JanVayu जनवायु
A citizen-led national archive of India's air quality crisis
About
JanVayu (जनवायु, "People's Air") is a citizen-led initiative to build India's first comprehensive public archive of the air quality crisis: its data, its victims, its policies and its public memory.
It keeps a record and runs no campaign.
Live at https://www.janvayu.in
Recent highlights (September 2026)
Forty-three years of PM2.5 are now on the site for 783 of 785 districts. The figures are annual and seasonal (four seasons), run from 1980 to 2022, and come from LongPMInd (Wang et al., Earth System Science Data, 2024; CC BY 4.0). The airshed panel charts them. New Delhi stays flat near 55 µg/m³ through the 1990s, climbs from 1999 and levels off near 90. In winter it rises from about 65 in the 1980s to about 138 in 2013–2022. These decade means match the ones in Hawa Ka Hisab's "Destruction of Delhi" report (51 for the 1980s, 87 for the 2010s), but that report is built on the same LongPMInd data, so the match is a consistency check and not independent confirmation. The series is a reconstruction. It should not be subtracted from the roughly 1 km 2024 satellite layer to show change.
A new panel, "Your airshed, or your town?", asks where a district's air comes from. Across districts, 89.2% of the variance in annual PM2.5 lies between states, and the remainder lies within them. NCAP sets targets city by city. If most of a city's burden arrives from its airshed, a city acting alone can only reach what is left. Pick a district and the panel shows the split. It says plainly that the local gap names no cause.
Two global land-pressure maps, Biodiversity Intactness and Human Footprint (Impact Observatory / Vizzuality), were tested and left out. Across districts they correlate with PM2.5 at −0.61 and +0.62, better than anything else on the site. Once differences between states are held fixed, the added explanatory power (incremental R²) falls to +0.007 and +0.008. At 100 m resolution they mostly show where the Gangetic Plain is. The dataset that looked like our best predictor, and was a map of the Gangetic Plain also describes the join error that nearly led to the wrong table being published.
The monthly update of the current-year air layer had never run. The layer says it is rebuilt on the 3rd of each month, but its first scheduled run failed because a folder it needed to write to did not exist. That is fixed, and the fix was tested with the workflow's own command.
Earlier highlights (August 2026)
The map's air figures are the 2024 satellite annual mean, because SatPM2.5 V6GL03 has published nothing newer. A second layer, at district level, now answers the question everyone asks about this year so far. It uses CAMS estimates via Open-Meteo, corrected against the satellite series for 2024 (r = 0.91, held-out RMSE 5.76 µg/m³ over 200 splits), and is rebuilt on the 3rd of each month. It is never merged with the annual layer, never drawn below district level and never called a measurement.
The site now leads with PM2.5 and not AQI, after a fair criticism at a conference. AQI is a unitless index that reports only its worst pollutant and cannot be averaged over a year. Every Indian limit, health study and NCAP target is written in µg/m³ of PM2.5. The reasoning is in Why JanVayu reports PM2.5 first and AQI second.
Field Testimony grew from 142 to 250 voices, across 107 cities and 14 languages. Field-collected quotes carry their collection date and mode. Every speaker consented and is named as they asked, and code-mixed speech is marked as spoken. The wall is shuffled daily, so no voice stays at the bottom.
The social feeds now say what they can and cannot do. Reddit works again through its public Atom feed, after its JSON interface began refusing requests from data-centre addresses. YouTube is read through channel feeds, and searched when a free Data API key is set. X and Instagram are links out, because neither can be read without a paid or authenticated API. Every X link is checked against X's public embed endpoint before it goes live.
Three things on the site now check themselves. Every stated figure is recomputed from the data, and a mismatch stops a release; this caught a photo count and a ward count that had each been five releases out of date. The homepage blog list is generated from the blog itself. The current-year layer rebuilds monthly.
Two new posts: How to read the JanVayu map, a plain reader's manual, and What the air looks like, on why a data site carries 31 photographs of something 2.5 µm across.
Features
| # | Feature | Description |
|---|
| 1 | The first screen | Asks what you are breathing. You enter a city, ward or village, or let it read the nearest station, and it replies with an instruction instead of a number. The headline is the verdict, with the reading, the CPCB band, and the same answer for a child, for asthma or COPD, and for going running. Bands follow the CPCB National AQI PM2.5 24-hour sub-index thresholds (30/60/90/120/250 µg/m³) |
| 2 | Ask JanVayu on the homepage | A question box under the first screen, with nine worked examples, in 10 Indian languages. Answers appear in place with their source named, and default to the place you just searched. /ask/ keeps conversation history and offers the installable app |
| 3 | Role-based views | An optional view for 12 audience roles: parent, student, researcher, policymaker, journalist, citizen, activist, doctor, teacher, NGO, business owner, and woman/caregiver. It is offered from the header switcher and never stands in front of the homepage |
| 4 | Simple language mode | A toggle in the header that switches the whole site to simple language. The choice lasts for the browser session |
| 5 | Glossary (Ctrl+K) | A searchable glossary of air quality terms, opened with Ctrl+K |
| 6 | Intro tour | A guided walkthrough of the main sections for first-time visitors |
| 7 | Real-time AQI dashboard | Live PM2.5 and AQI for 157 Indian cities from WAQI and CPCB, with Beijing, London and Singapore alongside for comparison. 33 cities refresh every 10 minutes, and the rest are fetched when you select them |
| 8 | Interactive AQI map | A map with station-level AQI markers across India. You can switch on layers from indianopenmaps.com: live-AQI shading by Lok Sabha constituency ("the air your MP answers for") and by district, assembly-constituency boundaries, and a pollution-sources layer showing landfills, dumpsites, coal mines, CPCB red and orange category industrial parks, and SEZs |
| 9 | Health impact research | Curated evidence from Lancet Countdown 2025, Harvard and IHME studies |
| 10 |
Key statistics (July 2026)
| Metric | Value | Source |
|---|
| Annual PM2.5 Deaths | 1.72 million | Lancet Countdown 2025 |
| Economic Cost | $339.4 billion (9.5% GDP) | Lancet Countdown 2025 |
| India's Global Share | India and China each recorded more than 2 million air-pollution deaths in 2023 (of 7.9 million worldwide) | State of Global Air 2025 |
| Most Polluted Capital | New Delhi (82.2 µg/m³) | IQAir 2025 |
| Most Polluted City | Loni, India (112.5 µg/m³) | IQAir 2025 |
| Cities Meeting WHO Guideline | Only 14% globally | IQAir 2025 |
| India Average PM2.5 | 48.9 µg/m³ (nearly 10 times the WHO limit) | IQAir 2025 |
| Life Expectancy Loss (India) | 3.5 years | AQLI 2025 |
How it is built
JanVayu is a light site with no front-end framework. Social and news feeds are gathered on the server:
┌─────────────────────────────────────────────────────────┐
│ Client (Browser) │
│ Single-page HTML app · Chart.js · Leaflet.js · WAQI │
└──────────────────────────┬──────────────────────────────┘
│
HTTPS (Netlify CDN)
│
┌──────────────────────────▼──────────────────────────────┐
│ Netlify Functions │
│ │
│ Scheduled (cron) On-demand (API) │
│ ┌──────────────────┐ ┌───────────────────────────┐ │
│ │ scheduled-fetch │ │ reddit-feed.js │ │
│ │ (every 4 hours) │ │ twitter-feed.js │ │
│ │ │ │ news-proxy.js │ │
│ │ daily-digest │ │ instagram-feed.js │ │
│ │ (8 AM IST daily) │ │ feed-status.js │ │
│ └────────┬──────────┘ │ subscribe.js │ │
│ │ └─────────────┬─────────────┘ │
│ │ │ │
│ ▼ ▼ │
│ ┌─────────────────────────────────────────────┐ │
│ │ Netlify Blobs (Cache) │ │
│ │ Feeds cached as JSON · Strong consistency │ │
│ └─────────────────────────────────────────────┘ │
│ │
│ ┌─────────────────────────────────────────────┐ │
│ │ Resend (Email Delivery) │ │
│ │ Daily AQI digest to subscribers │ │
│ └─────────────────────────────────────────────┘ │
└──────────────────────────────────────────────────────────┘
Main design choices:
- The whole front end is one
index.html with its CSS and JavaScript inline. There is no build step, bundler or framework.
- Social media and news APIs are fetched through Netlify Functions, which avoids browser cross-origin limits and keeps API keys private.
- A scheduled function fetches all feeds every 4 hours and stores them in Netlify Blobs. The on-demand API functions read from that store, so responses are instant.
- The browser calls the WAQI API directly every 10 minutes. The token is a free-tier public key.
- A scheduled function runs daily at 8:00 AM IST (2:30 AM UTC), reads subscribers from Blobs, fetches current AQI and sends personalised HTML emails through Resend.
Automatic update schedule
| Task | Frequency | Mechanism |
|---|
| Social/news feed refresh | Every 4 hours | scheduled-fetch.mjs (Netlify Scheduled Function) |
| Daily AQI email digest | Daily at 8:00 AM IST | daily-digest.mjs (Netlify Scheduled Function) |
| Live AQI dashboard refresh | Every 10 minutes | Client-side JavaScript (WAQI API) |
| Current-year district air layer | 3rd of each month | current-year-air.yml → scripts/build-current-year-air.py |
| Stated figures vs the data | Every push and PR | guard-site-figures → scripts/check-site-figures.py |
| Homepage blog list vs the blog | Every push and PR | scripts/build-blog-index.py --check |
| Social + news sweep for the Voices panel | Mondays, 04:00 UTC | Scheduled Claude routine (opens a branch for review) |
Project structure
JanVayu/
├── index.html # Main website (single-page application)
├── favicon.svg # Site favicon
├── package.json # Node.js dependencies (Netlify Blobs, Resend)
├── netlify.toml # Netlify build & deploy configuration
├── .gitignore # Ignored files (node_modules/, .netlify/)
├── CNAME # Custom domain configuration
├── README.md # This file
├── CONTRIBUTING.md # Contribution guidelines
├── CODE_OF_CONDUCT.md # Community standards
├── LICENSE # MIT (code) + CC BY-NC-SA 4.0 (content)
├── blog/ # Blog (Docsify-powered, Markdown posts)
│ ├── index.html # Docsify blog config
│ ├── _sidebar.md # Blog navigation
│ ├── README.md # Blog home page
│ └── posts/ # Blog posts (YYYY-MM-DD-slug.md)
├── downloads/ # Downloadable reports and datasets
├── netlify/
│ └── functions/ # Netlify serverless functions
│ ├── scheduled-fetch.mjs # Cron: fetches all feeds every 4 hours
│ ├── daily-digest.mjs # Cron: sends daily AQI email digest
│ ├── reddit-feed.js # API: serves cached Reddit posts
│ ├── twitter-feed.js # API: retired; it read Nitter, whose public instances are gone, and nothing calls it
│ ├── news-proxy.js # API: serves cached news articles
│ ├── instagram-feed.js # API: serves cached Instagram posts
│ ├── feed-status.js # API: reports feed freshness and health
│ ├── subscribe.js # API: manages email subscriptions
│ └── blob-store.js # Shared: Netlify Blobs store helper
└── node_modules/ # Dependencies (git-ignored)
Technical stack
| Layer | Technology | Purpose |
|---|
| Frontend | Vanilla HTML / CSS / JavaScript | Zero-dependency single-page application |
| Charts | Chart.js | AQI trends and health data visualizations |
| Maps | Leaflet.js + OpenStreetMap | Interactive AQI station maps |
| AQI Data | WAQI API | Real-time air quality from the stations WAQI aggregates for India |
| Serverless | Netlify Functions | Server-side API proxying and scheduled tasks |
| Caching | Netlify Blobs | Persistent JSON cache with strong consistency |
| Email | Resend | Transactional email delivery for daily digests |
| Hosting | Netlify (auto-deploy from GitHub) | CDN, edge functions, scheduled functions |
| Domain | janvayu.in | Custom domain via Netlify DNS |
Environment variables
The following environment variables must be configured in the Netlify dashboard for full functionality:
| Variable | Required | Description |
|---|
RESEND_API_KEY | Yes | API key from Resend for sending email digests |
RESEND_FROM | Yes | Verified sender email address (e.g., [email protected]) |
BLOB_TOKEN | Yes | Netlify personal access token for Blob store access |
NETLIFY_SITE_ID | Yes | Netlify site identifier (used by Blob store and scheduled functions) |
GROQ_API_KEY | Yes | Groq API key for AI features; runs the openai/gpt-oss-120b open model (free at console.groq.com). Optional GROQ_MODEL overrides the model without a code change. |
Note: The WAQI API token is the WAQI_TOKEN environment variable in Netlify. It is not in the repository, and the browser reaches WAQI through /.netlify/functions/waqi-proxy. See environment variables.
Self-hosting and local development
Prerequisites
Setup
# Clone the repository
git clone https://github.com/JanVayu/JanVayu.git
cd JanVayu
# Install dependencies
npm install
# Set environment variables (create a .env file or configure in Netlify CLI)
export RESEND_API_KEY="your_resend_api_key"
export RESEND_FROM="[email protected]"
export BLOB_TOKEN="your_netlify_personal_access_token"
export NETLIFY_SITE_ID="your_netlify_site_id"
# Start local development server with Netlify Functions support
netlify dev
The site will be available at http://localhost:8888. Netlify Dev emulates the serverless functions locally so you can test the full stack.
Without Netlify Functions
If you only need the front-end (no social feeds or email digests):
# Serve index.html with any static file server
npx serve .
# or
python3 -m http.server 8000
The AQI dashboard will work without any server-side setup since it calls the WAQI API directly from the browser.
Data sources
JanVayu integrates public data sources, including:
Roadmap
Full phased roadmap: docs/wiki/Roadmap.md · tracked on GitHub Issues.
Recently shipped (v26.6.242): the Delhi government's claim of 211 good-to-moderate days in 2026 is checked against CPCB's own bulletins in Delhi's 211 days of good-to-moderate air, checked day by day: the 2026 count holds to within a day, the years it was compared with are full-year totals, and on the same dates 2026 still leads 2023 by 7 days. The CPCB bulletin layer runs to 19 October and covers 300 cities and 517,157 city-days. Every page now carries a share bar and a preview card, blog posts have static pages a link previewer can read, and figures that were not data (PM2.5 back-calculated from AQI, a demo dataset, synthetic charts) were removed (v26.6.240 to 241). Third-party code on the docs and blog readers is pinned with integrity hashes, Pretext is served from this site, and unit tests now run in CI.
Before that (v26.6.235): every standalone page now loads the shared stylesheet and follows the theme control, and the choice is remembered between pages (v26.6.232). Five of the six pollutant pages printed a random value under a real unit from 26 April to 22 September 2026. They now show the per-pollutant sub-index that WAQI reports, labelled as one, and Five of our six pollutant pages were printing random numbers. CPCB's own daily AQI bulletins, 2015 to 2026, cover 297 cities and 507,334 city-days, with the monitor count beside every figure and no trend line, because the number of stations behind each city grew over the period (v26.6.199 to 202). Weather-normalised trends for 44 cities replaced the single Delhi-NCR run as the national view (v26.6.193), and a colour-contrast check now sweeps every page in both themes before a pull request can merge.
Earlier (v26.6.155): PM2.5 now leads everywhere, after a criticism at a conference that was fair. AQI is a unitless index that reports only its worst pollutant and cannot be averaged over a year, while every Indian limit, health study and NCAP target is written in µg/m³ of PM2.5. The decks, dashboard and section copy now say so, and Why JanVayu reports PM2.5 first and AQI second. How to read the JanVayu map is a reader's manual for the atlas, not an engineering note. The Field Testimony wall grew from 142 to 250 voices across 107 cities and 14 languages, with field-collected quotes stamped by collection date and mode and code-mixed speech marked as spoken. The Reddit feed works again through the public Atom feed after Reddit's JSON API began refusing data-centre addresses, and X/Twitter was removed rather than left as an empty tab.
Before that (v26.6.153): the boundary atlas is complete. A single Boundaries menu covers 983,149 areas across seven levels (state, district, block/mandal/tehsil, gram panchayat and village on the rural side, city and ward on the urban side), and each carries the same nine measures: annual PM2.5, the same air split into four seasons, surface heat from a national Landsat mosaic, and tree, green and built-up cover from ESA WorldCover. Six levels ship as PMTiles read by HTTP range request. Villages carry the same numbers in per-district TopoJSON, because a 267 MB tile archive cannot ship. A Compare panel plots any two measures against each other for whatever is on screen. The per-city ward panel is retired, and #ward-map points at the map.
Next up:
- Publish the boundary tiles as a release.
scripts/fetch-tiles.mjs is written and unwired. Publishing the archives would take about 159 MB (data/tiles) out of the working tree, and would let villages use PMTiles like the other six levels instead of the per-district loader.
- Reduce the repository's weight. The working tree is ~395 MB (excluding
.git; data/villages is ~192 MB and data/tiles ~159 MB). A shallow-clone or data-split path would help contributors.
- Answer whole-country correlations. The Compare panel covers only what is on screen, and says so. A precomputed statistics file would let it answer nationally.
- Cover the states no ward source reaches, Manipur, Mizoram, Srinagar and Siliguri among them. An RTI to West Bengal Municipal Affairs is the realistic route for Siliguri.
- Fix Thiruvananthapuram's last ward with no heat figure. It sits in a Landsat coverage seam, and the national mosaic closed five of six.
Contributing
We welcome contributions from:
- Researchers, with datasets, papers and analysis
- Journalists, with investigations and verified reports
- Developers, with code, visualizations and tools
- Citizens, with testimonies, local documentation and translations
- Designers, with accessibility and communication work
See CONTRIBUTING.md for guidelines and the Collaborator Statement of Work for detailed workstream descriptions.
Governance
JanVayu grew out of the #AQIForJanHit campaign of the All India Professionals' Congress (AIPC), which describes itself as a political platform, and was designed and built under AIPC's Dr. Manmohan Singh Fellows Programme. Its code, data, sources and editorial rules are public, and the same sourcing rules apply to every government and party.
Editorial decisions are guided by:
- Factual accuracy and verification
- Source transparency
- Respect for affected communities
- Accessibility across languages and regions
Name change note
This project was previously known as "Vayu Smriti" (वायु स्मृति). It was renamed to JanVayu (जनवायु) for better linguistic inclusivity across India's diverse language communities.
Forking and reuse
JanVayu is designed to be forked for other cities, regions, or countries. Designed to run on the free tiers of WAQI, Groq, Resend and Netlify (free-tier limits change; check each provider).
See FORKING.md for a complete guide to what to change, which API keys you need and how to attribute the project.
If you create a fork, let us know and we'll list it in the README.
License
- Code: MIT License. Use it freely, modify it, redistribute it.
- Content and documentation: CC BY-NC-SA 4.0. Share with attribution, non-commercially, under the same license.
- Data: individual sources keep their original licenses.
See LICENSE for details. If you fork this project, please credit JanVayu as the upstream source.
Contact
Support
JanVayu is a public interest project. If you wish to support:
- Contribute data or expertise (see Contributing above)
- Report a problem by opening an issue
- Share the website
JanVayu is built on the principle that public memory is a prerequisite for public accountability.
जनवायु, क्योंकि हवा सबकी है।
