AI Research Agent

Literature review,
grounded in the
papers it actually read.

Confluex searches Semantic Scholar, arXiv, and PubMed, ranks what matters with composite scoring, and writes you a survey you can defend — every [N] in the draft is traceable to an abstract in the local cache.

225M+
Papers
indexed
3–5
LLM calls
per review
$0.20
Average
cost
02Walkthrough

From a topic to a defended survey,
in two minutes.

Watch a single run end-to-end — keyword expansion, four parallel searches, the cosine-filter cut, composite ranking, the LLM call that writes the actual prose, and the quality-judge handshake that decides whether to ship or re-synthesise.

Confluex · Deep Research MaxDemo recording
Demo recording · 02:14Deep Research MaxLive cache · 225M papers
03Sources

Pulled from the canon.
Not the open web.

Every Confluex run starts with a parallel search across the three databases serious researchers actually trust, plus a local SPECTER2 / MiniLM vector index for sub-second pre-filtering. No SEO blogs. No press releases. No hallucinated journals.

primary api
Semantic Scholar
200M+papers, with citation graph
preprints
arXiv
2.4M+abstracts mirrored locally
biomedical
PubMed
36M+via E-utilities
specter2 · minilm
Local index
Chromavector store, sub-second filter
05Sample output

An excerpt,
not a mood board.

Every review exports as .docx, .tex, and a compiled .pdf. Citations are numbered, traceable, and survive an advisor's red pen — because each one comes from an abstract the system actually read.

Review · 2026-05-13Model · gpt-4o · 20 papersQuality · 0.87 avg

2.  Thematic Analysis of Retrieval-Augmented Approaches

The first wave of retrieval-augmented generation systems treated retrieval as a one-shot upstream step, concatenating top-k passages into a fixed context window before generation[1]. Subsequent work relaxes this rigidity: iterative retrievers re-query the corpus after each generation step[3], while self-reflective variants emit explicit retrieval tokens to decide when additional context is warranted[7].

For long-form scientific writing the trade-off is sharper. Static retrieval is fast and reproducible but tends to over-represent the first few sub-topics; iterative schemes adapt but compound latency and citation drift[4, 9]. A growing body of evidence suggests that the highest-quality surveys come not from richer prompting but from better candidate sets: composite scoring functions over relevance, citation count, and recency consistently outperform pure semantic ranking on coverage and novelty[12, 14].

… continues for five more sections — methodology comparison, key findings, gaps, and conclusion.

06Pricing

A plan for every
stage of research.

From a single coursework essay to a shared lab library — pick a tier, upgrade or downgrade whenever the work shifts. Top-up packs and seat add-ons live on the full pricing page.

Free

For curious afternoons.

$0/mo
  • check1–2 active projects
  • checkUp to 5 PDF uploads
  • check1 short report / month
Start free

Student

For coursework & first papers.

$8/mo
  • check50 PDF uploads
  • check10 Deep Search reports / mo
  • checkBasic Writer outputs
Get Student

Lab / Team

For groups thinking together.

$22/mo · per seat
  • checkShared projects & libraries
  • checkPooled Deep Search credits
  • checkAdmin dashboard
Set up the lab
07Get started

Stop reading papers
you didn't need
to read.

Confluex · AI20K-026 · Built with LangGraph + FastAPI + Next.js