Multimodal search foundation
How to use
Jina AI.
Building multilingual or multimodal semantic search, retrieval-augmented generation, relevance reranking, classification, and web-to-LLM ingestion workflows.
Jina AI is a Search foundation platform for multilingual and multimodal embeddings, reranking, classification, web reading, and deep search APIs. Building multilingual or multimodal semantic search, retrieval-augmented generation, relevance reranking, classification, and web-to-LLM ingestion workflows. This guide covers the whole path in one place: official access, a first session that produces something reviewable, the checks that make output trustworthy, and the permissions worth limiting before you connect real work.
AI should make the work easier to inspect. If the workflow removes the source, the owner, or the review step, redesign the workflow.
Access & setup
Find, install, and sign in to Jina AI
Get into the official Jina AI experience with the right account and a setup you understand.
- 01
Start at https://jina.ai/ and confirm the domain before entering account or payment information.
- 02
Availability: Jina AI is a browser-managed API platform accessed through HTTPS endpoints and compatible SDKs; selected models can also be evaluated or deployed under their individual licenses.
- 03
A Jina API key for authenticated use, a selected model and license, a vector store or search stack where needed, authorized source data, relevance evaluations, and server-side secret handling.
- 04
Sign in with the account you intend to keep using, then review plan, data, notification, and permission settings.
- 05
Run one low-risk test task before connecting sensitive files, repositories, or workspace data.
- Use official download pages.
- Review permissions during setup.
- Keep installers and applications updated.
First session
Your first useful Jina AI session
Learn the interaction loop using a small task with a clear outcome.
- 01
Define the users, corpus, languages, modalities, query types, relevance labels, freshness, privacy class, latency, recall, precision, and cost targets.
- 02
Choose an embedding model for candidate retrieval and a reranker for final ordering, recording dimensions, context, license, task mode, and model version.
- 03
Create a versioned evaluation set with difficult positives and negatives, index authorized content, and run retrieval and reranking as separate measurable stages.
- 04
Review relevance, missing sources, stale pages, language balance, prompt injection, citations, latency, and cost before integrating results into any answer or action.
- State the outcome before the background.
- Provide the real source material.
- Review the result before expanding the task.
Quality control
Check the quality of Jina AI output
Establish that a retrieved answer is actually supported by the sources it cites.
- 01
Restate the question precisely; retrieval quality degrades fastest on vague or compound queries.
- 02
Open the cited sources and confirm each one genuinely contains the claim attributed to it.
- 03
Check the publication and capture dates — stale sources are the most common cause of confidently wrong answers.
- 04
Look for what is missing: retrieval returns what matches the query, not what would contradict it.
- 05
Confirm quotations are verbatim and that summaries have not reversed a qualification or a negation.
- Open the citation; do not trust that a link supports the claim beside it.
- Check dates on every retrieved source.
- Treat an absent contradiction as unverified, not as agreement.
Privacy & permissions
Use Jina AI safely
Control what you index and what leaves your systems when you query it.
- 01
Classify the corpus before indexing: confidential, personal, and regulated material needs an explicit decision.
- 02
Confirm where the index is stored, who can query it, and how long content is retained.
- 03
Scope crawling and scraping to sources you are permitted to collect, and respect robots and terms of use.
- 04
Strip secrets and personal data before indexing rather than filtering them at query time.
- 05
Review query logs — the questions asked can be as sensitive as the documents indexed.
- Embeddings and rerankers produce similarity and relevance estimates, not verified truth. Model versions, dimensions, licenses, context limits, rate limits, and task-specific prefixes differ, and changing a model normally requires re-embedding and reevaluation. Reader and search output can contain hostile page instructions, copyrighted text, personal data, and stale facts. Keep keys server-side, respect source access rules, minimize stored content, and preserve citations and human review.
- Follow your organisation's approved-use policy.
- Never treat fluent output as authorization to act.
Core workflows
Step-by-step ways to use Jina AI for the work it does best.
Each workflow below has its own outcome, steps and review checkpoints.
Combine embeddings and reranking with an explicit relevance benchmark, source controls, and citation-preserving answer generation.
↓ 02 WorkflowCreate a controlled web-research collector with Jina ReaderConvert permitted web pages into reviewable evidence without turning page text, stale results, or hostile instructions into trusted commands.
↓Workflow 01
Build a multilingual retrieval pipeline with Jina AI
Combine embeddings and reranking with an explicit relevance benchmark, source controls, and citation-preserving answer generation.
Outcome: Return a small, well-ordered evidence set that measurably serves real queries across required languages.
- 01
Define users, languages, corpus, source authority, chunking, freshness, access rules, query classes, relevance labels, latency, recall, precision, and cost targets.
- 02
Choose a compatible embedding and reranker, record model versions, dimensions, context, task settings, license, and build a hard evaluation set with judged positives and negatives.
- 03
Normalize and chunk authorized documents without losing titles, URLs, dates, permissions, or section boundaries, then embed and index versioned content.
- 04
Retrieve a broad candidate set and rerank a smaller set, measuring each stage separately for language balance, false positives, missing evidence, latency, and cost.
- 05
Pass only cited excerpts to generation, treat retrieved text as untrusted data, enforce access at query time, and reevaluate before any model, chunking, or corpus change.
What good use looks like
- Measure retrieval and reranking separately.
- Preserve source and permission metadata.
- Re-embed after model changes.
Workflow 02
Create a controlled web-research collector with Jina Reader
Convert permitted web pages into reviewable evidence without turning page text, stale results, or hostile instructions into trusted commands.
Outcome: Collect a bounded, cited research packet whose access, freshness, and factual claims can be verified.
- 01
Define the research question, allowed domains, time window, jurisdictions, source hierarchy, robots and terms requirements, personal-data rules, and stopping criteria.
- 02
Keep the Jina key server-side, validate every requested URL, block private networks and redirects, set size, timeout, rate, and budget limits, and log source metadata.
- 03
Use search for discovery and Reader for selected pages, storing the original URL, title, retrieval time, relevant excerpt, and conversion warnings rather than an uncited text dump.
- 04
Ignore instructions embedded in pages, cross-check consequential claims with primary sources, detect duplicate or stale material, and review copyright and personal-data exposure.
- 05
Produce a cited packet with confidence and gaps, delete unnecessary page content on schedule, test blocked pages and parser failures, and require human approval before downstream publication.
What good use looks like
- Allowlist destinations and block private IPs.
- Treat page text as untrusted input.
- Keep URL and retrieval time with every excerpt.
Official references
Check the current product documentation.
Features, plan limits, availability, and data controls change. These official pages are the starting points used for this guide.