About
About Valyrian News Network
Transparency in Automated Journalism
Our Mission
Valyrian News Network exists to increase public understanding of global events by reading the world's local news in its own languages and synthesising it into clear, sourced English-language reporting. Most international coverage is filtered through a handful of wire services; VNN goes back to the front pages residents actually read in each country and brings that perspective to a global audience.
How VNN Works
VNN is a fully automated pipeline of specialised AI agents built on Valyrian Tech's SERENDIPITY framework. Each stage runs on a schedule, and every story moves through six distinct steps before it reaches this site.
Gathering
Country-specific gatherer agents visit local news sources on a daily schedule and collect headlines, URLs and categories — in the local language. Each covered country is gathered around its own local news cycle, so the pipeline reads each country's front page roughly when residents do.
Compiling
A compiler agent clusters articles that report on the same event (not just similar topics) across multiple outlets, translates non-English content into English, and produces a structured story with title, summary, category and key points. As more sources publish throughout the day, stories are recompiled to incorporate fresh reporting.
Ranking
Compiled stories are ranked by importance through a consensus vote among seven independent ranking agents, each bringing a different editorial lens — balanced perspective, historical context, analytical reasoning, impact assessment, global implications, human interest and trend analysis. Votes are cast and aggregated using the open-source Hivemind Protocol, a Condorcet-style ranked-choice voting system that stores ballots on IPFS and verifies them with Bitcoin-signed messages. The result is a single, auditable publication order per country.
Research
A research agent works through the ranked stories one at a time. For each story it:
- • Fetches the full text of every source article
- • Searches for additional reporting, prioritising local-language sources
- • Extracts key facts, direct quotes, timelines and the people involved
- • Produces a research document where every claim is attributable to a source URL
Writing
A writer agent transforms the research into a publication-ready article following a strict structure: lead, context, developments, analysis and conclusion. Articles are constrained by a validator that checks:
- • Word count and required metadata fields
- • Title and meta-description length limits
- • That every URL appears in the research document (no invented links)
- • That sources are cited on first mention, with no duplicate links
Publishing & Revision
Before publication, a publisher agent verifies every URL in the article with a live browser check. If any link is broken, fabricated or insufficiently sourced, the article is rejected with specific revision notes and sent back to a revision agent that rewrites it. Only articles that pass URL verification and quality checks are committed to this website. Stories that cannot be salvaged after repeated attempts are discarded rather than published.
Our Standards
Source Integrity
- ✓ Every URL in a published article is verified live before publication
- ✓ Writers can only link to URLs that appear in the underlying research document
- ✓ Local-language sources are preferred over second-hand translations where available
- ✓ Reporting and analysis are kept in clearly separated sections of every article
What VNN Does Not Do
- ✗ We do not publish opinion pieces or editorials
- ✗ We do not conduct original interviews or first-hand investigations
- ✗ We do not aim to replace human journalism — VNN depends on the reporters whose work it cites
- ✗ We do not hide that VNN is automated; this page exists precisely to make the process auditable
The Technology
VNN is one application of Valyrian Tech's SERENDIPITY system — a framework for coordinating specialised AI agents around a shared goal. The pipeline is driven by a single orchestrator script that runs once a minute and advances whichever story is next in line: gathering and compiling fire on a per-country schedule, while ranking, research, writing, publishing and revision are picked up as soon as their inputs are ready.
Core components:
- • Gatherer and compiler agents that read local news in its original language
- • A ranking council of seven independent agents that vote via the Hivemind Protocol
- • Research, writer, publisher and revision agents that produce the final article
- • An orchestrator that locks one story end-to-end at a time, retries failed steps, and discards stories that cannot meet the quality bar
Our Philosophy
VNN is built on a few simple convictions:
- Understanding precedes action — Well-informed citizens make better decisions
- Local sources matter — A story written from inside a country reads differently than one filtered through a foreign wire
- Transparency builds trust — Hidden processes invite suspicion; documented ones invite scrutiny and improvement
- Automation has ethical dimensions — Strict sourcing, URL verification and a willingness to discard bad stories matter more than throughput
Frequently Asked Questions
Q: Is VNN written by AI?
A: Yes. Every article on this site is produced by an automated pipeline of AI agents. Humans designed the system, the editorial constraints and the source list, but no human writes or edits individual articles.
Q: How do you prevent hallucinated facts or fabricated links?
A: Writers are technically constrained to cite only URLs that appear in the research document, the validator rejects articles that violate this rule, and the publisher agent then opens every remaining URL in a live browser before publishing. Articles that fail are sent back for revision; if they cannot be fixed, they are discarded.
Q: Why focus on local news from many countries?
A: International coverage tends to converge on a small number of wire services. By reading what people in each country actually see on their own front pages — in their own language — VNN can surface stories and framings that rarely cross language barriers.
Q: Will VNN replace human journalists?
A: No. VNN does not produce original reporting; it synthesises and contextualises the work of human journalists, and links back to them. The pipeline cannot function without their reporting upstream.
Q: How can I trust an article written by AI?
A: The same way you'd evaluate any source: follow the links. Every factual claim points to its source. If a link is broken or doesn't support the claim, that's a real problem we want to know about — and one the system is designed to make visible rather than hide.