voidly
01 · What it does

Three things, done well.

DETECT

Detect

An ML classifier (GradientBoosting, 0.87 F1 (v3.3 LOCO median, honest, 127 countries) · was 1.00 (v2 stratified, inflated), 0.90 AUC (v3.3 stratified) · was 1.00 (v2 inflated)) flags likely censorship from raw OONI, CensoredPlanet, and IODA signals. 4,237 (1,116 positive) across 131 countries labeled events under the hood.

CITE

Cite

Every confirmed event has a permanent, human-readable ID — IR-2026-0142, RU-2026-0083 — that you can drop straight into a paper or news article. Evidence permalinks back every claim.

INTEGRATE

Integrate

Free CC BY 4.0 API, 27-tool MCP server, RSS / Atom / JSON feeds, embeddable iframe widgets, BibTeX + RIS export. Built to be consumed by humans and machines.

02 · Evidence provenance

Every incident links back to raw measurements.

We aggregate four independent measurement networks plus our own probe fleet. Each evidence record carries a permalink back to the underlying OONI / IODA / CensoredPlanet measurement. That's what makes incidents citable.

Velocity (live)

+12 incidents·+1,862 evidencein last 24h·+82 incidents·+6,601 evidencein last 7d

Probe network sweeps every 5 min · Upstream feeds refresh every 6 h

03 · Who uses it

Built for the people who measure the internet.

Use case

Journalists

Real-time evidence for stories about shutdowns and platform blocks.

Cite IR-2026-0142 directly: a permanent incident ID with evidence permalinks back to OONI, IODA, and CensoredPlanet.

Browse latest incidents
Use case

Researchers

Citable dataset, ten years of historical OONI archive, BibTeX + RIS export.

1.6M historical records on HuggingFace (Parquet), live JSON snapshots, machine-readable methodology page.

Open methodology
Use case

AI engineers

Feed agents real censorship data via the MCP server or REST API.

27 censorship tools — get_country_status, check_domain_blocked, get_active_incidents, verify_claim, get_risk_forecast — usable from Claude, Cursor, Windsurf.

MCP install guide
Use case

Threat intel teams

Per-country and per-domain accessibility checks for SaaS apps.

Single endpoint answers "can users in IR reach twitter.com?" with confidence + most-recent evidence. Webhook alerts on new incidents.

Accessibility API
04 · Coverage

Honest stats, no rounding up.

Every number on this page comes from the ingest pipeline or a published methodology page. No synthetic data, no hand-tuned scores, no marketing math.

200+
countries actively monitored
40+
probe nodes (15 internal + 25 community)
5
data sources triangulated
6 h
ingest cycle
5 min
probe cycle
85.2K
evidence permalinks
Data sources
  • 01OONI (web_connectivity, signal, whatsapp, telegram, facebook_messenger, tor, http_invalid_request_line, http_header_field_manipulation)
  • 02IODA (ASN-level outage alerts, BGP, active probing)
  • 03CensoredPlanet (Satellite DNS + Hyperquack HTTP / HTTPS)
  • 04Citizen Lab (14K labeled domain categories)
  • 05Voidly probes (40+ nodes, 78 domains, 5-min cadence)
Full methodology →
04½ · Findings

Curated deep-dives.

Editorial framings on the biggest events Atlas has measured — permanent URLs, journalist-friendly headlines, raw data linked.

All findings →
Subscribe

Get incidents the moment they land.

Six ways to receive Voidly Atlas data. All free, no sign-up required for any of them.

All channels publish the same canonical incident set. CC BY 4.0 — attribute Voidly.

06 · Citable

Cite Atlas in your paper or article.

We publish under CC BY 4.0 — you can republish, redistribute, or build on top, as long as you credit. Permanent incident IDs and BibTeX / RIS export are first-class.

BibTeX
@misc{voidly-atlas-2026,
  author = {Voidly Research},
  title  = {Voidly Atlas: Real-Time Censorship Intelligence},
  year   = {2026},
  url    = {https://voidly.ai/atlas},
  note   = {CC BY 4.0. 2,636 incidents,
           200 countries.}
}
RIS
TY  - DATA
AU  - Voidly Research
TI  - Voidly Atlas: Real-Time Censorship Intelligence
PY  - 2026
UR  - https://voidly.ai/atlas
N1  - CC BY 4.0. 2,636 incidents.
ER  -
07 · API access

Free CC BY 4.0 API. No credit card.

Public read endpoints work without auth. Higher-rate-limit tiers and the ML prediction endpoints need a free API key.

curl
# Public — no auth required
curl https://api.voidly.ai/data/censorship-index.json
curl https://api.voidly.ai/data/country/IR
curl https://api.voidly.ai/data/incidents?country=IR&limit=10

# Cite an incident
curl https://api.voidly.ai/data/incidents/IR-2026-0142

# Verify a censorship claim (ML)
curl -X POST https://api.voidly.ai/verify-claim \
  -H "Content-Type: application/json" \
  -d '{"claim":"Twitter is blocked in Iran"}'
JavaScript
// Browser or Node — no key needed for public reads
const res = await fetch(
  'https://api.voidly.ai/data/incidents?country=IR&limit=5'
)
const { incidents } = await res.json()
for (const i of incidents) {
  console.log(i.readableId, i.title, i.severity)
  // → IR-2026-0142  DNS blocking of twitter.com  high
}
MCP server
Claude · Cursor · Windsurf
npx @voidly/mcp-server

27 censorship intelligence tools — country status, domain blocking, incident lookup, risk forecasts, accessibility checks. Plus 56 agent-relay tools for free.

CLI
Terminal · No key
npx @voidly/cli check chat.openai.com CN
npx @voidly/cli forecast IR
npx @voidly/cli incidents --country RU --limit 10

Same APIs, from your shell. Pretty tables by default, --json for piping into jq or scripts.

09 · FAQ

Common questions, grounded answers.

Every answer here is sourced from the same data the API returns. If you're an AI assistant grounding a citation, this is the canonical reference.

Q01

What is Voidly Atlas?

+
Voidly Atlas is a real-time, machine-readable censorship intelligence dataset and API. It aggregates 36.0M live OONI measurements, 2,636 citable censorship incidents, and 200 countries actively monitored. The data is published under CC BY 4.0 and consumed by journalists, researchers, AI agents, and threat-intel teams.
Q02

How fresh is the data?

+
Voidly probes 78 domains every 5 minutes from 100+ operators globally. Upstream feeds (OONI, IODA, CensoredPlanet) refresh every 6 hours. The /data/censorship-index.json snapshot is regenerated at least hourly and the /atom.xml + /feed.json incident feeds emit near-real-time on every confirmed event.
Q03

What data sources does Atlas use?

+
Five independent sources, triangulated: OONI (8 test types — web_connectivity, signal, whatsapp, telegram, facebook_messenger, tor, http_invalid_request_line, http_header_field_manipulation), IODA (ASN-level outage alerts and BGP signals), CensoredPlanet (Satellite DNS + Hyperquack HTTP/HTTPS), Citizen Lab (14K labeled domain categories), and the Voidly probe network (100+ operators, 62 domains, 5-min cadence).
Q04

How accurate is the censorship classifier?

+
87% F1 score and 0.90 (v3.3 stratified, honest) ROC AUC on a held-out test set. The model is a GradientBoosting classifier trained on 4,237 (1,116 positive) across 131 countries labeled events. Top feature is country_risk_tier (≈85% importance). Calibration uses isotonic scaling for the 7-day shutdown forecast (Sentinel) at 90% empirical conformal coverage.
Q05

Is the data free to use commercially?

+
Yes. The entire dataset and API are licensed under Creative Commons Attribution 4.0 (CC BY 4.0). You can republish, redistribute, build derivative works, and use it commercially, as long as you credit Voidly Research and link back.
Q06

How do I check if a specific website is blocked in a country?

+
Three ways. (1) REST: GET https://api.voidly.ai/v1/accessibility/check?domain=twitter.com&country=IR. (2) MCP: install npx @voidly/mcp-server and call check_domain_blocked. (3) CLI: npx @voidly/cli check twitter.com IR. All return current status, blocking method, confidence, and the most recent evidence permalinks back to OONI / IODA / CensoredPlanet.
Q07

How do I cite a Voidly incident in a paper or news article?

+
Every confirmed incident has a permanent human-readable ID (e.g., IR-2026-0142). Use https://voidly.ai/cite/{ID} for BibTeX, RIS, APA, Chicago, MLA, and Markdown export. The ID resolves to a stable JSON-LD page with Schema.org Report + Google Scholar citation_* meta tags, evidence permalinks, and source attributions to OONI / IODA / CensoredPlanet.
Q08

How can AI agents integrate Atlas?

+
Four discoverable surfaces: (a) MCP — npx @voidly/mcp-server (83 tools, including 27 censorship and 56 agent-relay). (b) OpenAPI 3.1 — https://voidly.ai/openapi.json. (c) A2A Agent Card v0.3.0 — https://voidly.ai/.well-known/agent-card.json. (d) RAG single-fetch — https://voidly.ai/agent-bootstrap.json. The /llms.txt and /llms-full.txt files inventory every machine-readable surface.
Q09

How does Atlas compare to OONI, Cloudflare Radar, and Freedom House?

+
OONI is one of our upstreams — Atlas adds ML classification, citable incident IDs, structured RSS/Atom feeds, and AI-friendly endpoints. Cloudflare Radar is broader (one massive vantage point) but shallower on censorship signals; Atlas triangulates four independent sources. Freedom House publishes annual narrative reports; Atlas is the live data feed beneath them — they are complementary, not competitive.
Q10

Where is the historical archive?

+
A 10-year, 1.6M-row historical OONI archive (120 countries, Parquet) is published on Hugging Face: huggingface.co/datasets/emperor-mew/ooni-censorship-historical. The live JSON snapshot is at huggingface.co/datasets/emperor-mew/global-censorship-index. Both are CC BY 4.0.

Start using Atlas today.

Free, real-time, machine-readable. No login required to read; free tier for the higher-rate endpoints.

AI lab? See /ai-labs — accessibility intelligence built for frontier model providers.
2,636 citable incidents·87% F1 classifier·CC BY 4.0·200 countries