Platform
Every primitive you need for RAG
From raw documents to grounded answers — a complete toolkit, one API, zero infrastructure to manage.
# Ask a question, grounded in your data
POST /v1/playground/chat
{
"knowledge_base_id": "kb_3f82a1",
"question": "What is the refund policy?",
"model": "gpt-4-turbo"
}
# → answer, sources, tokens, costIngestion & indexing
Turn any source into a searchable index.
Multi-format parsing
PDF, Markdown, HTML, Word, and Notion exports — parsed, chunked, and embedded automatically.
Semantic chunking
Boundary detection preserves meaning across paragraphs and sections, no manual tuning required.
Vector indexing
Production-grade vector storage with hybrid BM25 plus dense retrieval out of the box.
RAGDCl
Find the right passages, every time.
Semantic search
Query embeddings match documents by meaning, not just keyword overlap.
Reranking
Second-pass reranking models re-score candidates for final precision before generation.
Citations & sources
Every answer returns the source passages it drew from, with scores and metadata.
Generation
Grounded answers with citations and control.
Any model
Swap GPT-4o, Claude, Gemini, or open-source models with a single parameter.
Streaming responses
Token-level streaming so users see answers as they generate — sub-second first token.
Structured outputs
Constrained generation returns JSON that matches your schema, ready for downstream use.