README
An LM Studio plugin that gives local LLMs four web-oriented tools built on @lmstudio/sdk. Web and image search are backed by DuckDuckGo (via the browser-fingerprinting impit HTTP client), website visits use @mozilla/readability for article extraction plus turndown for clean Markdown output, and images referenced by the image-oriented tools are downloaded into the chat's working directory so the assistant can display them inline.
DuckDuckGo web search.
| Parameter | Type | Notes |
|---|---|---|
query | string | Required. |
page | int 1–100 | Optional, defaults to 1. Enables pagination. |
Returns an array of [title, url] pairs. Results are cached by (query, safeSearch, page). The results-per-page cap is controlled by the plugin's limitWebResults toggle and webMaxResults slider (default 10, max 30) — these are plugin-only settings and are not exposed as tool parameters. Disabling limitWebResults returns every result DuckDuckGo includes on the page. Safe-search mode is plugin-only (default moderate).
DuckDuckGo image search. Requires a VQD token scraped from the DuckDuckGo homepage.
| Parameter | Type | Notes |
|---|---|---|
query | string | Required. |
page | int 1–100 | Optional, defaults to 1. |
Matching images are downloaded into the chat's working directory and returned as local file paths. If a download fails, the remote URL is returned for that slot instead so the assistant always gets something displayable. The results-per-page cap is controlled by the plugin's limitImageResults toggle and imageMaxResults slider (default 10, max 100) — these are plugin-only settings and are not exposed as tool parameters. Disabling limitImageResults returns every image DuckDuckGo includes on the page. Safe-search mode is plugin-only (default moderate).
Fetches a URL and returns its title, first-level headings (h1/h2/h3), and a readable content excerpt produced by @mozilla/readability. The content is returned as Markdown by default — headings, lists, inline links, and inline images are preserved — or as plain text when the plugin's contentFormat setting is switched to "text". Use the View Images tool to download any images of interest.
| Parameter | Type | Notes |
|---|---|---|
url | URL | Required. |
findInPage | string[] | Optional search terms that bias which content slices are returned when the page exceeds the character budget. Strongly recommended. |
Both the visible-text character budget (contentLimit) and the output format (contentFormat) are plugin-only settings — they are not exposed as tool parameters so the model cannot override the user-set or default values. The response includes contentLength, the character count of the full extracted content prior to truncation. When contentLength > content.length the content was truncated — refine findInPage and re-call, or raise contentLimit in the plugin settings.
Downloads images from an explicit URL list, from images scraped off a page, or both. Provide at least one of imageURLs or websiteURL.
| Parameter | Type | Notes |
|---|---|---|
imageURLs | URL[] | Optional explicit list to download. |
websiteURL | URL | Optional page to scrape for images. |
maxImages | int 1–200 | Optional; caps how many scraped images are downloaded when websiteURL is supplied. |
Returns an array of per-image records:
{ "filename": "typescript.svg", "alt": "TypeScript logo", "title": "Click to enlarge", "image": "" }
When a download fails, the record carries an error field in place of image. alt and title are populated from the source page's <img> attributes when images are scraped via websiteURL; explicit imageURLs arrive without that metadata and surface both fields as empty strings.
Install the plugin through LM Studio's plugin browser. Once enabled, the four tools become available to any model that supports tool calls.
Requires Node.js ≥ 22 (for fetch and AbortSignal.any).
git clone https://github.com/packern/web-tools.git cd web-tools npm install npm run dev # runs `lms dev`
npm run push publishes the plugin to the LM Studio Hub (lms push).
All fields are exposed in the LM Studio plugin UI. Two sentinel conventions apply: -1 means "use the built-in default" and — for the knobs where it makes sense — 0 means "disabled / no limit / no delay".
| Field | Range | Default | Purpose |
|---|---|---|---|
| Search Results Per Page | 0–10 | 0 → 5 | Page size for web and image search. 0 lets the assistant decide. |
| Safe Search | strict / moderate / off / Auto | Auto → moderate | DuckDuckGo safe-search mode. |
| View Images: Max Images | -1–200 | -1 → 10 | Maximum images scraped when View Images receives a websiteURL. |
| Visit Website: Content Character Limit | 0–100 000 | 0 → 10 000 | Visible-text character budget for the page excerpt. |
| Visit Website: Content Format | Markdown / Plain text | Markdown | Output format of the content field. Markdown retains headings, lists, and inline links; Plain text strips syntax and preserves only line breaks. |
| Search Cache TTL | -1–3600 s | -1 → 900 s (15 min) | How long web/image search results stay cached. 0 disables. |
| VQD Token Cache TTL | -1–3600 s | -1 → 600 s (10 min) | How long DuckDuckGo VQD tokens stay cached. 0 disables. |
| Website Cache TTL | -1–3600 s | -1 → 600 s (10 min) | How long fetched HTML stays cached. 0 disables. |
| Min Interval Between Requests | -1–30 s | -1 → 5 s | Minimum gap between outbound DuckDuckGo requests. 0 disables rate limiting. |
| Max Retries Per Request | -1–10 | -1 → 3 | Retry attempts (after the first try) for every outbound request. 0 disables retries. |
| Retry Initial Backoff | -1–30 s | -1 → 1 s | Base backoff before the first retry. 0 = no delay. |
| Retry Max Backoff | -1–300 s | -1 → 30 s | Upper bound on exponential-backoff delay. 0 = no delay. |
| VQD to Image API Delay | -1–10 s | -1 → 2 s | Delay inserted between the VQD scrape and the image-search API call. 0 = no delay. |
Three disk-backed cacache stores, all persisted under ~/.lmstudio/plugin-data/lms-plugin-duckduckgo-cache/:
searchCacheTtlSeconds).vqdCacheTtlSeconds).websiteCacheTtlSeconds).Caches survive plugin reloads. To fully clear them, stop LM Studio and delete the cache directory above.
A shared bottleneck-backed rate limiter enforces a 5-second minimum gap between outbound DuckDuckGo requests by default. Every outbound HTTP request is wrapped by p-retry with randomized exponential backoff (factor 2) — three attempts with a 1-second to 30-second window by default. All four knobs are configurable; set the interval or retry count to 0 to disable either behavior.
With the plugin enabled you can explicitly ask the assistant to search the web, fetch images, visit a page, or view images, but you can also just ask a question whose answer requires the web and the assistant will pick the right tool on its own.
npm run dev — run the plugin in LM Studio dev mode (lms dev).npm run push — publish to the LM Studio Hub (lms push).npm run lint / npm run lint:fix — ESLint on src/**/*.ts.npm run format / npm run format:check — Prettier.npm run knip — dead-code / unused-export check.A local pre-commit hook at .git/hooks/pre-commit can run lint, format:check, and knip sequentially and abort the commit on any failure. The hook is not committed to the repo — fresh clones need to reinstall it. Bypass with git commit --no-verify when necessary.
No test suite is configured.
Built on top of Daniel Sig's original
lms-plugin-duckduckgo and
lms-plugin-visit-website plugins, now
merged into a single tool suite and extended with the View Images tool, Markdown-formatted
website content, findInPage-biased content slicing, configurable rate limiting and retry,
and a persistent cacache-backed store for VQD tokens, search results, and fetched HTML.
MIT — see LICENSE.
