Text chunk (RAG) x402 REST usage
Split text into overlapping chunks for RAG ingestion - by characters (default) or by exact LLM tokens. Returns the chunks plus offsets. Deterministic, no model needed.
0.001 USDC per call
per call