Cito

Cito is a hybrid search engine designed to provide efficient and unrestricted access to a vast academic corpus. It is specifically built for AI agents and researchers who require deep literature exploration without encountering the rate limitations often imposed by traditional academic APIs. The engine indexes over 236 million papers, incorporating both keyword and dense vector (SPECTER2) retrieval methods, fused with RRF and reranked by a cross-encoder for enhanced accuracy. The problem Cito addresses is the significant bottleneck created by restrictive API rate limits on academic search engines. For AI agents performing extensive literature reviews, these limitations can cause research runs to stall, hindering progress. Existing platforms like Google Scholar lack APIs altogether, while others, such as Semantic Scholar, impose very low request limits (e.g., 1 request per second), which are insufficient for automated, in-depth research tasks. Cito was developed to overcome these limitations by indexing the corpus independently and offering more generous access. Cito's core functionality revolves around its hybrid search approach. It combines traditional keyword-based search (BM25) with advanced semantic search using SPECTER2 dense vectors. These results are then fused using Reciprocal Rank Fusion (RRF) and further refined by a cross-encoder model. This multi-faceted approach ensures that search results are both comprehensive and highly relevant, capturing nuances that single-method searches might miss. The engine is served from a single CPU box, providing fast response times, typically under half a second. Key features include a free web search interface that requires no signup, making it immediately accessible to anyone. For developers and AI agents, Cito provides a plain JSON API with significantly more generous rate limits compared to upstream services. Free API keys offer 100 requests per minute, a substantial improvement over the typical 1 request per second. This allows for more fluid and extensive automated research. Another critical feature is the native MCP (Machine-Readable Content Protocol) endpoint. This integration allows AI agents, such as Claude Code, to directly search the literature without complex setup or intermediary steps. This seamless integration is designed to facilitate deep literature research directly within the agent's workflow, eliminating common friction points. Cito is explicitly designed as a retrieval engine, not a chatbot. Its primary output consists of ranked lists of academic papers, complete with abstracts, citation counts, open-access PDF links, and DOIs. The interpretation and synthesis of this information are left to the user's agent or their own analytical process. This focus on pure retrieval ensures that the engine is optimized for speed and accuracy in finding relevant research papers. Cito operates by indexing a massive corpus of academic papers, including keyword data and dense vector embeddings. It employs a sophisticated retrieval pipeline that fuses multiple search strategies to deliver highly relevant results. The system is optimized for speed and efficiency, capable of returning results in under half a second, even from a single CPU instance, addressing the performance concerns of automated research tasks. The primary benefit for users is the ability to conduct deep, uninterrupted literature research. By removing API rate limit frustrations, Cito empowers researchers and AI agents to explore academic databases more thoroughly and efficiently. The generous API limits and direct agent integrations streamline the research process, saving time and enabling more comprehensive analysis. Concrete use cases for Cito include AI agents performing extensive literature reviews for academic papers, researchers quickly gathering background information on a topic, or developers integrating academic search capabilities into their applications. For instance, an AI agent like Claude Code can use the MCP endpoint to initiate a deep dive into a research area, retrieving relevant papers and their metadata without hitting rate limits. Cito is targeted at developers building research agents, AI researchers, and academics who frequently engage in literature reviews. It offers a free web search and a JSON API with generous limits. While specific tech stack details beyond the use of SPECTER2 dense vectors, RRF, and cross-encoders are not explicitly detailed, the mention of serving from a single CPU box suggests an efficient, optimized deployment. Pricing is free. In summary, Cito provides a powerful, unrestricted hybrid academic search engine that overcomes the limitations of existing APIs, enabling efficient and deep literature research for both human users and AI agents.