PromptVault: An AI Prompt Manager With No Backend

TL;DR - Key Takeaways

1. PromptVault is a free-forever prompt manager for AI tools — a Manifest V3 browser extension plus a web app — that stores your reusable prompts in your own Google Drive, not on anyone's server.
2. There is no backend. Prompt data lives in a vault.json in a PromptVault folder in your Drive; media bytes upload to the same folder; the vault keeps only a lightweight reference. All compute is client-side.
3. drive.file least-privilege scope — the app can only ever see the files it created. It cannot read the rest of your Drive, which also keeps it out of Google's restricted-scope verification.
4. Right-click to paste a saved prompt straight into any text field, or copy from the toolbar popup and paste anywhere — including pages where extensions can't inject. Save selected text or any web image into your vault from the context menu.
5. OAuth implicit grant, no client secret — sign-in uses chrome.identity.launchWebAuthFlow with a public client ID; tokens stay in local storage and are revoked on disconnect. No telemetry, no third-party calls.
6. Shared core, two front-ends — a vanilla-JS MV3 extension and a Svelte web app both build on one @promptvault/core package; the web app ships to GitHub Pages, the extension as a signed CRX.

Why I Built This

If you use Claude, ChatGPT, or Gemini every day, you accumulate prompts. Prefaces that set the tone. Guardrails you paste before a risky task. A persona. A "explain it like I'm a senior engineer" block. Snippets you reach for constantly.

Mine were scattered across notes apps, a prompts.txt, and my actual clipboard history. Every "prompt manager" I tried wanted me to sign up for an account, trust their servers with the text I paste into AI tools, and — eventually — pay a subscription to keep my own writing.

That's backwards. A prompt library is just text and a few images. It should live somewhere I already own, cost nothing to run, and never touch a third-party server.

So I built one with a hard constraint: no backend, ever. If there's no server, there's no data to leak, no bill to pay, and nothing for me to shut down later.

What It Is

PromptVault has two front-ends that share one library:

• A browser extension (Chromium: Chrome, Edge, Brave, Arc, Opera) that lives where you actually type. Right-click to paste a prompt into the page, or open the toolbar popup to search and copy.
• A web app (tech-sumit.github.io/promptvault) that manages the same library — full CRUD, categories, search, import/export, drag-and-drop media.

Both read and write the same vault in your Google Drive, so your prompts follow you between the two.

Items are organized into six fixed categories — preface, guardrail, system, personal, snippet, and media — each with a title, body (Markdown), optional link, tags, and an optional media attachment.

Architecture: Bring Your Own Storage

The whole design follows from one rule: the user's data lives in the user's storage. There is no application database because there is no application server.

graph TD
    subgraph drive ["Your Google Drive — /PromptVault folder"]
        VJ["vault.json<br/>prompts + metadata"]
        MED["media files<br/>(images you attach)"]
        VJ -. "DriveRef: id, link, thumb" .-> MED
    end

    subgraph clients ["Client-side only — no backend, no database"]
        EXT["Browser Extension<br/>MV3: service worker · content<br/>popup · dashboard"]
        WEB["Web App<br/>Svelte + Vite → GitHub Pages"]
        CORE["@promptvault/core<br/>vault · drive · merge · model · markdown"]
    end

    EXT --> CORE
    WEB --> CORE
    CORE -- "read / write vault.json<br/>(revision + merge)" --> VJ
    CORE -- "multipart upload · blob fetch" --> MED
    AUTH["Google OAuth — implicit grant<br/>drive.file + userinfo.email"] --> CORE

The storage model

A media item's bytes are uploaded to your Drive's PromptVault folder; the vault stores only a lightweight DriveRef — the file id, a link, and a thumbnail reference. Prompt text and metadata live in a single vault.json in the same folder. The app never holds your data; it holds a pointer into your Drive and a short-lived access token.

Sync and conflict resolution

vault.json is read and written through the Drive API with revision tracking. On write, the client passes the expected headRevisionId; if Drive has moved on (you edited from another device), it raises a ConflictError, and the client merges instead of clobbering. The merge is last-write-wins per item (newest updated timestamp wins) with tombstones so a delete on one device propagates instead of resurrecting. The extension also caches locally and syncs on a periodic alarms tick plus a debounced push after edits.

One core, two front-ends

@promptvault/core is the shared brain — the vault, the Drive client (multipart upload, blob download, vault read/write), the merge logic, the data model, and a small XSS-safe Markdown renderer. The extension is vanilla JS (Manifest V3, no framework, loads as "unpacked" with zero build step beyond bundling). The web app is Svelte + Vite. Both depend on the exact same core, so behavior can't drift between them.

Inside the extension

• Service worker — builds the right-click menus, routes a chosen prompt to the active tab, and runs background Drive sync.
• Content script — pastes into the page. Text is inserted at the caret via the native value setter plus a bubbling input event, so React-controlled editors (ChatGPT/Claude/Gemini) actually register the change. Images are re-encoded to PNG (Chrome's clipboard write() only accepts PNG) and pasted via execCommand — which works in a content script thanks to the clipboardWrite/clipboardRead permissions.
• Popup / dashboard / onboarding — search-and-copy, full management with Markdown rendering and lazy Drive-image previews, and a one-click "Connect Google Drive" flow.

The extension ID is pinned via a manifest key so dev builds and the published build share one identity, and CI produces a signed CRX for self-distribution alongside the Web Store zip.

How to Use It

The web app (nothing to install)

1. Open tech-sumit.github.io/promptvault/app.
2. Click Sign in with Google. PromptVault creates a PromptVault folder in your Drive.
3. Add items — text prompts, links, or drag-and-drop an image to attach media.
4. Hit Copy on any item and paste it wherever you're working.

The browser extension

1. Load it (Web Store, or load the dist/ folder unpacked) and click Connect Google Drive — one click, no setup, no Client ID to create.
2. Paste into any field: right-click in a composer → Paste from PromptVault → pick an item. Text drops straight in at the caret; an image is copied so you press ⌘/Ctrl+V.
3. Copy from the popup: open the toolbar popup, search, and Copy — then paste anywhere, including chrome:// pages and built-in AI boxes where no extension can inject.
4. Capture from the web: right-click selected text → Save selection to PromptVault, or right-click any image → Save image to PromptVault (it uploads to your Drive and adds a media item).

Your prompts sync through your own Drive, so the extension and the web app always show the same library.

Security and Privacy

This is the part I care about most, because a prompt manager sees text you trust to AI tools. The design is built so that there is almost nothing to trust me with.

• No backend, no database. Your prompts and media live in your Google Drive. I run no server, so there is no copy of your data anywhere I control.
• drive.file least-privilege scope. The app can only access files it created — the PromptVault folder and nothing else. It is technically incapable of reading the rest of your Drive. (This also keeps it clear of Google's restricted-scope review.)
• userinfo.email only, used solely to show which account you connected.
• OAuth implicit grant, no client secret. Sign-in uses chrome.identity.launchWebAuthFlow with a public client ID. There is no secret embedded in the extension — there's nothing to extract from the package.
• Tokens stay local. The access token is cached in chrome.storage.local (extension) / localStorage (web). Disconnect revokes the token and clears it.
• No telemetry, no analytics, no third-party calls — the only network requests are to the Google APIs you authorized.
• Untrusted text is escaped. Saved and captured content is HTML-escaped before rendering (the Markdown renderer escapes first, then formats, and only allows http(s) links) — no innerHTML of raw user text.
• Open source. The whole thing is on GitHub; you can read exactly what it does before you connect anything.

The summary: bring your own storage, least privilege, no secrets, no servers.

Tech Stack

What's Next

• Chrome Web Store listing (the package and signed CRX are already built by CI).
• Per-site auto-insert for image attachments into the major AI composers.
• Optional end-to-end encryption of vault.json for users who want their prompts unreadable even to Google.

PromptVault is open source — if a no-backend, your-storage prompt manager is the kind of thing you'd use, the code and the live app are both linked below.

Live app: tech-sumit.github.io/promptvault Source: github.com/tech-sumit/promptvault