ChatGPT-Shortcut (AiShort): a searchable prompt library you can self-host

The problem AiShort solves, and the people it is aimed at

Prompt reuse across a team tends to fail for a boring reason: prompts live in chat histories, notes apps and screenshots, so nobody can find the one that worked last month. AiShort addresses that with a browsable catalog. The README describes the project as a prompt management tool and states it ships 5000+ prompts across writing, coding, office work, learning, design and marketing, with 26 scenario tags for filtering. The target user is not a prompt researcher. It is someone who pastes text into a chat box or a coding tool and wants a starting point rather than a blank page. The README also names Cursor explicitly alongside ChatGPT, Claude and Gemini, which matters because the same prompt often has to be pasted into a different surface depending on the task. The project is written in TypeScript and built on Docusaurus, so the artifact you deploy is a documentation-style static site rather than a service you call. That single design decision shapes everything else in this review: there is no API layer to integrate with, and no runtime that evaluates a prompt for you.

How the library, the extension and the account layer fit together

Three pieces are visible in the repository material. The first is the Docusaurus site, which holds the prompt catalog and the documentation under the same build. The second is the browser extension for Chrome, Edge and Firefox, which opens a sidebar with Alt + Shift + S so prompts are reachable while a chat page is open. The third is an optional account layer that the README separates from the anonymous experience. Without an account you get search, tags and one-click copy. After login, the feature list adds My Collection with drag-and-drop sorting and custom tags, custom prompts, community sharing with votes and comments, JSON export, and levels from L0 to L9 that accrue as you share. That split is the clearest architectural statement in the material: the catalog is public and read-mostly, while personal organisation and contribution sit behind authentication. A Tampermonkey script called ChatGPT Shortcut Anywhere is offered as an alternative to the extension for summoning the sidebar on any site. What the material does not describe is how prompts are stored at the data layer, whether the community layer is a separate service, or how conflicts are resolved when two people edit the same shared prompt. If you plan to run this for a team, treat those as open questions to answer from the deployment guide rather than assumptions.

Getting an instance running: the paths the documentation lists

The README points to a Deployment Guide and names four routes: Vercel, Cloudflare Pages, Docker, or local. For the Vercel route there is a deploy button whose repository URL is https://github.com/rockbenben/ChatGPT-Shortcut/tree/main, which clones the repository into your own Vercel account. Docker and local are listed but the README does not print the commands, so the exact docker run invocation and the local development command have to come from https://www.aishort.top/en/docs/deploy rather than from this article. The same applies to any environment variables the self-hosted build expects. What the README does state is that there is an Offline Edition aimed at air-gapped enterprise or government networks, which needs no backend and no account, and stores data locally in the browser. That last point is a real constraint rather than a footnote: in the offline edition the collection lives in the browser profile, so clearing site data or moving to another machine means starting over unless you exported first. The extension is distributed through the Chrome Web Store, Microsoft Edge Addons, Firefox Add-ons and GitHub Releases, and the sidebar shortcut is Alt + Shift + S across those builds. If you are evaluating rather than deploying, the hosted instance at aishort.top requires no signup and no install, which is the fastest way to judge whether the catalog's contents match your work.

Where the design shows its limits

The catalog model has a failure mode that the feature list does not address. A prompt is stored as text plus tags and a language, and the material gives no sign of a versioning scheme tied to a model. Prompts that work on one model version can regress on the next, and a shared library with community voting has no mechanism described for marking a prompt as stale after a model change. Voting tells you what other people liked, not whether the prompt still produces the output you need today. There is a second constraint in the language layer. The README advertises 18 languages with full UI and prompt translation, and the badge row lists the README translations. Translated prompts are a different artifact from the original, and the material does not explain who reviews a translated prompt or how a translation is kept in sync when the source prompt is edited. For a single user this is invisible. For a team standardising on one prompt across regions it is the thing that breaks. A third limit is scope: this is a copy-and-paste tool. If your workflow needs a prompt to be called programmatically, parameterised with variables, or asserted against in CI, AiShort does not describe any of that, and the JSON export is a data dump rather than an integration surface.

How it differs from a prompt framework such as Promptfoo

The nearest comparison is not another prompt gallery but a tool like Promptfoo, and the difference is what each one treats as the unit of work. AiShort treats the prompt as content: you find it, copy it, paste it, and optionally save it into a collection with tags. Promptfoo treats the prompt as a test subject: you define prompts and test cases in configuration, run them against providers, and get a pass or fail result you can put in a pipeline. AiShort has no evaluation step at all in the material provided. That is not a defect in AiShort, it is a different job. If your question is "what should I paste into Cursor for this refactor", AiShort answers it. If your question is "did my prompt change break the output on the model we ship with", AiShort has nothing to say and Promptfoo is built for exactly that question. The practical consequence is that the two can coexist: keep the library for discovery and personal reuse, and keep the evaluation harness separate, accepting that a prompt promoted from the library into the harness has to be maintained in two places unless you script the JSON export.

Maintenance, release cadence and what the MIT licence covers

The repository is active rather than archived. The most recent push recorded in the material is 2026-09-10, and the release list shows v4.5.0 on 2026-08-29, v4.4.5 on 2026-08-03 and v4.4.3 on 2026-07-15. That is roughly a release per month across the visible window, which for a Docusaurus site plus extension usually means content and UI changes rather than breaking API changes, though the material does not break the changelog down by type. The upgrade cost for a self-hosted instance depends on how far you have diverged from upstream: if you deploy the repository as-is through the Vercel button, updating means pulling the new main branch, and your own prompts live in the account or browser layer rather than in the repository. If you fork and edit the catalog files directly, every upstream release becomes a merge. The licence is MIT, which permits commercial use, modification and redistribution provided the copyright notice and permission notice are retained. The README does not state a separate licence for the prompt text itself, and prompt collections can carry their own terms, so if you intend to redistribute the catalog inside a product, confirm the status of the prompt content with the maintainer rather than assuming the repository licence settles it. Nothing here is legal advice; it is a pointer to the question you need answered before shipping.

Who should install it, and what to check before committing

The fit is a small team or an individual who pastes prompts into chat and coding tools all day, wants a shared vocabulary of tags, and is comfortable running a static site. The browser extension with Alt + Shift + S is the part that changes daily behaviour, because it removes the tab switch. The anti-fit is anyone who needs prompts under version control with automated checks, or who needs a programmatic interface; the material describes neither. Before committing, verify three things in order. First, open the hosted instance and search for the tasks you actually do, because a catalog of 5000+ prompts is only useful if the ones you need are in it and are current. Second, read the deployment guide at aishort.top/en/docs/deploy for the Docker and local commands that this README omits, and confirm which storage the self-hosted build uses for accounts and collections. Third, if your network is air-gapped, confirm the Offline Edition's browser-local storage model is acceptable, since it makes export the only migration path between machines.

Editorial conclusion

Adopt AiShort if you want a shared, searchable prompt catalog with a browser sidebar and are willing to run the Docusaurus build yourself. Do not adopt it if you need per-model prompt versioning, automated evaluation, or a CI gate on prompt changes; nothing in the repository material supports those. Verify first: the deployment guide's Docker and Cloudflare Pages paths, whether the offline edition covers your network, and how the JSON export maps to your own storage before you migrate any prompts into it.

Official sources

1. License: MIT
2. Project website
3. README
4. Releases
5. rockbenben/ChatGPT-Shortcut on GitHub