Home

Deploy Jina AI's embedding, reranker, and reader models inside customer environments that cannot reach the internet.

For sales/SA/field engineers walking a customer through their first deployment, start with Why Air-Gap, then Quick Start. For developers integrating the API into an application, start with API Reference.

At a glance

    PHASE 1 (network)                              PHASE 2 (offline)
    ─────────────────                              ─────────────────

  ┌────────────────┐    USB / SCP / disk    ┌────────────────┐    port 8080
  │ connected host │   ──────────────────►  │ air-gapped host│   ──────►  app
  │                │       .tar.gz          │                │
  │ jina-on-prem.py │                        │  docker load   │  OpenAI / Cohere
  │   bundle       │                        │  docker run    │  Gemini / Voyage
  └────────────────┘                        └────────────────┘
        │                                          │
        ▼                                          ▼
  download weights                          serve embeddings,
  + deps from HF Hub                        reranking, readers
  docker build                              zero outbound calls

That's the whole product. The connected machine has internet to fetch model weights and dependencies. Everything is baked into a single Docker image and exported as a .tar.gz. The offline machine only needs Docker.

What's supported

• 28 models: Jina embeddings (v5, v4, v3, v2), rerankers, ColBERT, CLIP, ReaderLM, VLM. See Model Catalog.
• 4 API schemas simultaneously: OpenAI, Cohere, Google Gemini, Voyage AI - drop-in for any client.
• Multimodal: text + image + audio + video on omni/clip/v4 models.
• GPU and CPU: same model can be packaged either way.
• Elasticsearch inference service: works as a service: openai endpoint out of the box.

Pick your starting point

You are...	Start here
An SA/sales engineer evaluating jina-on-prem for a customer	Why Air-Gap, then Customer Scenarios
Comparing this against Ollama / vLLM / ONNX / hosted API	Comparison vs alternatives
A field engineer deploying at a customer site	Quick Start, then Sizing & Hardware
A developer integrating the API	API Reference
Building a new bundle from scratch	Bundling Guide
Rolling out a new model version	Versioning & Updates
Hitting an error	Troubleshooting, FAQ

License note

Most Jina v5/v4/v3 models are CC-BY-NC-4.0: commercial use needs a license. Contact Elastic sales. v2 and v1 models are Apache-2.0 and free for any use. Per-model license is in the Model Catalog.