Research moves fast. resmon does, too.

resmon is a free desktop app. It asks arXiv, PubMed, OpenAlex, Crossref, and 21 more on a schedule or when you ask, keeps every paper it finds on your computer, and records what each source did on every run. This is its morning view, drawn from a real collection.

Download resmon Install guide Version 1.0.0 for macOS, Windows, and Linux. Free, under the MIT license.

073090180 days since collected
new Since Today
a watch’s papers
more papers nearby

The plate: all 1,751 papers on the map, placed by how much wording they share, worked out on the computer that holds them. Drawn from the engine's own map and Today, exported unedited from a collection built on November 11, 2026: three watches' papers from arXiv, and three searches' from the sources they asked. The papers and the day's runs are real. For the watches' earlier papers, the day each was collected, and so how bright it is drawn, is set from its publication date.

New Since Today 16

Diffusion models

12 new

Ran today at 1:19 PM in 1 s · 51 found · 1 of 1 sources answered

5

Bayesian inverse problems and epidemic-threshold uncertainty for a heterogeneous SIS reaction-diffusion model

Duc-Lam Duong arXiv D4

From the abstract We study Bayesian inference of persistence from early observations of a heterogeneous susceptible--infected--susceptible (SIS) reaction--diffusion model with standard incidence.
6

Blurring the Block Boundaries: Overlapping Block Denoising for Efficient Diffusion Language Models

Amir Mann, Jonathan Fhima, Shaked Brody, et al. arXiv B4

From the abstract Diffusion Language Models (DLMs) have emerged as effective alternatives to autoregressive models for latency-critical tasks, enabling faster token generation. Recent advances employ block diffusion with self-editing capabilities, where each block refines tokens iteratively before advancing to the next one.
7

Cognitive Mapping of Learning by Thinking Mechanisms in Large Language Models

Dong Han, Jin Zhang, Kunyue Yang arXiv G5

From the abstract Large language models (LLMs) increasingly exhibit capabilities that resemble human Learning by Thinking (LbT), the acquisition of new insight through internal processes such as explanation and reasoning, without new external input.
8

From Video Frames to Robot Meta-Frames: A Unified Latent Interface for World Action Models

Guoqiang Liang, Yiqi Lin, Yanzhe Chen, et al. arXiv B3

From the abstract Pretrained video diffusion models (VDMs) offer a powerful foundation for World Action Models (WAMs), which jointly predict robot actions and future observations. Building such models requires injecting low-dimensional robot signals (e.g.
9

Location Matters: Sparse Latent-Variance Attacks for Face Privacy against Diffusion Personalization

Shuting Yang, Shuxin Li, Linyan Dai, et al. arXiv C3

From the abstract Text-to-image diffusion models can learn and reproduce an individual's identity from only a few reference photos, creating risks of unauthorized and misleading content generation.
10

RB-LDC: Redundancy-Balanced Latent Coding for Noise-Robust Diffusion Latents

Hyunseok Jeong, Jaeho Jeon, Young-Sik Kim arXiv C4

From the abstract Latent diffusion models decode approximately correct latents, making decoder sensitivity to perturbations relevant to image quality. After whitening, forward diffusion corrupts every direction at the same signal-to-noise ratio (SNR), despite unequal decoder sensitivities.
11

Mixed Signals: Weight-Space Programming for EEG and ECG

Dara Varam, Taraneh Aminosharieh Najafi, David Atienza arXiv F5

From the abstract Weight-space programming allows language models to read a trained neural network's weights and edit them. This opens the door to ``training'' by reasoning rather than through gradient descent.
12

EgoHMP: Achieving Precise Human Motion Prediction in 3D Scenes via Egocentric Cues

Xin Zhao, Pengzhan Zhou, Ziyi Li, et al. arXiv A3

From the abstract The complex interactive dependencies between human behavior and 3D scenes make accurately predicting human intent a major challenge for human motion prediction (HMP). While recent studies use human gaze to infer intent, gaze signals suffer from ambiguity, drift, and a reliance on eye-tracking hardware.
13

When to Start Denoising? Adaptive Denoising Entry for Generative Video Compression

Zifu Zhang, Yuqiao Li, Ding Ding, et al. arXiv B2

From the abstract Generative video compression offers a promising solution for compressing and reconstructing videos under extremely low bitrate constraints, among which the diffusion-based solutions still achieve the best performance by providing the powerful generative latent prior.
14

Parallel Broyden Methods for Efficiently Evaluating Nonlinear State Space Models

Ian Christopher Tanoh, Scott W. Linderman arXiv B5

From the abstract Nonlinear state space models are typically evaluated sequentially, limiting their ability to exploit modern parallel hardware. Recent work has shown that state space models can be evaluated in parallel by reformulating evaluation as a root-finding problem, which can be solved with a parallel form of Newton's method.
15

Dark Matters: VAE-Aware HDR Encoding for SDR-to-HDR Video Reconstruction

Donghoon Cha, Bomseok Park, Hansl von Kwon, et al. arXiv B2

From the abstract SDR-to-HDR reconstruction is an ill-posed problem in which clipped highlights, crushed shadows, and quantization artifacts must be recovered while preserving the original content.
16

AdjointFlow: Accelerating Scientific Inverse Problems with Physics-Based Transport

Ethan Hanold, Nicholas M. Boffi, Qin Li, et al. arXiv C6

From the abstract Inverse problems are ubiquitous across scientific applications ranging from seismic imaging to astrophysics to nondestructive testing, yet classical solvers often struggle with nonconvexity and repeated evaluations of expensive physical models.

Graph neural networks

4 new

Ran today at 1:19 PM in 0 s · 17 found · 1 of 1 sources answered

1

Hierarchical Graph Representation Learning with Pooling-Induced Substructures

Luca Sbicego, Xiaowen Dong, Dorina Thanou arXiv E2

From the abstract Hierarchical graph-level tasks that require reasoning over substructures that extend beyond local neighborhoods are posing challenges for standard Graph Neural Networks (GNNs).
2

Resilient Integrated Access and Backhaul Network Planning via Graph Neural Network Curriculum Learning

Jie Zhang, Haochen Sun, Mingzhang Zhou, et al. arXiv B6

From the abstract Integrated Access and Backhaul (IAB) is a key enabler for cost-effective millimeter-wave (mmWave) densification, yet practical deployments are strongly constrained by the spatial availability of fiber-connected IAB donors.
3

Analog quantum approximate optimization algorithm for solving the weighted MaxCut problem on three-dimensional single-atom Rydberg array

Junxi Chen, I. I. Beterov arXiv F5

From the abstract We theoretically investigate an analog implementation of the quantum approximate optimization algorithm (QAOA) for the weighted maximum cut problem (MaxCut) problem.
4

Design and Over-the-Air Evaluation of a Digital Twin-Driven Site-Specific Neural Receiver for 5G NR on NVIDIA DGX Spark

Qiuyuan Chen, Chengyang Yao, Liuchang Zhuo, et al. arXiv C5

From the abstract We design and evaluate a site-specific neural receiver for the 5G New Radio (NR) uplink.

Nothing new from

Protein structure

Ran today at 1:19 PM in 0 s · 0 found · 1 of 2 sources answered

CORE: no API key configured Why

CORE. Selected, but the API key it requires was not configured, so this source returned nothing. It did not contribute to the search.

The sources

25 sources

Every source resmon asks, with what the app tells you before a search: whether it needs your key, the rate resmon holds itself to, how it combines your terms, and the finest window of dates it can answer. Open a source for the rest.

The facts come from the catalog in resmon’s source code. What each source’s terms let resmon keep, and why some sources are not here, is in Sources and their terms.

What every source did

A quiet morning and a broken search must never look the same. resmon records what every source did on every run, and says why a source returned nothing: it answered and had nothing, it could not be asked, or it never answered. This is the Source Wall of the collection above, starting with its verdict, word for word.

“4 of 12 sources have not answered any run resmon has a record of. Semantic Scholar rate-limited resmon's last request.”

Crossref, EuropePMC, HAL, OpenAlex, PLOS, PubMed, and Zenodo answered when last asked and are no longer in the most recent run's saved selection.

What each of 12 sources did on each of 6 runs, recorded November 11, 2026. Choose a mark to read it.
Source	Wed, Nov 11						When last asked
Source	Run 52 1:18 PM	Run 53 1:18 PM	Run 54 1:18 PM	Run 55 1:19 PM	Run 56 1:19 PM	Run 57 1:19 PM	When last asked
arXiv							answered1:19 PM EST
CORE							no API key configured1:19 PM EST
Crossref							answered1:18 PM EST
DBLP							could not answer this window1:18 PM EST
DOAJ							did not answer1:18 PM EST
EuropePMC							answered1:18 PM EST
HAL							answered1:18 PM EST
OpenAlex							answered1:18 PM EST
PLOS							answered1:18 PM EST
PubMed							answered1:18 PM EST
Semantic Scholar							rate-limited1:18 PM EST
Zenodo							answered1:18 PM EST

DOAJ · Deep Sweep, “protein structure” · 1:18 PM EST

did not answer

DOAJ could not be queried: HTTP 502 after 1 attempt. This is not a zero — the source did not answer.

Papers
In this run’s selection: Yes

52 Deep Sweep, “protein language models”, 4 sources 1:18 PM EST · 240 returned · 232 new
53 Deep Dive, “equivariant graph neural networks”, 1 source 1:18 PM EST · 80 returned · 80 new
54 Deep Sweep, “protein structure”, 10 sources 1:18 PM EST · 93 returned · 93 new
55 Watch, Graph neural networks, 1 source 1:19 PM EST · 17 returned · 4 new
56 Watch, Protein structure, 2 sources 1:19 PM EST · 0 returned · 0 new
57 Watch, Diffusion models, 1 source 1:19 PM EST · 51 returned · 12 new

answered The source replied and returned records.
answered, zero The source replied and genuinely had nothing. A quiet field, not a fault.
answered in part The source answered every page; resmon stopped at its own time budget or the run's cap.
could not be asked No key, retired, no author query, or a window the source cannot answer.
rate-limited The source was asked and told resmon to slow down (HTTP 429). It is up; try again later.
failed The source was asked and the exchange broke.
outcome not recorded resmon did not observe an outcome. This is not a success and not a failure.
not asked on this run This run's saved selection did not name this source. Nothing was asked of it.

The Source wall of the same collection on November 11, 2026, exported unedited: three searches and each watch's run of the day, every one asked of the public sources for real.

The tour

3 min 41 s

0:00 Today What is new since you last looked, on a map of your whole collection.
0:24 A Deep Dive One source, asked in depth, and its papers on a map of their own.
0:47 A Deep Sweep Several sources at once, and how each one answered.
1:16 Look Up a Paper Every source that holds a paper, from its DOI.
1:42 Watches A search that runs on a schedule, and the calendar of what is due.
2:09 Library Your own PDFs, read in focus mode.
2:30 Ask A question about your collection, answered by your own Codex sign-in.
3:04 Health What every source did on recent runs, and what needs attention.
3:25 Getting Around The command palette, the shortcuts, and a swipe back.

Recorded on November 6, 2026, in a copy of the collection drawn above. The searches in it asked the public sources for real; the clicks are scripted, at a person’s pace.

What it does

Today: What arrived since you last marked all read, on a map of the collection. The map says whether it places papers by shared wording or by your embedding model. New papers are ringed and numbered, with the same numbers in the list beside it.
Watches: A search resmon runs on a schedule, keeping every paper it finds. A watch can follow words, a person’s new papers, or retractions and corrections on a person’s papers. The Calendar shows when each is due, and says so when nothing will run it.
Search: Search a collection, ask the sources, or find one paper. A Deep Dive asks one source in depth and a Deep Sweep asks several at once, with each source’s limits stated first. Every source is asked for each of your terms, and Match: Any Term or All Terms decides which papers are kept. Look Up a Paper looks up a DOI, arXiv id, PMID, or title and says how each copy matched. Export every collection match as BibTeX, RIS, or CSV.
Runs: Running now, waiting for a free slot, and the history. Watch each source’s progress, see why a run failed, and select runs to export or run again. The monitor in the top bar keeps progress and unseen failures in view.
Library: One gallery for your reading queue, files, and projects. Keep PDF, TXT, MD, and TeX files, read them in the app, and pin exact passages in Evidence. Get the PDF downloads copies only under a stated open license and the source and host rules. The reading queue works on Windows; kept files and Evidence do not.
Ask: An assistant that reads your watches, runs, and papers. Through Claude Code or Codex under your own sign-in, or your own provider key. It uses the same tools an outside program can, and nothing is billed through resmon.
Health: Whether the watching itself is intact. The Source Wall marks what every source did on each recent run. The Watchdog says what resmon can tell is wrong with its watching, and a check finds retractions and newer versions of papers you hold.
Outside the window: A command line, an MCP server, and a local HTTP API. The resmon command ships inside the app and installs from its menu. Tools that speak the Model Context Protocol can run watches and searches and read the collection.
Delivery and backup: Reports where you read them. Every run writes a report as a web page, Markdown, and a PDF where a LaTeX engine is installed, with its papers and search record. A watch can email it (Gmail and iCloud Mail settings are filled in for you), post it to a signed webhook, or put it in a folder or a feed on your own disk. Back up the database to a folder you keep.

Keep several collections in one database, open them in separate windows or macOS tabs, and arrange the sections in Settings › Top Bar. The guide covers each screen, and Settings › Tutorials has short recordings.

On your computer

No account: There is nothing to sign up for, and no resmon server to sign in to.
Collection: One database for your collections, watches, runs, and reading, with reports and Library files on your own disk.
Your keys: Kept in your operating system’s keychain, never in the database, and sent only to the service they belong to.
No telemetry: resmon sends no usage data. Searches go to the sources they ask; Ask, summaries, embeddings, PDF downloads, and delivery use the connections you choose. Backups are written to a folder you keep.

Get it

Version 1.0.0 · MIT license

Platform	File	Tested
macOS, Apple Silicon	DMG	Used daily.
macOS, Intel	DMG	Built and opened by the release check on every release; not used daily.
Windows (x64)	Installer (.exe)	Built and opened by the release check; not used daily. Library files and Evidence are unavailable.
Linux (x64)	AppImage	Built and opened by the release check; not used daily.

Download resmon

The macOS builds are not signed, so the first launch needs one extra step. Windows and Linux builds update themselves. The install guide covers both, and running from the source.