# Voidly — Historical Fact Sheet This dated record describes the project as it was documented in August 2026. For current company identity and product descriptions, use [About Voidly](https://voidly.ai/about) and the [brand kit](https://voidly.ai/branding). Product availability and payment permissions must be checked on the current product pages. Last updated: 2026-08-23 ## Company - **Name**: Voidly - **Founded**: 2024 - **Team**: Two-person team, AI-augmented engineering. The full stack — data ingestion, ML, public APIs, agent SDK, payments, frontends, mobile, and infrastructure — is built and shipped lean. - **Mission**: Make internet censorship measurable, accessible, and actionable for AI systems, researchers, journalists, and developers. ## Products | Product | One-liner | | ------------------- | -------------------------------------------------------------- | | **Voidly Atlas** | Network intelligence with documented country coverage. | | **Voidly Relay** | E2E encrypted agent-to-agent messaging. | | **voidpay** | Hosted encrypted zero-value private-session staging; no wallet or payment route. | | `+` Voidly Sentinel | Shutdown early-warning forecasts (XGBoost + isotonic). | | `+` Voidly Messenger| E2E encrypted messaging PWA at `msg.voidly.ai`. | | `+` AI Censorship Index | Only live public dataset of where major AI services are reachable, by country. | | `+` `@voidly/mcp-server` | 84 tools, MCP-Registry listed. | ## August 2026 observations and current sources The measurements below with explicit August dates are historical snapshots. For current country, incident, evidence, and probe totals, read the linked public responses. - **38.8M** OONI measurements indexed all-time (38,779,164 on 2026-08-25 — sum of row totals in ). This spans **125** countries, not 130: five of the indexed countries (BS, SR, DK, KP, ER) have no row in the all-time snapshot and contribute only to the rolling window below. - **347K** OONI measurements in the rolling 30-day scoring window (346,718 on 2026-08-25 — `metadata.liveWindowMeasurements` at ). This is the window the published censorship scores are computed over; it is a *different corpus* from the line above and roughly 112× smaller. Never quote one under the other's label. > **Correction, 2026-08-25.** These two lines replace a single line that read > "**16M** live OONI / CensoredPlanet / IODA measurements (rolling 30 days)". > Every part of it was wrong except the window length. 16,000,000 was not a > measurement — the API route served `Math.max(derived, 16000000)`, so the > constant was returned on every request the route ever answered, 46× above its > own derived basis of 346,718 and simultaneously 2.4× *below* the cumulative > corpus a reader would take "16M processed" to mean. The source attribution was > wrong too: this count is OONI only. CensoredPlanet and IODA feed the evidence > and incident corpora below, not this number. - **3.5B+** measurements published by **OONI** — the upstream corpus Voidly indexes *from*. This is OONI's number, not Voidly's, and it is not a cross-source total. Check it yourself: `curl https://api.ooni.io/api/_/global_overview` → `measurement_count` (3,599,731,972 on 2026-08-25; it grows continuously). > **Retracted 2026-08-25:** this line read "**2.2B+** underlying raw measurement > corpus aggregated" and was published as an OONI + CensoredPlanet + IODA > aggregate. It was false in the direction nobody audits: the claimed > three-source total sat ~1.4 **billion below OONI alone**. An inflated number > invites scrutiny; a deflated one reads as conservative, which is how this one > survived for months on a press page. No honest cross-source sum is available > either — IODA publishes no measurement count, and CensoredPlanet's "60B+" is > undated prose on a webpage, not an endpoint. So the corpus is now quoted from > the one component that answers a live request, attributed by name. - **1.6M** historical records — 10-year OONI archive. - **Incident and evidence counts**: the [live incident statistics](https://api.voidly.ai/v1/censorship-index/incidents/stats) separate all tracked records from citable censorship, evidence-backed, and multi-source subsets. The total incident count is not a count of verified censorship. Stable human-readable IDs are available for individual records. - **Country index coverage**: read the `metadata.totalCountries` field and `countries` array in the [live index](https://api.voidly.ai/data/censorship-index.json). The index country count is not the probe network's country count. - **AI Censorship Index** — a live public dataset of where major AI services (ChatGPT, Claude, Gemini, Perplexity, Hugging Face) are reachable, measured network-level where observations exist. See [current AI service results](https://voidly.ai/ai-blocked) for the measured coverage. - **Probe network**: the [live roster](https://api.voidly.ai/v1/probe/network) publishes active core and community nodes; the [probe statistics](https://api.voidly.ai/v1/probe/stats) publish measured country coverage. These populations differ from index coverage. - **Classifier evaluation**: use the [current model card](https://api.voidly.ai/v1/classifier/info) and its LOCO mean headline, training-label composition, and small-sample median caveat. The older v2 score was retired in May 2026 after a label-derived feature was found to inflate it. - Framework adapters for **LangChain, CrewAI, AutoGen and LlamaIndex** (published to PyPI; each targets the voidpay credit rail, which answers HTTP 410 while that rail is withheld, so they are not currently callable end-to-end). > **Corrected 2026-08-25:** this line read "LangChain, CrewAI, AutoGen, > LlamaIndex, Haystack, Vercel AI, OpenAI-compat, and A2A" — eight names for > four adapters. There is no Haystack adapter anywhere in the tree (`git > ls-tree` returns zero matching paths; every code hit is a local variable > named `haystack`) and no PyPI package; Vercel AI has no adapter package; > and "OpenAI-compat" pointed at `openai-action/openapi.yaml`, a ChatGPT GPT > Action spec rather than a framework adapter. Only names with shipped, > published code are listed. ## Open Source - GitHub org: - npm: - `@voidly/mcp-server` — 84 tools (registry latest 2.17.0 on 2026-08-23) - `@voidly/agent-sdk` — Double Ratchet, X3DH, ML-KEM-768 PQ hybrid (registry latest 3.45.0 on 2026-08-23) - HuggingFace: - `emperor-mew/global-censorship-index` — live JSON - `emperor-mew/ooni-censorship-historical` — 1.6M-row Parquet archive ## Licensing - **Data**: CC BY 4.0 (every dataset published by Voidly) - **Code**: Mixed — repository root, contracts, and supporting libraries are MIT (`/LICENSE`, `/voidly-contracts/LICENSE`, etc.). The `@voidly/agent-sdk` package ships under a proprietary "All Rights Reserved" license — npm distribution is permitted, but the source is not open-licensed. Other packages indicate their license in their individual `LICENSE` file. ## Citation (BibTeX) ```bibtex @misc{voidly2026, author = {{Voidly}}, title = {{Voidly: Censorship intelligence + agent infrastructure for the post-AI internet}}, year = {2026}, howpublished = {\url{https://voidly.ai}}, note = {Open data, CC BY 4.0. Current country coverage: https://api.voidly.ai/data/censorship-index.json. Current incident and evidence totals: https://api.voidly.ai/v1/censorship-index/incidents/stats.} } ``` For a specific incident, cite the readable ID, e.g.: ```bibtex @misc{voidly_IR_2026_0142, author = {{Voidly}}, title = {{Censorship Incident IR-2026-0142}}, year = {2026}, url = {https://voidly.ai/censorship-index/incidents/IR-2026-0142}, note = {Voidly Open Censorship Data, CC BY 4.0} } ``` ## Contacts | Audience | Email | | ------------------- | ---------------------- | | Press / media | `press@voidly.ai` | | Investors | `investors@voidly.ai` | | Researchers / academia | `research@voidly.ai` | | Founder (direct) | `founder@voidly.ai` | ## 1-pager PDF Not yet generated. A printable 1-pager PDF for fundraising / press is pending. In the meantime, this file (`FACTS.md`) and the press page () cover the same ground. Email `press@voidly.ai` for an on-demand version. ## Useful links - Live data hub: - Live dashboard: - API docs: - Methodology: - Agent SDK: - AI Censorship Index: - Topic pages: