VeritasFeed is a self-hosted news and open-source intelligence platform. It ingests from news APIs, social networks, RSS, the open web and chat, scores every source and article for trust and political bias, and fuses it with a real cyber threat-intelligence layer: IOCs, CVEs, MITRE ATT&CK and breach tracking.
Self-host or SinonTech-hosted. Bias-aware. Air-gap capable.
Native VeritasFeed clients. The macOS and Linux desktop apps are available now in alpha; Windows, iOS and Android are on the way. Contact us to join the alpha.
Source Families
News APIs, social networks, RSS, web scraping, chat, PDF, email, paste sites, and threat-intel feeds.
API Modules
REST API across ingestion, NLP, scoring, search, alerting, feeds, notifications and the threat-intel subsystem.
Schema Migrations
A mature data model behind trust scoring, story clustering, threat intel and multi-tenant isolation.
Database Isolation
Tenant isolation enforced by PostgreSQL row-level security, not just application filtering.
Deployment Modes
Docker, Kubernetes, and fully air-gapped, with federated SSO built in.
VeritasFeed pulls from where the story actually breaks (not just a wire API) and treats every source as something to be scored, not trusted by default.
Five news APIs (GDELT, Guardian, Mediastack, NewsAPI, New York Times) and four social networks (Bluesky, Mastodon, Reddit, X), each a real collector.
RSS polling, full web scraping with readability extraction, plus Discord and Telegram channels, PDF documents, email and paste sites.
Every source and article carries a trust score and a political-bias position, with source-reliability tracking over time, so you see the lean, not just the headline.
The same wire story republished across fifty outlets is detected and grouped, so a single event does not read as fifty independent confirmations.
Articles cluster into stories and narratives, with trending detection, so you follow how a story develops rather than scrolling a flat timeline.
Integration with Google Fact Check surfaces claims that have already been assessed, alongside the articles making them.
The language processing runs on real models, and beneath the news sits a genuine cyber threat-intelligence subsystem, useful whether you are a newsroom, an analyst team or a SOC watching the open source.
Categorisation, sentiment (VADER and a DistilBERT model), named-entity recognition (spaCy), summarisation and M2M-100 translation, all running via ONNX on your own hardware, with graceful fallback when a model is not present.
Elasticsearch-backed full-text search with saved searches and alerts, so a topic, entity or keyword pings you the moment it appears.
IOC extraction and feeds, CVE/NVD synchronisation, MITRE ATT&CK tagging, threat-actor tracking, YARA and Sigma matching, and breach attribution.
MISP and TAXII 2 export push structured intelligence into your existing tooling; shareable feeds and reports move it to people.
Route alerts to email, webhooks, Slack, a SOAR platform or web push, wherever your team watches.
SAML, OIDC, OAuth, SCIM and WebAuthn/MFA, with a hash-chained audit log: enterprise identity without a third-party dependency.
| News sources | GDELT, The Guardian, Mediastack, NewsAPI, New York Times |
|---|---|
| Social & chat | Bluesky, Mastodon, Reddit, X, Discord, Telegram |
| Other ingestion | RSS, web scraping (readability), PDF documents, email/IMAP, paste sites, IOC threat-intel feeds |
| NLP | Categorisation, sentiment (VADER + DistilBERT), NER (spaCy), summarisation (LexRank/TextRank), M2M-100 translation — via ONNX Runtime, on-prem |
| Intelligence | Per-source and per-article trust + political-bias scoring, source-reliability tracking, wire-story detection, story clustering and trends |
| Threat intel | IOC feeds, CVE/NVD sync, MITRE ATT&CK tagging, threat-actor tracking, YARA/Sigma matching, breach attribution, MISP + TAXII 2 export |
| Search | Elasticsearch 8 full-text search, saved searches, alerting |
| Notifications | Email, webhook, Slack, SOAR, web push, in-app |
| Identity | SAML, OIDC, OAuth (Google/Microsoft), SCIM, WebAuthn/MFA |
| Databases | PostgreSQL 16 (66 migrations), Redis 7, Elasticsearch 8 |
| API | 84 route modules |
| Multi-tenancy | Organisation → tenant → sub-tenant with PostgreSQL row-level security (database-enforced isolation) |
| Backend stack | Python 3.12, FastAPI, SQLAlchemy 2 async, Celery workers, ONNX Runtime |
| Frontend stack | React 18 with TypeScript, Vite, Redux Toolkit, TanStack Query, Tailwind, i18n (English, Spanish, French) |
| Deployment | Docker Compose, Kubernetes (Helm), cloud VMs, virtual environments, air-gapped |
A self-hosted news and open-source intelligence platform. It ingests from nine source families (news APIs, social networks, RSS, the open web, chat, documents, email, paste sites and threat-intel feeds), runs a real NLP pipeline over the content, scores every source and article for trust and political bias, and fuses it with a cyber threat-intelligence layer. It runs on PostgreSQL, Redis and Elasticsearch on your own infrastructure.
Every source carries a reliability profile tracked over time, and every article is scored for trust and placed on a political-bias spectrum. Because the same wire story republished across many outlets is detected and grouped, one event does not read as many independent confirmations. The point is to show you the lean and the provenance, not to tell you what to think.
Real. Sentiment uses VADER and a DistilBERT model, entity recognition uses spaCy, translation uses Meta's M2M-100, and summarisation uses extractive algorithms, all running through ONNX Runtime on your own hardware. When a model artifact is not present, the pipeline falls back to lighter methods rather than failing. A few advanced features (abstractive narratives, daily briefings) are still deterministic placeholders, and we do not describe those as finished AI.
Because open-source intelligence and threat intelligence overlap. VeritasFeed extracts indicators of compromise, synchronises CVEs, tags content against MITRE ATT&CK, tracks threat actors, and matches YARA and Sigma rules, then exports via MISP and TAXII 2. A SOC can watch the open source for early signals; a newsroom can ignore that layer entirely. Both run on the same platform.
By PostgreSQL row-level security: the database itself enforces that a tenant only ever sees its own rows, set per session. That is stronger than application-layer filtering, because isolation does not depend on every query being written correctly. The model is organisation → tenant → sub-tenant, suited to an agency or MSSP running intelligence for many clients.
Closed beta. It is a mature, feature-broad platform: nine source families, a real ONNX NLP pipeline, 84 API modules and 66 database migrations, covered by 353 backend tests. Request access and we will walk you through the sources relevant to you, the NLP models, and deployment before you commit anything.
VeritasFeed is in closed beta. Request access for deployment documentation, an architecture walkthrough, and a look at the sources and models most relevant to you.
Talk to an engineer about running your intelligence platform on infrastructure you own, or let us host it for you and still own every byte.