plot, violin, map, calendar heatmap, graph, sunburst, tree, parallel-coordinates, and more — all via one <XChart> tag.
• KPIs & tables — <Counter> cards with trend sparklines and change badges, sortable/paginated <Table> with CSV export, inline <Value>, drag-and-drop <PivotTable>.
• Filters — <Dropdown> (single & multi), <Search>, <DateRange>, <Toggle>, plus a project-wide global date filter. They sync to the URL, so every dashboard view is shareable.
• Live data — mark a query live and its components stream fresh results over a WebSocket. Works against any connector (it polls).
• LLM commentary — <Ask data={q} ask="What stands out?" /> renders a model's answer next to the chart (Mistral, Claude, OpenAI, or OpenRouter).
• A semantic metric layer — define measures once in YAML, then <BarChart metric={sales.revenue} by={sales.region} />.

---

Built for coding agents

A Dashdown dashboard is just Markdown + SQL + component tags — the ideal medium for a coding agent to write. A full chart is one tag; a query is a few lines of SQL. An agent can describe a whole dashboard in a handful of lines, so it builds far more per token than hand-coding a UI — and you review a small, plain-text diff in a PR instead of thousands of lines of JavaScript.

Every project is scaffolded to be agent-ready out of the box:

• AGENTS.md — a tool-agnostic guide that any agent (Claude Code, Cursor, Codex, …) reads on open, so it knows the platform without you explaining it. It's a map: a cheat-sheet plus a table of contents into per-topic references/*.md shards, loaded only when a task needs them (no 50k-token manual for every edit).

• A Claude Code skill — .claude/skills/dashdown-authoring/ routes a task to the right reference and the command that verifies it.

• Facts from the CLI, not from memory — the agent checks instead of guessing, so it doesn't invent attributes or config keys:

  dashdown components            # introspected attribute catalog for every component
  dashdown check                 # config loads + every page renders?
  dashdown query "SELECT …"      # inspect the real data / schema
  dashdown screenshot /page      # did the chart canvases actually draw? (non-zero if not)
  

Already have a project? dashdown skill drops the guide in (--refresh to update it to the current release). A static build also publishes llms.txt / llms-full.txt for agent hosts that fetch docs over the network.

→ Coding agents for the full read-edit-verify loop.

---

Connect to your data

CSV · JSON · Parquet · DuckDB · MotherDuck · PostgreSQL · MySQL/MariaDB · SQL Server / Azure SQL · Snowflake · BigQuery · Excel · Google Sheets · Microsoft Fabric / Power BI (DAX) · Cube.

# sources.yaml
main:
  type: csv
  directory: data

Backend drivers install as extras (pip install 'dashdown-md[postgres]'), so the core stays lean. Need something else? Write a connector in a few lines of Python, or ship it as a pip-installable plugin.

---

Share it anywhere

• Static export — dashdown build pre-renders the whole site to plain HTML + JSON snapshots. Host it on Netlify, Vercel, GitHub Pages, S3 — no server, no Python.
• PDF — dashdown pdf renders a presentation-quality deck (cover page, one widget per row, clean page breaks) via headless Chromium.
• Embed — drop any page into Notion, a wiki, or a CMS as an auto-resizing, chrome-less iframe, with signed page-scoped tokens when auth is on.
• Auth — lock a dashboard behind HTTP Basic or an API key with one config block.

---

Documentation

📖 Read them online at direndai.github.io/dashdown — itself a Dashdown site, built and deployed from docs/ by CI.

The full docs are themselves a Dashdown dashboard. Run them locally:

dashdown serve docs

Or read the source under docs/pages/ — getting started, configuration, writing pages, components, connectors, queries, the semantic layer, real-time data, filters, formatting, exporting, embedding, auth, and extending. The docs/ directory is itself a runnable Dashdown project, so it doubles as a worked example: dashdown serve docs.

---

Telemetry

Dashdown sends an anonymous usage event on dashdown serve / dashdown build (Dashdown version + OS only — never your data, queries, paths, or connector names), so we can see how many people use it. It's on by default and trivial to turn off:

dashdown telemetry off          # or: DASHDOWN_TELEMETRY=0 / DO_NOT_TRACK=1

dashdown telemetry status shows exactly what would be sent. See docs/pages/telemetry.md for the full policy.

---

License

Copyright © 2026 Dirend AI.

Dashdown is free and open source under the GNU AGPL-3.0-or-later. Use, modify, and self-host it (including commercially) at no cost; the only obligation is that users of your version can get its source.

A commercial license is available if AGPL doesn't fit — e.g. embedding Dashdown in a closed-source product. See LICENSING.md for the plain-English breakdown, and CONTRIBUTING.md before opening a PR.