Evidence — Code-Driven BI Dashboards

Overview

Evidence, the open-source BI framework that generates beautiful, interactive dashboards from SQL queries and Markdown. Helps developers build data reports as code, deploy them as static sites, and create self-service analytics without heavy BI tools.

Instructions

Project Setup

# Create a new Evidence project
npx degit evidence-dev/template my-dashboard
cd my-dashboard
npm install
npm run dev    # Dashboard at localhost:3000

Writing Reports

Evidence reports are Markdown files with embedded SQL:

<!-- pages/sales-overview.md — Sales dashboard page -->
# Sales Overview

```sql monthly_revenue
SELECT
  date_trunc('month', created_at) AS month,
  SUM(amount) AS revenue,
  COUNT(*) AS orders,
  SUM(amount) / COUNT(*) AS avg_order_value
FROM orders
WHERE created_at >= '2025-01-01'
GROUP BY 1
ORDER BY 1

Revenue has grown {fmt(monthly_revenue[monthly_revenue.length - 1].revenue, 'usd')} this month, a {pct_change} change from last month.

<LineChart data={monthly_revenue} x=month y=revenue yFmt=usd title="Monthly Revenue" />

<BarChart data={monthly_revenue} x=month y=orders title="Orders per Month" />

Revenue by Product

SELECT
  p.name AS product,
  SUM(oi.quantity * oi.unit_price) AS revenue,
  SUM(oi.quantity) AS units_sold
FROM order_items oi
JOIN products p ON p.id = oi.product_id
JOIN orders o ON o.id = oi.order_id
WHERE o.created_at >= current_date - interval '30 days'
GROUP BY 1
ORDER BY 2 DESC
LIMIT 10

<DataTable data={product_breakdown} rows=10> <Column id=product title="Product" /> <Column id=revenue title="Revenue" fmt=usd /> <Column id=units_sold title="Units Sold" fmt=num0 /> </DataTable>

<BarChart data={product_breakdown} x=product y=revenue yFmt=usd swapXY=true title="Top 10 Products by Revenue" />

### Data Source Configuration

Connect to your databases:

```yaml
# sources/mydb/connection.yaml — PostgreSQL connection
name: mydb
type: postgres
host: localhost
port: 5432
database: analytics
user: evidence_reader
password: ${POSTGRES_PASSWORD}    # From environment variable
ssl: true

# sources/duckdb/connection.yaml — DuckDB for file-based analytics
name: local
type: duckdb
filename: data/analytics.duckdb

# sources/bigquery/connection.yaml — Google BigQuery
name: warehouse
type: bigquery
project_id: my-gcp-project
dataset: analytics
credentials_path: ./gcp-key.json

# sources/csv/connection.yaml — CSV files
name: csvdata
type: csv
path: data/        # All CSV files in this directory become tables

Interactive Components

<!-- pages/cohort-analysis.md — Interactive filters -->
# Cohort Analysis

<!-- Dropdown filter -->
<Dropdown name=time_period title="Time Period">
  <DropdownOption value="7 days" valueLabel="Last 7 Days" />
  <DropdownOption value="30 days" valueLabel="Last 30 Days" />
  <DropdownOption value="90 days" valueLabel="Last 90 Days" />
</Dropdown>

<Dropdown name=segment title="Customer Segment" data={segments} value=id label=name />

```sql cohort_data
SELECT
  date_trunc('week', first_purchase) AS cohort_week,
  weeks_since_first AS week_number,
  COUNT(DISTINCT customer_id) AS customers,
  SUM(revenue) AS revenue
FROM customer_cohorts
WHERE first_purchase >= current_date - interval '${inputs.time_period}'
  AND segment = '${inputs.segment.value}'
GROUP BY 1, 2
ORDER BY 1, 2

<Heatmap data={cohort_data} x=week_number y=cohort_week value=customers title="Customer Retention by Cohort" />

<!-- Big number KPIs --> <BigValue data={cohort_data} value=customers title="Total Customers" fmt=num0 comparison=revenue comparisonTitle="Total Revenue" comparisonFmt=usd />

### Templated Pages

Generate pages dynamically from data:

```markdown
<!-- pages/products/[product_name].md — One page per product -->
# {params.product_name}

```sql product_detail
SELECT * FROM products WHERE slug = '${params.product_name}'

SELECT
  date_trunc('day', o.created_at) AS date,
  SUM(oi.quantity) AS units,
  SUM(oi.quantity * oi.unit_price) AS revenue
FROM order_items oi
JOIN orders o ON o.id = oi.order_id
JOIN products p ON p.id = oi.product_id
WHERE p.slug = '${params.product_name}'
  AND o.created_at >= current_date - interval '90 days'
GROUP BY 1
ORDER BY 1

<LineChart data={product_sales} x=date y=revenue yFmt=usd /> <LineChart data={product_sales} x=date y=units />

### Deployment

```bash
# Build static site
npm run build

# Deploy to any static hosting
# Vercel
npx vercel

# Netlify
npx netlify deploy --prod --dir=build

# Evidence Cloud (managed hosting)
npx evidence deploy

# Schedule refreshes with cron
# Evidence rebuilds queries at build time
# Set up a cron job to rebuild periodically:
# 0 */4 * * * cd /app/dashboard && npm run build && cp -r build /var/www/dashboard

Examples

Example 1: Creating a weekly SaaS metrics report

User request:

Build an Evidence dashboard that shows our weekly SaaS metrics — MRR, churn rate, new trials, and conversion rate — from our PostgreSQL database.

The agent scaffolds an Evidence project, configures the PostgreSQL connection in evidence.plugins.yaml, creates SQL queries for each metric (select date_trunc('week', created_at) as week, sum(amount) as mrr from subscriptions...), builds a Markdown page with <LineChart>, <BigValue>, and <DataTable> components, and adds date range inputs with <DateRange> for filtering.

Example 2: Building a templated customer health page

User request:

I need a per-customer detail page in Evidence that shows usage trends, support tickets, and renewal date for each customer.

The agent creates a templated page at pages/customers/[customer_id].md, writes SQL queries that filter by ${params.customer_id}, adds a customer index page with <DataTable> linking to each detail page, and includes <BarChart> for usage and <Alert> components for upcoming renewals.

Guidelines

1. SQL is the source of truth — Write queries directly in Markdown; no abstraction layer between you and the data
2. Use parameterized queries carefully — Evidence supports ${inputs.x} but sanitize inputs; prefer dropdown constraints over free text
3. One page per topic — Keep reports focused; use navigation for different areas (sales, product, customer)
4. Templated pages for catalogs — Use [param].md for product pages, customer profiles, or any entity-level reports
5. Read-only database users — Connect Evidence with a read-only database user; it should never write data
6. Version control reports — Evidence projects are code; store in Git, review in PRs, deploy via CI
7. Build-time queries — Queries run at build time, not on page load; schedule rebuilds based on data freshness needs
8. DuckDB for local analysis — Use DuckDB as a data source for CSV/Parquet files; no database server needed