A research workspace that keeps papers, their versions, and research context connected—from discovery to reuse.
Gewu Insight brings continuous literature discovery, organization, and reading into one workspace. It organizes scientific works as persistent, stateful research objects: each paper retains a shared identity as metadata, publication versions, supporting files, project associations, and analyses accumulate. Researchers and authorized AI clients can return to the same record, inspect earlier materials, and continue their work. This provides an inspectable scientific memory layer, with questions and scientific judgment directed by the researcher.
Keep research objects visible, evidence traceable, and insight cumulative.
Website · Paper · Quick start · Explore the workspace · User Guide · AI & MCP · Citation · License
Version 1.0.0 · macOS and Windows · Source-available research software
Academic/non-commercial research use is permitted under the custom LICENSE. Commercial use requires separate prior written authorization.
One paper, connected files: inspect its versions, choose a Default, and read its metadata alongside your Library.
- Follow the literature. Track journals, research topics, and authors through three complementary monitors. Save relevant papers into your own collections.
- Organize around your research. Use Library folders and project Spaces; the same paper can appear in several contexts with one shared identity.
- Keep the evidence together. Read PDFs and Markdown notes, compare versions, and retain supplements, annotations, data, and code in Paper Files.
- Continue with context. Revisit saved reading material, maintain your own publication record, and optionally let an authorized AI client read papers and append analyses with source references.
Download the installer for your platform from the release page's Assets. Both packages include Python and application dependencies. GitHub's Source code ZIP is for source-based setup, not an App installer.
This repository is currently private. Its GitHub release links require an account with repository access; contact info@gewu-insight.com if you need access or an installation package.
| Platform | Requirements | Download |
|---|---|---|
| macOS | macOS 14 or later; Apple silicon or Intel | macOS 1.0.0 DMG |
| Windows | Windows 10 version 1607+ (build 14393+) or Windows 11; Intel/AMD x64 | Windows 1.0.0 Setup |
- Download Gewu-Insight-macOS-1.0.0.dmg and open it.
- Drag Gewu Insight.app to Applications, eject the disk image, and open the installed App.
- Choose Create New Database or Open Existing Database. Keep your Gewu Database in a separate folder outside the App. The App menu's Database Location… shows the active folder.
- Download Gewu-Insight-Windows-1.0.0-x64-Setup.exe and run Setup.
- Choose the program and Gewu Database folders. To retain an existing collection, select Use existing Gewu Database and its root folder.
- Open Gewu Insight from the Start menu. Python and .NET are bundled; Setup needs internet access if Microsoft WebView2 must be installed.
Read the release's checksum and first-launch notes. The macOS package uses ad-hoc integrity signing and is not Apple-notarized; the Windows installer is unsigned. Optional LibreOffice and TeX tools are installed separately.
Try one paper: drag a PDF into a Library folder, or visit an article in Browser and choose Save to Gewu. Inspect Paper Files and Paper Details, then create a Space for a project. Open Help → User Guide for the manual bundled with your platform. The website also provides both platform guides.
Updating an existing installation: close the App and back up the selected Gewu Database and its files first. Replacing the App preserves the separate database location. Older source-folder installations should follow the migration instructions before updating. For source-based setup, use Manual setup.
If startup fails, follow your platform's release notes and User Guide. Installed
macOS App logs are in ~/Library/Logs/Gewu Insight; do not delete the database
to repair startup.
Six main tabs connect discovery, reading, organization, and your own publications. The Library tab provides access to Library, independent Spaces, and shortcuts. Spaces are project roots at the same organizational level as Library. These screenshots show the real macOS App; select an image to inspect it at full size.
Library & Spaces Organize papers and standalone files in folders and project Spaces. Browse List or Table View, search a collection, and sort by Time added. Shortcuts provide another route to existing folders. |
My Publications Maintain your publication list, authorship roles, ordering, and citation evidence. Export to Word, LaTeX, PDF, or an Overleaf project. Manual roles and ordering survive metric refreshes. |
Journal Monitor Follow selected journals and arXiv updates. Browse by date, venue, and field, inspect figures and abstracts, and save papers with their source information. Your subscription choices persist across updates. |
Subject Monitor Build reusable topics from search terms and review their dated results. Selecting a calendar date opens saved results; refreshing a topic is a separate action, with its status visible. |
Author Monitor Follow selected researchers through supported scholarly profiles, browse their publications, and identify new unread papers. Refreshes respect source limits and show cooldowns when a service restricts access. |
Browser Read publisher pages and PDFs in the built-in browser (WebKit on macOS), with tabs, bookmarks, and local history. Save to Gewu lets you review the paper's identity, destination, and file type before importing it. |
For an existing collection, Folder Loader imports a directory tree. Portable folder export includes files and a manifest for importing their relationships into another Gewu workspace; see the User Guide.
Browser tab controls, bookmark import and sync, downloads, and system Passwords behavior are covered in the User Guide and Browser documentation. Website links from the other workspaces open in Browser; on macOS, local PDFs open in Preview.
A paper's Gewu Key connects its identifiers, metadata, files, and placements across the workspace. Adding that paper to another folder or Space reuses the shared record. Start with metadata and add files later under the same Key.
Paper Files keeps the Published version, accepted manuscript, preprints, supplements, annotated copies, and other attachments associated with that paper. Choose which file is Default; your choice is retained when other versions arrive. Use Quick Look to preview PDFs and Markdown, or open local PDFs in Preview on macOS.
Supported corrections and retractions stay linked to the original paper. Refresh Info checks explicit Crossref update relations; coverage depends on the source, and saved papers are not continuously monitored for notices. See publication updates.
Managed documents remain real files. Move and rename them through the App so records and disk paths stay coordinated. See Paper Files and managed file layout for classifications, storage and file operations.
AI Access is optional and off by default. Gewu provides a local API and an MCP adapter so an AI client can work with explicitly authorized collections. Enabling access does not itself call a model, upload a library, or start an analysis job.
In Setting (macOS) or Settings (Windows) → AI Access, create a named connection and choose its collection, expiry and permissions. Use the generated configuration with a client supporting local stdio MCP. The setup guide covers Codex and DeepSeek Harness; compatible clients such as Claude Code can use the same interface.
A typical workflow is: select a paper → read its authorized source files and earlier analyses → save a new Markdown analysis → review it in Paper Files. A later authorized session can retrieve the same record and continue from the retained sources and notes. Saving is an explicit client action; there is no automatic synchronization of AI conversations.
Reading, saving analyses, and organizing/importing have separate permissions. Analysis saves append notes with source references without overwriting existing Paper Files. You can revoke connections, and upgrades do not expand existing grants. The AI Access guide explains setup, metadata-only imports, file attachment, Resolve, provenance and the limits of each operation.
Your AI client may send authorized content to its chosen model provider. These grants govern API/MCP access; they do not restrict a program's independent operating-system access to local files. Review generated analyses against their sources.
SQLite is the source of truth for paper identities, relationships, file records, and workspace state. PDFs and attachments remain on disk; reports and exports do not replace the database.
The installed App and your Gewu Database are stored separately. With the App and its service stopped, back up the complete selected database folder, any additional managed file roots, and externally linked originals. Copying the App or source checkout alone does not back up your research. A pre-upgrade SQLite snapshot or a portable folder export does not replace this full backup. See backup and recovery and the platform's bundled User Guide.
The workspace uses a local-first architecture, with online services supporting discovery, metadata, citation counts, and publisher access. Availability can depend on a service's limits, the network or VPN exit address, and institutional access. Gewu uses pacing and cooldowns; it does not bypass paywalls or access controls. Configure optional service keys in Setting (macOS) or Settings (Windows) → Literature Data Source.
Gewu Insight was originally developed for macOS, which remains the primary platform for development and testing. The Windows version was subsequently ported from the macOS implementation.
- macOS — a standalone desktop release with Finder, Preview, Quick Look, Dock, and native drag-and-drop integration.
- Windows — a separately packaged desktop release for Windows 10/11 x64, with its own installer and User Guide. See the Windows release notes for requirements and platform details.
Windows validation is currently more limited. Windows 11 has undergone limited testing in a virtual machine. Compatibility fixes for Windows 10 have been guided by feedback from external testers, while direct testing and regression verification by the development team on Windows 10 remain limited.
The evaluation reported in the accompanying paper was conducted on macOS.
The Python, SQLite, and web-interface core is separated from native desktop integration. Screenshots in this README show macOS; use your platform's release notes and User Guide for installation and native behavior.
| Start here | What it covers |
|---|---|
| macOS User Guide · Both platform guides | Daily workflows, settings, shortcuts, backup, and troubleshooting; each App includes its own guide under Help → User Guide. |
| AI Access & MCP | Client setup, collection grants, reading, analysis saves, imports, moves, and renames. |
| Browser: Save to Gewu | Reviewing article identity, PDF candidates, file types, and import destinations. |
| Browser: keyboard & bookmarks | Navigation, bookmark import/export, and explicit one-way sync. |
| Browser: PDFs · Passwords | PDF viewing and the boundaries of system password integration. |
| Managed file layout | Default files, Associated Files, aliases, and recovery boundaries. |
| macOS release & installation · Windows release & installation | Current installers, requirements, checksums, and first-launch notes. |
| Older source-installation migration | Back up and separate research data from a legacy source checkout before updating. |
| Development guide | Manual setup, architecture, CLI commands, builds, updates, and verification. |
| Distribution guide | Package formats, separate database storage, signing, and release checks. |
| Licensing explained · Contributing | Research use and sharing, reserved commercial rights, and contribution terms. |
The HTML manual is the source for the bundled PDF. Guide coverage and screenshot provenance are recorded separately from the product overview.
If you use Gewu Insight in your research, please cite the accompanying paper.
Kong, L. & Wu, Z. A Stateful Research Information Infrastructure for Scientific Discovery. arXiv:2610.14538 (2026). Read the paper.
@misc{kong2026stateful,
title = {A Stateful Research Information Infrastructure for Scientific Discovery},
author = {Lingyuan Kong and Zhengran Wu},
year = {2026},
eprint = {2610.14538},
archivePrefix = {arXiv},
primaryClass = {cs.DL},
url = {https://arxiv.org/abs/2610.14538}
}Citation metadata is also available in CITATION.cff for GitHub's Cite this repository feature.
The manuscript's evaluation uses the fixed software snapshot
commit 539d00d.
Use that snapshot when reproducing the reported results.
Gewu Insight is known in Chinese as 格物洞见. The name draws on 格物致知 and expresses a research outlook shaped by the practice of physics.
- 格物 · Gewu — begin with concrete phenomena, experimental observations, and questions from nature; keep research in contact with the world it seeks to understand.
- 洞见 · Insight — seek the patterns and mechanisms behind phenomena through physical pictures, intuition, imagination, and insightful guesses. Mathematics serves as a tool for developing and expressing that understanding.
The workspace supports this practice by keeping papers, versions, supporting files, and notes connected, so researchers can revisit experimental results, compare physical pictures, and pursue emerging ideas. Questions, interpretation, and scientific judgment remain with the researcher. Read more in the User Guide's explanation of the name.
Investigate nature through experimental phenomena; seek insight through physical pictures and intuition.
（从实验现象中格物，以物理图像和直觉求洞见。）
Designed and developed by Lingyuan Kong and Zhengran Wu.
Developed with coding assistance from OpenAI Codex.
Copyright © 2026 Lingyuan Kong and Zhengran Wu. All rights reserved.
Gewu Insight is source-available research software under the custom Gewu Insight Academic and Non-Commercial Research License, not an open-source license. The LICENSE permits academic/non-commercial research use, including downloading, installing, running, studying, and modifying it. Original and modified versions, derivatives, and patches may be shared free of charge for permitted research under the same terms, retaining notices and identifying modifications. This includes public research forks and pull requests.
Commercial use, commercial redistribution, sublicensing, and integration into commercial products or services require separate prior written authorization. Contact info@gewu-insight.com. Independent research outputs do not inherit the software license merely by using Gewu. A pull request does not transfer its author's copyright or grant additional commercial rights; see the contribution policy.
Third-party components, scholarly content, and services retain their own terms. See third-party notices and the dependency and asset audit, including outstanding font and artwork permissions.