Multi-corpus routing for research agents

This points to the next product step for deep research agents, moving from generic web search to query routing across specialized data sources. Today Manus can already orchestrate long, multi step research through Parallel, but for medical, legal, and financial questions it still often works by searching the open web, opening pages one by one, and reading them. The real upgrade is direct access to the right corpus first, then reasoning on top of it.

• The practical issue is not just speed. It is source quality. In the interview, the PM describes Manus finding medical studies through ordinary web search, while wanting direct journal access instead. That matters because open web results are increasingly crowded with SEO pages and AI rewritten summaries rather than primary material.
  1 sacra
• There is already evidence that domain specific corpora create real product separation. Exa is described as offering domain streams such as financial filings and analyst reports. In medicine, OpenEvidence built traction by pairing an AI interface with licensed access to journals like NEJM and JAMA, turning better retrieval into a differentiated product.
  1 sacra 4 sacra
• This also helps explain where value sits between Manus and Parallel. Parallel supplies the research backbone, but Manus controls the orchestration layer that decides which tools to call for which task. If domain routing becomes core, the winning agent will look less like one search box and more like a dispatcher choosing the right database, connector, and workflow for each question.
  1 sacra 3 sacra

From here, deep research products are likely to evolve into multi corpus systems, with one path for open web discovery and others for journals, filings, case law, internal documents, and live data. As that happens, the strongest products will be the ones that can automatically pick the right knowledge base for the job and fuse the results into one reliable answer.