Pay only for the tokens you use

Reading, browsing, faceted search, contributing, and discussion are free, forever. The AI tools — natural-language semantic search and multi-paper synthesis — call a large language model, and every model call costs us money per token.

So AI features are metered. You pre-load credits, and each request deducts the exact token cost. We pass through the model's published per-token price and add a small 15% margin that covers our servers, search infrastructure, and the embeddings that power retrieval. No subscription, no minimum — when your balance runs out, the free features keep working.