FAQai

FAQai is a RAG dataset generator that turns PDFs, DOCX, TXT, Markdown files, or website content into grounded chunks, question-answer pairs, and test datasets for retrieval-augmented generation systems. It is built for developers and teams building RAG pipelines who need evaluation and adversarial data rather than just a model, since FAQai frames most RAG failures as a data problem rather than a model problem.

A user uploads a document, up to 25 MB depending on plan, or points FAQai at a public web page or a same-domain crawl of up to 500 pages. FAQai then chunks the document and generates canonical question-answer pairs, query variants, an evaluation dataset, adversarial test questions, quality insights, and a tailored RAG config with a system prompt and code snippets, all from that one document.

FAQai produces five dataset types: RAG chunks, canonical Q&A, query variants, evaluation datasets, and adversarial datasets, plus a separate quality-insights analysis across 9 categories. Scanned PDFs are processed with OCR on paid plans, and a RAG Playground lets a user chat with the processed document to test retrieval quality before exporting. Results export to 16 formats and 8 vector database integrations, including Pinecone, Qdrant, and pgvector, and are also reachable through a REST API, Python and TypeScript SDKs, and an MCP server.

FAQai offers a free plan with 10 pages per month, JSON export only, and no credit card required, plus three paid tiers: Basic at $9/month, Starter at $19/month, and Professional at $39/month, each raising page quotas, file-size limits, API calls, and OCR allowance. Documents are encrypted at rest with AES-256 and in transit with TLS 1.3, are never used to train AI models, and FAQai is GDPR compliant with data deletion available at any time.

Use cases

• Turning technical docs or knowledge bases into RAG training data
• Surfacing edge cases and misleading queries that expose hallucinations
• Letting AI tools like Claude and Cursor generate datasets via MCP
• Generating a RAG dataset straight from a website with no file