PaperGuru
Sign inDownload
Compare

The AI paper studio that actually finishes the paper

“Autonomous AI Scientists” make great livestreams. Deep Research agents make great briefings. Neither gives you a submittable paper. PaperGuru runs the full pipeline — multi-source literature search, AI writing grounded in real citations, a live LaTeX editor, and 100+ venue templates — with a published track record at ICML, FSE, AEI and TOSEM.

Start freeSee the 10-way comparison
AEI · TOSEM
Journal papers accepted / in major revision
ICML · FSE
Top-tier conference papers accepted
< 1%
Verified citation hallucination rate
100+
Venue templates ready to submit
Track record

Accepted and in-review at the venues that matter

Papers drafted with PaperGuru are already through peer review at top-tier venues across software engineering, machine learning, and engineering informatics. This is what separates a writing tool from a demo.

AEI
Advanced Engineering Informatics
Journal paper accepted
Elsevier flagship journal in engineering informatics
TOSEM
ACM Transactions on Software Engineering and Methodology
Major revision
CCF-A · one of software engineering's top three journals
ICML
International Conference on Machine Learning
Accepted
CCF-A · one of ML's top-three conferences alongside NeurIPS and ICLR
FSE
ACM International Conference on the Foundations of Software Engineering
Accepted
CCF-A · flagship software engineering conference

Venues and statuses verifiable on request. We do not name authors or paper titles on a public page; we respect double-blind review and author confidentiality.

Full comparison

PaperGuru vs 9 auto-research products and open-source projects

Every row is a concrete capability a researcher actually needs. Every claim below is sourced from the vendor’s own documentation, paper, or public benchmark.

Capability
PaperGuru
This is us
Sakana AI Scientist v2
Autonomous agent
HKUDS AI-Researcher
HK open source
Analemma FARS
Shanghai demo
Google AI co-scientist
Research preview
FutureHouse Platform
Lit-review agents
OpenAI Deep Research
ChatGPT feature
SciSpace
Closest suite
Stanford STORM
OSS (wiki-style)
gpt-researcher
OSS (web reports)
Searches real peer-reviewed literature
OpenAlex / Semantic Scholar / arXiv / PubMed / DBLP, not random web pages
YesPartialPartialPartialPartialYesNoYesNoNo
AI drafts a full paper (not just answers)
YesYesYesYesNoNoNoPartialPartialNo
Live LaTeX editor with real-time PDF preview
You can actually change the draft and see it compile
YesNoNoNoNoNoNoNoNoNo
100+ venue templates (NeurIPS, ICML, CVPR, ACL, Nature, Cell, IEEE, ACM …)
Yes1 (ICLR-style)1 (ICLR-style)NoNoNoNoPartialNoNo
Cross-discipline (CS · biomed · physics · humanities · economics)
YesML onlyML only (5 sub-fields)AI research onlybiomed onlybiomed / chemgeneral Q&AYesgeneralgeneral
Choose Claude · GPT · Gemini · Grok · community models
YesPartialPartialNono (Gemini only)Nono (GPT only)PartialPartialYes
Desktop app for Mac & Windows
YesNoNoNoNoNoNoNoNoself-host
Cloud collaboration (real-time co-editing)
YesNoNoNoclosed betaYesNoYesNoNo
Human-in-the-loop editing during the run
You can steer, edit, regenerate — not just accept what the agent emits
YesNoNoNoNoPartialPartialPartialPartialNo
Anyone can sign up today
YesDIY (GPU + CLI)DIY (Docker + GPU)no (observe only)no (Trusted Tester)YesYesYesyes (demo)self-host
Transparent public pricing
Yes$15–20 / run APInot disclosednot disclosednot availablefree tier + TBDYesYesfreefree (self-host)
Published papers at peer-reviewed venues
AEI · TOSEM · ICML · FSE1 workshop (withdrawn)its own benchmarknone (arXiv only)wet-lab case studiesPaperQA2 benchmarkNoNoNoNo

Sources: repository READMEs, arXiv preprints, vendor blogs and pricing pages as of April 2026. See the sources list below for every citation.

Hallucination

Near-zero citation hallucination, by design

The single biggest reason researchers distrust AI writing is fabricated citations. PaperGuru is built around four defensive layers so the problem is engineered out rather than papered over.

01

Grounded in real literature

Every citation PaperGuru inserts is resolved against OpenAlex, Semantic Scholar, arXiv, DBLP, or PubMed before it is written into your draft. If a paper cannot be found in the live scholarly graph, it never enters your bibliography.

02

DOI + BibTeX round-trip

Citations are stored with their canonical DOI or arXiv ID and re-validated at export time. Broken DOIs, invented authors, and phantom journals are caught before the PDF is compiled.

03

Full-text grounding, not abstract guessing

When you ask the AI to support a claim, it reads the open-access full text and grounds the generated sentence to a concrete span. No “the paper argues…” hallucinations when the paper says no such thing.

04

Human-auditable diff

Every AI edit is surfaced as a diff you can accept, reject, or rewrite. Nothing lands in your paper without you seeing it — the opposite of “one-shot auto-submit” agents.

> 99%
AI-inserted citations verified as existing, peer-reviewed sources in our internal evaluation (n = 1,000+ citations, sampled across disciplines)
Philosophy

Five ways PaperGuru is not an “AI Scientist”

01

We don't try to replace researchers — we make you 10× faster

“Fully autonomous AI Scientists” sound impressive but today they land at workshop rejections at $1,000 per paper, with hallucinated citations and frozen scope. PaperGuru treats you as the author. The AI searches, drafts, cites, and compiles; you decide what is actually true.

02

A real editor, not a one-shot agent

Sakana, HKUDS AI-Researcher, and FARS all emit one PDF and then stop. If you spot a mistake, you re-run for $15–$1,000. PaperGuru gives you a LaTeX editor with live PDF preview, tracked AI diffs, and the ability to regenerate any paragraph or figure on demand.

03

Every discipline, not just ML

AI Scientists target arXiv CS listings. PaperGuru supports NeurIPS, ICML, ICLR, CVPR, ACL, EMNLP, NAACL, Nature, Cell, NEJM, The Lancet, IEEE TPAMI, ACM CHI, PRL, JACS, AER, APSR, PMLA and 100+ more — with the actual venue LaTeX class files preloaded.

04

Near-zero citation hallucination, by design

We route every AI-inserted citation through the scholarly graph with DOI verification and full-text grounding. Deep Research tools like OpenAI, Perplexity, and Gemini cite blogs and press releases; we cite papers that actually exist.

05

Transparent pricing you can read on one line

$199.99 / year for Pro. $500 / year for Plus. $2,000 / year for research labs. No “contact us” gate, no $104k per batch like FARS, no invite-only Trusted Tester programs like Google AI co-scientist.

Head-to-head

What each competitor actually does today

PaperGuru vs Sakana AI Scientist v2

Sakana’s AI Scientist is a fully autonomous agent: given a topic, it ideates, runs ML experiments, emits one ICLR-style LaTeX PDF for roughly $15–$20 in API cost. It is an OSS research artifact, not a product — no editor, no venue switching, no literature search UI. One v2 paper passed peer review at the ICLR 2025 ICBINB workshop; all three submissions were withdrawn per a pre-agreed protocol, and Sakana’s own team confirmed none met the main-conference bar. PaperGuru is the opposite design: a human-in-the-loop editor backed by the full scholarly graph, with accepted papers at ICML, FSE, AEI and TOSEM.

PaperGuru vs HKUDS AI-Researcher

AI-Researcher (HKU Data Intelligence Lab, 5.2k GitHub stars, NeurIPS 2025 Spotlight) produces full-length ICLR-style papers plus runnable code for five pre-configured ML sub-fields: VQ-VAE, graph recommendation, diffusion / flow matching, generic GNNs, and reasoning. Outside these sub-fields you must hand-build the benchmark corpus. Onboarding is rough — the open repo has 56 open issues including broken Docker images, Windows path bugs, and committed API keys. PaperGuru supports every discipline out of the box and you can sign in and write immediately; no GPU, no Docker, no per-sub-field corpus engineering.

PaperGuru vs Analemma FARS

FARS (Shanghai-based Analemma) livestreamed 100 short AI-research papers in 228 hours on a 160-GPU cluster at roughly $1,000 per paper and 114 M tokens per paper. Its papers score 5.05 on Stanford’s Agentic Reviewer — above average ICLR submissions (4.21) but below the acceptance threshold (5.39). There is no code, no paper, no API; you can watch but not use it. PaperGuru is available today, costs $199.99–$2,000 per year, and has a peer-reviewed track record in published venues rather than a reviewer-model score.

PaperGuru vs Google AI co-scientist

Google’s AI co-scientist is a six-agent Gemini-powered system that generates biomedical research hypotheses, validated in wet-lab case studies (AML drug repurposing, liver fibrosis, antimicrobial resistance). It is not a product — access is restricted to an invite-only Trusted Tester program for research organisations, no API, no published pricing, biomed-only. It stops at hypothesis; there is no paper, no LaTeX, no template. PaperGuru complements rather than competes here: AI co-scientist proposes the study, PaperGuru writes it up into a submittable manuscript.

PaperGuru vs FutureHouse / PaperQA2

FutureHouse’s platform (Crow, Falcon, Owl, Phoenix) is the strongest open literature-QA service on the market; its open-source PaperQA2 library (8.4k stars) holds SOTA on the RAG-QA Arena science benchmark and measurably beats PhD researchers on literature search precision. It answers questions; it does not write papers, format to a venue, or compile LaTeX. The natural workflow is PaperQA2 surfaces the literature → PaperGuru writes, cites and submits. We treat FutureHouse as an upstream partner, not a competitor.

PaperGuru vs OpenAI / Perplexity / Gemini Deep Research

Deep Research tools produce multi-thousand-word markdown briefings with inline citations, in 5–30 minutes, for general web topics. They are excellent for analyst-style synthesis but two things rule them out for academic submission: they cite the open web (blogs, press releases, Reddit) rather than peer-reviewed papers, and they output markdown or Google Docs, never venue-formatted LaTeX. PaperGuru grounds every citation in the scholarly graph and emits submission-ready LaTeX with the correct class files preloaded.

PaperGuru vs SciSpace

SciSpace is our closest suite-shaped competitor — it has paper search, PDF chat, an AI writer, and the legacy Typeset template library. The gap is that SciSpace’s writer is a paragraph-level assistant, not an end-to-end drafting pipeline, and its template library is a starting point rather than an integrated editor experience. PaperGuru’s LaTeX editor, live PDF preview, AI diffs, and peer-reviewed publication record at ICML / FSE / AEI / TOSEM is the concrete differentiator.

PaperGuru vs Stanford STORM and gpt-researcher

STORM (28 k stars) and gpt-researcher (27 k stars) are excellent open-source long-form synthesis projects. They produce Wikipedia-style articles and markdown reports respectively. Stanford’s own README is explicit: STORM output “cannot be published without a significant number of edits” and is meant for the pre-writing stage. PaperGuru picks up exactly where these tools hand off: take the synthesis, turn it into a structured academic paper, format it for a specific venue, cite it from peer-reviewed sources, and compile to PDF.

Honesty

What PaperGuru is not

  • Not a fully autonomous scientist. We do not run wet-lab experiments, we do not train new models for you, and we will not submit a paper under your name while you sleep. That is a design choice, not a limitation.
  • Not a replacement for advisors or peer review. We make drafting, formatting, and citation faster; scientific judgement and responsibility stay with the human author.
  • Not a guarantee of acceptance. Publication depends on the novelty and rigor of your work. PaperGuru’s role is to make sure mechanical friction — LaTeX errors, broken references, venue formatting — never costs you a decision.
FAQ

Frequently asked

How is PaperGuru different from Sakana’s AI Scientist or HKUDS AI-Researcher?
Those are autonomous agents optimized for a single ICLR-style output PDF in machine-learning sub-fields. PaperGuru is a full academic writing studio: you search literature with a real UI, draft with AI inside a live LaTeX editor, preview the compiled PDF in real time, and pick from 100+ venue templates across every discipline. You stay in the author seat; the AI does the fetching, drafting, and formatting.
Is PaperGuru a “Deep Research” tool like OpenAI or Perplexity?
No. Deep Research products produce markdown reports citing the open web — useful for briefings, not submittable as papers. PaperGuru cites peer-reviewed literature with verified DOIs and produces submission-ready LaTeX with the correct class files for your target venue.
What does “near-zero hallucination” actually mean?
Three layers: (1) every citation resolves against OpenAlex, Semantic Scholar, arXiv, DBLP, or PubMed before it lands in your bibliography; (2) DOIs and BibTeX are re-validated at export time, so broken references never reach the compiled PDF; (3) when the AI supports a claim, it grounds the sentence against the open-access full text rather than inventing plausible-sounding prose. Internal evaluation on 1,000+ AI-inserted citations puts verified existence above 99%.
Has work drafted with PaperGuru been published?
Yes — at the time of writing, authors using PaperGuru have papers accepted or in major revision at ICML, FSE, AEI, and TOSEM (ACM TOSEM), among others. We obviously do not share author names without consent, but the list of venues is verifiable on request from our enterprise team.
Can I use Claude, GPT-5, Gemini, or Grok?
Yes. Free-tier users get a curated set of community models (Llama, Gemini flash, DeepSeek, etc.). Pro and above unlock Claude, GPT, Gemini Pro, and Grok. Pro also supports Bring-Your-Own-API-Key if you prefer to route spend through your own vendor account.
Is PaperGuru desktop or cloud?
Both. Native apps for macOS (Apple Silicon and Intel) and Windows x64, plus a cloud workspace for collaboration. The desktop app gives you offline-friendly LaTeX compilation with local file persistence; the cloud workspace gives your co-authors real-time co-editing.
Do you replace my advisor / my writing?
No. That is the explicit design line. We think the “one-shot AI paper” framing is dishonest about what current models can do. PaperGuru drafts sections, fetches citations, formats LaTeX, and flags weak arguments — you make the scientific judgement and put your name on the paper.
Sources

Everything on this page is linkable

Stop re-running agents. Start writing your paper.

Download PaperGuru, import a project, and submit to your target venue with the template already wired up. Free tier, no credit card, 500 bonus credits on signup.

Start free