Most retrieval‑augmented systems treat literature search as a single retrieval step followed by summarization, while real research continuously reshapes the understanding of the question, spawns new sub‑questions, and weighs evidence. Optimist Deep Research is an open‑source agent that models literature research as an iterative evidence‑acquisition process over an explicit, mutable research state.
The main question is decomposed into a tree of sub‑questions; each sub‑question has search intents that are translated into source‑specific queries for arXiv, OpenAlex, Crossref, PubMed, Semantic Scholar and the web. Results are merged into a canonical, provenance‑preserving document registry via a conservative four‑tier deduplication hierarchy. The registry is expanded by citation snowballing until a saturation criterion is met, then screened, ranked, and allocated to sub‑questions round‑robin. Documents are read in parallel into a structured evidence store that records claims with support, location, type and strength.
A critic reviews the state along eight dimensions and returns research actions rather than a binary verdict: it can add, split, rephrase, deprioritize or close sub‑questions, making the research plan both an input and an output. The final report is written section by section from the evidence store, and every citation resolves to a retrieved document. We describe the design, the deterministic machinery that keeps the language‑model pipeline auditable, and a case study that read 28 of 237 retrieved documents in about three and a half minutes. The case also exposed several weaknesses.
Review: The system brings research closer to human practice through explicit state management and iterative retrieval, yet it still needs finer‑grained evidence evaluation and better scalability for large document sets.