A library card for research agents: alphaXiv opens DeepWiki for Research
The layer offers multi-turn retrieval, keyword and embedding search across millions of papers, blogs and researchers, and now plugs into agents through Claude Connectors and ChatGPT plugins.
The known weakness of research agents is the bibliography: however strong the model, without access to the literature it reinvents the wheel. alphaXiv, the discussion and discovery platform built on arXiv, has opened a direct answer: DeepWiki for Research. Its slogan sums up the claim: let your research agents stand on the shoulders of giants.
The package offers three tool layers over millions of papers, blogs and researcher profiles: fast multi-turn retrieval (the agent deepens its search based on what it finds), classic keyword search, and semantic embedding search. The real news is distribution: the layer now connects directly through Claude Connectors and ChatGPT plugins, and is open to any compatible agent as an MCP server. A literature review shrinks to a single tool call inside the agent.
The demo example fits the day's spirit too: summarizing Google DeepMind's "AI co-mathematician" paper, the system describes it as an attempt to build Claude Code for math research; a persistent workbench instead of a chat window, harness thinking applied to mathematics. A small but clear sign that research is drifting toward the agent patterns of software.
Our own footnote: we wired this layer into our agent pipeline; verification and context sweeps on arXiv-sourced stories will now run through these tools as well. The winners of the agent era, it seems, will be those connected to the best library as much as those holding the best model.
