PromptEnhance

Select text in any editable field on any web page — a chat box, a form, anything — click the ✨ Enhance button that appears, and the selection is rewritten in place into a sharper prompt. Runs entirely as a browser extension; no local server required. You bring your own API key for whichever provider you want to use (or point it at a local Ollama or LM Studio install for a fully offline setup).

Supported providers: Anthropic (Claude), OpenAI, Google (Gemini), OpenRouter, Ollama, LM Studio.

📺 Watch a demo video

Requirements

• A Chromium-based browser: Google Chrome, Microsoft Edge, or Brave all work the same way. (Firefox is not supported — this extension uses Manifest V3 APIs.)
• An API key for whichever cloud provider you choose, or a local Ollama or LM Studio install if you'd rather run models on your own machine.

Load the extension

These steps are the same on every platform — only where the browser's menu lives differs slightly.

1. Open your browser and go to the extensions page:
   • Chrome: chrome://extensions
   • Edge: edge://extensions
   • Brave: brave://extensions
2. Turn on Developer mode (a toggle, usually top-right of the page).
3. Click Load unpacked.
4. In the file picker, select this project's folder (the one containing manifest.json — e.g. PromptEnhance/, wherever you placed it on disk).
5. PromptEnhance should now show up in your extensions list and in the toolbar (pin it via the puzzle-piece icon if you don't see it).

Platform notes for step 4 (file picker)

• Windows (default): browse to wherever you saved the folder, e.g. C:\Users\<you>\Desktop\PromptEnhance.
• macOS: same dialog, just a macOS-style file picker — e.g. /Users/<you>/Desktop/PromptEnhance or wherever you cloned/unzipped it.
• Linux: same idea — e.g. /home/<you>/PromptEnhance. If Chrome was installed via Flatpak or Snap, its file picker may only show folders under your home directory by default; keep the project there, or grant the sandboxed browser access to wherever you placed it.

There is no build step — the folder is loaded as-is.

Configure a provider

1. Click the PromptEnhance toolbar icon (or right-click it → Options).
2. Pick a Provider.
3. Pick a Model from the dropdown — click Refresh models (or just enter your API key) to populate it from the provider's live model list. If you need a model that isn't listed yet, choose Custom (type manually)… at the bottom of the dropdown to type an id yourself.
4. Enter your API key (not needed for Ollama or LM Studio — see below).
5. Click Save.

Each provider remembers its own Model/API key/Base URL independently — switching the Provider dropdown loads that provider's own saved settings (pre-filling a sensible default Base URL the first time you pick a local provider) rather than showing whatever was last typed in for a different provider.

Keys are stored locally in the browser's extension storage — they are never sent anywhere except directly to the provider's own API.

Using Ollama instead of a cloud API

Ollama runs models locally, so there's no key and nothing leaves your machine.

1. Install Ollama:
   • Windows / macOS: download the installer from ollama.com/download.
   • Linux: curl -fsSL https://ollama.com/install.sh | sh
2. Pull a model, e.g. ollama pull llama3.1:8b.
3. Make sure the Ollama server is running (the installer starts it automatically on Windows/macOS; on Linux it typically runs as a systemd service — systemctl status ollama to check, or run ollama serve manually).
4. In PromptEnhance's Options, set Provider to Ollama (local). The default Base URL of http://localhost:11434 works for a standard local install — only change it if you've configured Ollama to listen elsewhere.
5. Click Refresh models to list what you've pulled, pick one, Save.

If Ollama rejects the extension's requests (403 / "Invalid API key")

Ollama's server checks the incoming request's Origin header against an allowlist and returns 403 Forbidden for anything not on it — a browser extension's chrome-extension://<id> origin isn't there by default. This has nothing to do with an API key (Ollama doesn't use one locally); you need to tell Ollama to accept requests from the extension via the OLLAMA_ORIGINS environment variable, then restart it:

• Windows: open a terminal and run setx OLLAMA_ORIGINS "*", then close and reopen Ollama (or restart the Ollama service/tray app) so it picks up the change. (You can also set this permanently via System Properties → Environment Variables.)
• macOS: if Ollama is running as the menu-bar app, run launchctl setenv OLLAMA_ORIGINS "*" in Terminal, then quit and reopen Ollama. If you run it via ollama serve in a terminal instead, export OLLAMA_ORIGINS="*" in that shell before starting it.
• Linux (systemd service): sudo systemctl edit ollama.service, add Environment="OLLAMA_ORIGINS=*" under [Service], then sudo systemctl daemon-reload && sudo systemctl restart ollama. Running manually instead? export OLLAMA_ORIGINS="*" before ollama serve.

* allows any origin (simplest for a local personal setup); if you'd rather scope it to just this extension, use chrome-extension://<your-extension-id> instead — find the id on chrome://extensions under the PromptEnhance card.

Using LM Studio instead of a cloud API

LM Studio also runs models locally with no key required — its built-in server just speaks the same API shape as OpenAI, so it plugs in the same way Ollama does.

1. Install LM Studio from lmstudio.ai (Windows, macOS, and Linux builds are all available from the same download page).
2. Download a model inside LM Studio (the "Discover" / search tab).
3. Start the local server: in LM Studio, go to the Local Server (or "Developer") tab, load the model you downloaded, and click Start Server. It listens on http://localhost:1234 by default.
4. In PromptEnhance's Options, set Provider to LM Studio (local). The default Base URL of http://localhost:1234 matches LM Studio's default — only change it if you've configured LM Studio to listen elsewhere.
5. Click Refresh models to list what's loaded, pick one, Save.

Reasoning models and "thinking" tokens

Reasoning-capable models (DeepSeek-R1, Qwen3, QwQ, etc.) run locally through Ollama or LM Studio sometimes leak their internal chain-of-thought straight into the response, wrapped in a <think>...</think> block, ahead of the actual answer. PromptEnhance asks the server not to think in the first place (best-effort, and only honored by models/templates that support it), and as a guaranteed fallback strips any <think>, <thinking>, or <reasoning> block from the response before it's inserted into your page — so you shouldn't see this leak regardless of the model. Cloud providers aren't affected: their reasoning is returned in a separate field the adapters already skip.

Confirmed working against a reasoning model in LM Studio; the same fallback applies to Ollama but hasn't been separately verified there yet.

Using it

1. Select text inside any textarea, input, or chat box (contenteditable) on any page.
2. A small ✨ Enhance button appears near your selection.
3. Click it. The button shows a loading spinner, then the selected text is replaced in place with the enhanced version.
4. If you don't like the result, Ctrl+Z (Cmd+Z on macOS) undoes it back to your original text, the same as any other edit in that field.

After editing the code

There's no build/watch process. After changing any file, go back to the extensions page and click the reload icon on the PromptEnhance card to pick up the changes. If you edited content/content-script.js, also refresh any tab you're testing in.

Timeouts

Each request the extension makes gives up after a fixed time and shows an error rather than hanging forever. These are hardcoded per provider (in background/providers/), not configurable from the UI:

Provider	Enhance request	Model list
Anthropic, OpenAI, Google, OpenRouter	30 seconds	15 seconds
Ollama, LM Studio	5 minutes	15 seconds

Cloud providers get a short timeout because a healthy API should respond in a couple seconds. Ollama and LM Studio get 5 minutes instead, since local inference has no comparable SLA — a cold model load (reading multi-GB weights into memory) or CPU-only generation can legitimately take minutes, especially for larger models. The model-list fetch stays short everywhere, since that's just metadata the server should already have on hand.

If you're regularly hitting the 5-minute local ceiling, that generally means the model itself is too large/slow for your hardware for interactive use — try a smaller or more quantized model rather than waiting it out.

Troubleshooting

• No button appears on selection: make sure you're selecting inside an actual input/textarea/contenteditable element, not plain page text.
• "No model configured" error: open Options and make sure Provider, Model, and (if required) API key are all filled in and saved.
• Model dropdown says "Failed to load models": double check the API key, or for Ollama/LM Studio, that the local server is actually running (and its server tab started, for LM Studio) and reachable at the Base URL shown.
• "Request timed out" error: see Timeouts above for how long each provider is given before this fires.
• "Invalid API key" error on Ollama, even with no key set: this is actually Ollama's origin allowlist rejecting the extension (HTTP 403), not a real key problem — see If Ollama rejects the extension's requests above.
• Enhance button just says "Reload page": the extension was reloaded/updated (e.g. via chrome://extensions) after this tab was already open. A page that was loaded before an update keeps talking to the old, now-disconnected extension context — Chrome doesn't re-inject the new content script into already-open tabs automatically. Refresh the tab (F5) and it'll work again; this only comes up right after an update, not during normal day-to-day use.
• Something else looks broken: on the extensions page, click the "service worker" link on the PromptEnhance card to open its console, and check the page's own DevTools console (F12) for content-script errors.

---

Copyright © 2026 digitalgods.ai. All rights reserved.