Jina AI
AIBuild search and RAG workflows with Jina AI: embeddings, reranking, web-page reading, web search, and zero-shot classification through one search-foundation connector.
What This Integration Enables
Jina covers retrieval as a pipeline. Web search finds sources, the Reader turns any URL into clean, model-ready text, embeddings vectorize content, reranking orders passages by true relevance, and zero-shot classification labels text without training. Together they are the ingredients of a grounded answer. The Reader is the quiet workhorse: it removes the navigation, ads, and boilerplate that make raw scraped pages useless to a model. But retrieval only becomes trustworthy when a person owns the answers that carry weight. An orchestration layer runs the search-read-rerank pipeline and decides which briefs go out automatically and which get a human check. FlowRunner is built for that layer.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Automated Research Briefs
An analyst requests a briefing on a company. The agent runs a web search, reads each result into clean text with the Reader, reranks the passages by relevance to the question, and assembles a sourced brief. The analyst gets a draft with links to every source and reviews it before use. The hours of opening and skimming sources collapse into a checkable draft.
Grounded Q&A Over the Web
A question needs an answer grounded in current public information. The agent searches, reads the top sources, reranks, and generates an answer citing the passages it used. Confident, well-sourced answers return directly; questions where the sources conflict or the retrieval is thin are routed to a person rather than answered on shaky ground.
Content Classification at Ingest
Incoming content needs labels without a trained classifier. The agent uses zero-shot classification to tag each item against your label set. Confident labels are applied automatically; ambiguous items are held for a person, so the taxonomy stays clean as new content arrives.
Human-in-Loop Highlight
A research brief assembled from web sources is a starting point a person should verify, not a finished fact. When Jina produces a brief or a grounded answer that will be acted on, FlowRunner routes it through a [human-in-the-loop](/concepts/human-in-the-loop/) step: the agent pauses, presents the answer with every source link, and sends it to the analyst via Slack. They verify the sources and confirm. The pipeline gathers and orders the evidence; a person owns the conclusion.
Agent Capabilities
7 actionsEmbeddings
1- Create Embeddings Generates vector embeddings for one or more input texts using Jina's embedding models (default jina-embeddings-v3).
Search & Ranking
1- Rerank Documents Reorders a set of candidate documents by semantic relevance to a query using Jina's reranker models (default jina-reranker-v2-base-multilingual).
Reader
3- Read URL Fetches a web page through Jina Reader and returns clean, LLM-ready content (markdown by default).
- Search Web Runs a web search through Jina Search and returns the top results already fetched and cleaned into LLM-ready content, so each result carries usable page text rather than just a snippet.
- Deep Search Runs an agentic deep-research search with jina-deepsearch-v1. Given a conversation (messages), it iteratively searches the web, reads pages, and reasons to produce a grounded, cited answer to complex questions.
Classification
2- Classify Texts Performs zero-shot classification of one or more input texts against a set of candidate labels using a Jina embedding or classifier model.
- Segment Text Tokenizes and splits long text into smaller chunks using Jina Segmenter. Returns the total token count and, when requested, the list of text chunks bounded by a maximum chunk length.
Start building with Jina AI
$100 in credits. No card required. Connect in minutes.