What happens when your AI-powered app confidently tells a customer something false? That gap between "the model answered" and "the answer was correct" is what Factagora's new API targets.
The service offers six endpoints designed to sit between your application and whatever large language model (LLM - the AI engine powering tools like ChatGPT or Claude) you're running. The core endpoint is the Fact Checker, which takes any text claim, checks it against multiple sources, and returns a structured verdict: TRUE, FALSE, or UNCERTAIN, along with a confidence score.
Six Endpoints, One Key
The full toolkit covers a range of verification needs:
- News Search: Real-time semantic news retrieval, ranked by source credibility
- Fact Checker: Word-by-word claim verification with confidence scores
- Evidence Finder: Retrieves both supporting and opposing evidence for a given claim
- Deep Research: Breaks down complex queries, gathers evidence, and returns structured reports
- Timeseries: Extracts time-ordered data points from URLs or uploaded files
- Causality Graph: Infers cause-and-effect relationships from documents
All six use a single API key and return structured JSON. Factagora reports under 200ms average response latency and a 99.9% uptime SLA. An OpenAPI specification is included, meaning it integrates with standard developer tooling without custom work.
Pricing runs on credits: 100 free credits on signup. Simple calls like news search cost 1 credit; heavier endpoints like deep research and causality graph cost 2-3 credits.
The main alternative for reducing AI hallucinations (when a model states false information with unearned confidence) is RAG, or retrieval-augmented generation - feeding relevant source documents to the model before it generates an answer. RAG works upstream, improving input quality. Factagora works downstream, checking outputs after the fact. The two approaches are not mutually exclusive.
The use case is clearest in high-stakes contexts: a legal research tool that cites case law that doesn't exist, a customer-facing chatbot that gives incorrect pricing, a financial assistant that states wrong figures. In those situations, a 2-credit fact-check call is considerably cheaper than the fallout. Factagora is positioning this as enterprise infrastructure, though the self-serve credit model and public documentation suggest they're open to individual developers as well.