ChatGPT-Shortcut (AiShort): a 5000+ prompt library you can self-host

The problem AiShort attacks: prompt reuse across tools

Most people write a prompt once, get a usable answer, and then lose the prompt in a chat history. The README frames the gap bluntly: "The gap between using AI and using AI well is one good prompt." AiShort is an attempt to make prompts a browsable, searchable asset instead of a thing you retype.

The audience is broad rather than specialist. The README lists writing, coding, office work, learning, design and marketing as covered areas, and the copy is aimed at people who use ChatGPT-style chat pages, coding tools such as Cursor, or their own API calls. The hosted site at aishort.top requires no signup, no paywall and no install, which tells you the maintainers expect casual visitors, not just developers. The repository matters to a narrower group: teams that want the same library behind their own domain, or on a network with no outbound access.

How the site, the extension and the prompt data fit together

The repository is a Docusaurus 3.10.2 site in TypeScript, with React 19 and Ant Design 6 in package.json. The build is not a plain docusaurus build. The scripts section defines prestart, predev and prebuild hooks that all run node scripts/generate.mjs, and build itself runs node scripts/buildPhased.mjs. Locale handling is centralised in scripts/i18nLocales.mjs, which the Dockerfile comment describes as the single source of truth for all locales.

That phasing exists for a concrete reason. The Dockerfile comment states that a single process building 18 languages ran out of memory, so yarn build now runs in segments with at most three locales per process. The same file pins the builder stage to $BUILDPLATFORM so the static output is produced once on amd64 instead of being rebuilt under QEMU for arm64, which the comment says caused concurrent builds and OOM kills. The final image is still nginx:stable-alpine serving /app/build on port 3000.

The extension is the other delivery path. It is distributed through the Chrome Web Store, Microsoft Edge Addons and Firefox Add-ons, opens a sidebar with Alt + Shift + S, and a Tampermonkey script called ChatGPT Shortcut Anywhere summons the same sidebar on any website. The README does not describe how the extension talks to the backend, so treat the sync behaviour as undocumented rather than assume it.

Installing and running your own instance with Docker

The repository ships a Dockerfile and a docker-compose.yml, and the compose file is the least ambiguous way to start. It pulls a prebuilt image from the GitHub Container Registry, maps port 3000 and sets the restart policy.

services:
  chatgpt-shortcut:
    container_name: chatgpt-shortcut
    image: ghcr.io/rockbenben/chatgpt-shortcut:latest
    ports:
      - "3000:3000"
    restart: unless-stopped

Run docker compose up -d from the repository root, then open http://localhost:3000. You should see the AiShort home page with the prompt card grid and the tag filters.

If you prefer to build the image yourself, the Dockerfile ends with the commands the maintainers document in a comment:

docker build -t chatgpt-shortcut .
docker run -d -p 3000:3000 --name chatgpt-shortcut chatgpt-shortcut

The build sets SKIP_GIT_INFO=true, which the comment explains disables showLastUpdateTime and uses the current time as buildDate, removing any dependency on git and shrinking the build context. Expect the build to be slow: it installs dependencies with yarn install --frozen-lockfile and then compiles the site for every locale.

For a first real use, go to the search box, press Ctrl or Command + K, and type a scenario such as coding or marketing. The README states there are 26 scenario tags to filter by. Click Copy on a card and paste the result into whichever chat or coding tool you use. The README lists Vercel, Cloudflare Pages, Docker and local deployment as options, and points at the Deployment Guide for the details of each.

Where AiShort stops being the right tool

The library is a catalogue of prompts written by the maintainers and the community. Nothing in the README suggests it reads your repository, your documentation or your past conversations to produce a prompt. If your need is a prompt that encodes your team's internal conventions, you are looking at a starting point, not a finished artifact.

The account boundary is the second constraint. The README divides features into a public set (one-click copy, tag and keyword search, 18 languages) and an advanced set that appears after login: 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. If you self-host, the README does not state whether login, community and export work without additional backend configuration, and the deployment documentation is the place that would have to say so. Plan for the public browsing and copy features unless you have confirmed the rest.

Language coverage is worth reading carefully. The README advertises full UI and prompt translation across 18 languages, and the repository carries matching README-lang files, but the actual prompt corpus behind a translated interface may be thinner than the interface itself. The README does not publish per-language prompt counts.

AiShort versus a prompt file in your own repository

The obvious alternative is a prompts.md file, or a directory of text files, committed next to the code. That approach wins on review: prompts change through pull requests, diffs are visible, and the file travels with the project. It loses on discovery. Nobody browses a markdown file to find out what prompts exist, and there is no tag filter or search key.

AiShort inverts those properties. Discovery is the product: 26 scenario tags, a Ctrl or Command + K search, and a card view with one-click copy. Contribution is looser. The community route is share, vote and comment, and the README ties participation to levels L0 through L9 rather than to a review process. For a team that already has a prompt file and is happy with it, AiShort adds a second place where prompts live, and the README does not describe an import path from an existing repository into a self-hosted instance. The JSON export runs in the other direction, out of AiShort.

Maintenance, upgrades and the MIT licence

The last push to the default branch was on 2026-09-10, and the most recent release is v4.5.0 from 2026-08-29, following v4.4.5 on 2026-08-03 and v4.4.3 on 2026-07-15. The repository is not archived. That release cadence is the practical upgrade cost: if you track the prebuilt ghcr.io/rockbenben/chatgpt-shortcut:latest image, a restart picks up whatever shipped most recently, which is convenient and also means you are not pinned to a version you tested. Pinning a specific tag is the alternative the compose file leaves open.

Building from source is heavier. The Dockerfile explains that the build was split into phases precisely because a single process compiling 18 locales ran out of memory, so a self-built image needs headroom on the build machine. package.json also carries an update script that runs npx --yes npm-check-updates -u followed by yarn install, which will move dependencies forward without a review step; running it is a deliberate choice, not routine maintenance.

The licence is MIT, and the repository includes a LICENSE file. That is permissive and permits commercial use and modification, but this is a description of the licence text, not legal advice. If you plan to redistribute a modified instance, read the LICENSE file and, if the stakes are high, take your own advice on attribution.

Editorial conclusion

Adopt ChatGPT-Shortcut if you want a prompt catalogue your whole team can browse, or if an air-gapped network forces you to run one locally: the Dockerfile and docker-compose.yml in the repository are the shortest path, and the documented Offline Edition stores data in the browser with no backend. Skip it if you need prompts generated from your own codebase or your own style guide, because the library is curated and community-contributed rather than derived from your material. Before rolling it out, verify which feature set you actually get, since collection, custom prompts, community and JSON export are listed as available after login, and check the Deployment Guide for the platform you intend to use.

Frequently asked questions

Official sources

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