How it works
Understanding EarlyForge's two-phase process — continuous preparation and on-demand generation.
How it works
EarlyForge runs on a two-phase process. Phase 1 runs continuously in the background (24/7), building and maintaining a rich knowledge base. Phase 2 generates a complete article on demand in about 75 seconds.
Process overview
flowchart LR
subgraph Phase1["Phase 1 - Continuous Preparation 24/7"]
TD[Trend Detection] --> DA[Source Discovery]
DA --> SC[Content Collection]
SC --> IDX[Knowledge Indexing]
end
subgraph Phase2["Phase 2 - On-demand Generation ~75s"]
BR[Brief or Trend] --> RECH[Intelligent Retrieval]
RECH --> GEN[AI Generation]
GEN --> PUB[Publish to WordPress / Ghost / etc.]
end
IDX -.-> RECHPhase 1 — Continuous preparation
This phase runs 24/7 in the background. Its job is to build and maintain a rich, up-to-date knowledge base that generation can draw from instantly.
Trend detection
Hundreds of sources are analyzed nonstop around the world — search, social, news, academic, finance, government, science, and reference feeds. Every signal is normalized, scored, and categorized across 28 categories with global coverage.
The scoring engine evaluates trends on five factors: velocity, diversity, recency, cohesion, and SEO potential.
Source discovery
EarlyForge takes detected trends and finds the best articles to use as sources. It queries multiple search engines and feeds, then applies AI relevance scoring to keep only the highest-quality sources.
Sources are chosen intelligently based on the topic's category, region, and editorial intent.
Content collection
EarlyForge collects content from the identified sources, respecting per-domain rate limits to stay a good citizen of the web. Clean article text is extracted automatically from any website.
Knowledge indexing
Collected content is indexed into a searchable knowledge base built for two kinds of lookup:
- Semantic search — find passages that match by meaning, not just keywords
- Entity search — understand the relationships between people, places, and concepts
The result is a rich, queryable knowledge base that powers near-instant article generation.
Phase 2 — On-demand generation
When you create a brief or select a trend, this phase generates a complete article in about 75 seconds.
Intelligent retrieval
EarlyForge runs a hybrid search that combines:
- Keyword matching
- Semantic search (by meaning)
- Entity search (relationships between concepts)
- Relevance re-ranking
This pulls the most relevant and recent passages from the knowledge base.
AI generation
The AI takes the retrieved context and writes an original article with:
- Inline citations (
[1],[2], ...) linking back to the source articles - SEO-ready structure — headings, meta description, and keywords
- Tone, length, and style configurable per site
- Real-time streaming so you can watch the article being written
Publish
The finished article is published to the destination you connected:
- WordPress (REST API)
- Ghost (Admin API)
- Medium
- Next.js Direct (direct insert with ISR revalidation for your own Next.js site)
- Webhooks (custom integration, signed with HMAC-SHA256)
You can publish to several destinations at once, manually or on a schedule.
Why two phases? By continuously building the knowledge base in Phase 1, generation in Phase 2 can be nearly instant (~75s). Without pre-built knowledge, every article would require real-time source collection, adding minutes of latency.
Root events — cross-lingual intelligence
EarlyForge goes beyond simple trend tracking. It clusters related trends across languages and countries into root events — unified representations of global news stories.
For example, if "Bitcoin crashes" trends in English, "Bitcoin s'effondre" in French, and "Bitcoin stuerzt ab" in German, EarlyForge recognizes these as the same underlying event and tracks its geographic propagation in real time.
This is powered by:
- Fast semantic search to find similar trends
- Entity disambiguation — "Apple" the company versus "apple" the fruit
- Adaptive time windows — tuned per category: short for fast-moving markets, much longer for science
To learn the vocabulary used throughout the product, continue to Key concepts.