Jina AI

Create embeddings, rerank search results, and perform neural search with Jina AI.

AI and MLAPI key5 actionsAll integrations

On this page

Connect
Call it from a workflow
Actions (5)
classify_content
deepsearch_query
extract_webpage_content
train_custom_classifier
web_search_summarization
Details

Connect

Auth type: API key (secret_text). Paste the API key (or token) from the provider's settings into the connect dialog. It is encrypted at rest and only decrypted on the server when an action runs.

In the editor: Backend panel, Integrations tab, then Connect. This connects it for the project.
For every project in a workspace: Workspace settings, Integrations.
Owners and admins choose which integrations members may connect in the admin panel (Integrations tab), and can attach their own OAuth app there.

Call it from a workflow

Pass the action name and its input. The call resolves with { success, output, error? }; a missing connection resolves with success false and the reason. See ctx.integrations.invoke and ctx.integrations.call.

Native workflow

// .workflow.js (native ctx)
export async function run(ctx) {
  const res = await ctx.integrations.invoke("jina_ai", "classify_content", {
    "model": "jina-clip-v2",
    "input": "<text>",
    "labels": []
  });
  if (!res.success) throw new Error(res.error);
  ctx.log.info("Jina AI answered", { output: res.output });
}

defineWorkflow module

// defineWorkflow module (@app/sdk ctx)
import { defineWorkflow } from "@app/sdk";

export default defineWorkflow({
  id: "jina-ai-classify-content",
  name: "Jina AI: Classify Text or Image",
  description: "Calls Jina AI when a row is created.",
  trigger: { type: "entity_event", entity: "tickets", event: "created" },
  async run(ctx, event) {
    const res = await ctx.integrations.call("jina_ai", {
      "action": "classify_content",
      "model": "jina-clip-v2",
      "input": "<text>",
      "labels": []
    }) as { success: boolean; error?: string };
    if (!res.success) throw new Error(res.error);
  },
});

From an app (connector proxy)

curl -X POST "$APP_ORIGIN/__appbricx/connector-proxy/jina_ai/classify_content" \
  -H "Authorization: Bearer $APP_KEY" \
  -H "content-type: application/json" \
  -d '{"props":{"model":"jina-clip-v2","input":"<text>","labels":[]}}'

Actions

classify_contentin catalog

Classify Text or Image. Assign categories to text or images using the Classifier API (zero-shot/few-shot).

Input	Type	Description
`model`required	one of "jina-clip-v2" \| "jina-embeddings-v3" \| "jina-clip-v1"	The model to use for classification.
`input`required	long text	Text or image URL to classify. URLs will be treated as images, other strings as text.
`labels`required	array	The labels to classify the content into.

deepsearch_queryin catalog

DeepSearch Query. Answer complex questions through iterative search, reading, and reasoning with the DeepSearch API.

Input	Type	Description
`model`required	one of "jina-deepsearch-v1"	Model
`prompt`required	long text	Prompt
`reasoning_effort`optional	one of "low" \| "medium" \| "high"	Constrains effort on reasoning for reasoning models. Reducing reasoning effort can result in faster responses and fewer tokens used on reasoning in a response.
`budget_tokens`optional	number	This determines the maximum number of tokens allowed for DeepSearch process. Larger budgets can improve response quality by enabling more exhaustive search for complex queries.
`max_attempts`optional	number	The maximum number of retries for solving a problem in DeepSearch process. A larger value allows DeepSearch to retry solving the problem by using different reasoning approaches and strategies.
`no_direct_answer`optional	boolean	Forces the model to take further thinking/search steps even when the query seems trivial. Useful if you're using DeepSearch in scenarios where you're certain the query always needs DeepSearch, rather…
`max_returned_urls`optional	number	The maximum number of URLs to include in the final answer/chunk. URLs are sorted by relevance and other important factors.
`response_format`optional	JSON	JSON schema for structured output format. Example: { "type": "json_schema", "json_schema": { "type": "object", "properties": { "numerical_answer_only": { "type": "number" } } } }
`boost_hostnames`optional	long text	A list of domains that are given a higher priority for content retrieval. Useful for domain-specific, high-quality sources that provide valuable content.
`bad_hostnames`optional	long text	A list of domains to be strictly excluded from content retrieval. Typically used to filter out known spam, low-quality, or irrelevant websites.
`only_hostnames`optional	long text	A list of domains to be exclusively included in content retrieval. All other domains will be ignored. Useful for domain-specific searches.

extract_webpage_contentin catalog

Extract Webpage Content. Convert a URL into clean, LLM-friendly Markdown using the Reader API.

Input	Type	Description
`url`required	text	The URL of the webpage to extract content from.
`format`required	one of "default" \| "markdown" \| "html" \| "text" \| "screenshot" \| "pageshot"	Format of the extracted content - how the webpage should be returned.
`remove_all_images`optional	boolean	Remove all images from the response.
`links_summary`optional	one of "none" \| "true" \| "all"	Create a "Buttons & Links" section at the end to help downstream LLMs or web agents navigate the page.
`images_summary`optional	one of "none" \| "true" \| "all"	Create an "Images" section at the end, giving downstream LLMs an overview of all visuals on the page.
`do_not_track`optional	boolean	When enabled, the requested URL won't be cached and tracked on our server.
`iframe_extraction`optional	boolean	Processes content from all embedded iframes in the DOM tree.
`shadow_dom_extraction`optional	boolean	Extracts content from all Shadow DOM roots in the document.
`follow_redirect`optional	boolean	Choose whether to resolve to the final destination URL after following all redirects. Enable to follow the full redirect chain.
`eu_compliance`optional	boolean	All infrastructure and data processing operations reside entirely within EU jurisdiction.
`json_response`optional	boolean	Return response in JSON format containing URL, title, content, and timestamp (if available).
`timeout_seconds`optional	number	Maximum page load wait time in seconds (0 means use default timeout, set a value if the default browser engine is too slow on simple webpages).
`css_selector`optional	long text	List of CSS selectors to target specific page elements (e.g., "body, .main-content, #article").
`wait_for_selector`optional	long text	CSS selectors to wait for before returning results (e.g., "#dynamic-content, .lazy-loaded").
`exclude_selector`optional	long text	CSS selectors for elements to remove from the results (e.g., "header, footer, .ads, #sidebar").

train_custom_classifierin catalog

Train Custom Classifier. Fine-tune a classifier with labeled examples for domain-specific tasks.

Input	Type	Description
`model`required	one of "jina-clip-v2" \| "jina-embeddings-v3" \| "jina-clip-v1"	The base model to use for training.
`access`required	one of "private" \| "public"	Visibility of the trained model.
`num_iters`optional	number	Number of training iterations to perform.
`training_data`required	array	Training Data

web_search_summarizationin catalog

Web Search Summarization. Perform a web search and retrieve summarized results using the Reader API.

Input	Type	Description
`query`required	text	The search query to perform.
`read_full_content`optional	boolean	Visit every URL in the search result and return the full content using Reader. Toggle on to enable more Reader-specific options.
`json_response`optional	boolean	The response will be in JSON format, containing the URL, title, content, and timestamp (if available). In Search mode, it returns a list of five entries, each following the described JSON structure.
`fetch_favicons`optional	boolean	This will fetch the favicon of each URL in the SERP and include them in the response as image URI, useful for UI rendering.
`preferred_country`optional	one of "Default" \| "US" \| "CA" \| "MX" \| "GB" \| "DE" \| "FR" \| "JP" \| "CN" \| "IN" \| "BR" \| "AU"	The country to use for the search. It's a two-letter country code.
`preferred_location`optional	text	From where you want the search query to originate. It is recommended to specify location at the city level in order to simulate a real user's search.
`preferred_language`optional	one of "Default" \| "en" \| "es" \| "fr" \| "de" \| "ja" \| "zh-cn" \| "hi" \| "pt" \| "it" \| "ko" \| "nl"	The language to use for the search. It's a two-letter language code.
`pagination`optional	number	The result offset. It skips the given number of results. It's used for pagination.
`in_site_search`optional	text	Returns the search results only from the specified website or domain. By default it searches the entire web.
`bypass_cached_content`optional	boolean	Our API caches URL contents for a certain amount of time. Set it to true to ignore the cached result and fetch the content from the URL directly.
`eu_compliance`optional	boolean	All infrastructure and data processing operations reside entirely within EU jurisdiction.

Details

Id: jina_ai
Category: AI and ML
Tags: embeddings, reranking, search, ai, neural-search
Package: @activepieces/piece-jina-ai@0.1.3
Tier: Community