Jina

Jina is not a first-class provider type in Aurora — it is reached through the OpenAI-compatible provider with a special base URL. The gateway recognizes the jina.ai host and applies a Jina-specific model-name transform that strips the jina-ai/ namespace from model IDs before the upstream call.

This page shows the activation pattern. For the provider matrix, see Providers Overview; for the Jina-specific behavior, see the OpenAI guide (the base-URL and override sections).

Activate

# This is an OpenAI-compatible provider pointing at Jina, not a Jina provider
OPENAI_API_KEY=jina_...
OPENAI_BASE_URL=https://api.jina.ai

Aurora auto-discovers an openai provider. The dashboard Providers page will show the host as https://api.jina.ai and the gateway will strip the jina-ai/ namespace from any model IDs it forwards to Jina.

Default models

Override the catalog with OPENAI_MODELS=jina-embeddings-v3,jina-reranker-v2-base-multilingual and set CONFIGURED_PROVIDER_MODELS_MODE=allowlist if you want to restrict to that list.

Multiple instances

You can run two Jina accounts side by side by creating two OpenAI-compatible providers:

OPENAI_JINA_PROD_API_KEY=jina_...
OPENAI_JINA_PROD_BASE_URL=https://api.jina.ai
OPENAI_JINA_STAGING_API_KEY=jina_...
OPENAI_JINA_STAGING_BASE_URL=https://api.jina.ai

Aurora exposes them as openai-jina-prod and openai-jina-staging. Use them in pools for high-throughput retrieval, or in workflows to control cache/audit per environment.

Embeddings

curl http://localhost:8080/v1/embeddings \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AURORA_MASTER_KEY" \
  -d '{
    "model": "openai/jina-embeddings-v3",
    "input": ["Hello from Aurora"]
  }'

openai here is the provider instance name; jina-embeddings-v3 is the model ID. The jina-ai/ namespace is stripped before the call.

Rerank

curl http://localhost:8080/v1/rerank \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AURORA_MASTER_KEY" \
  -d '{
    "model": "openai/jina-reranker-v2-base-multilingual",
    "query": "machine learning",
    "documents": [
      "Aurora is an AI gateway.",
      "Pizza is a popular Italian food.",
      "Neural networks learn from data."
    ]
  }'

See API: Rerank for the full request/response shape.

Gotchas

• Provider name is openai, not jina. The provider/model selector you pass to Aurora is openai/jina-embeddings-v3, not jina/.... If you have multiple OpenAI providers, use their instance names — openai-jina-prod, openai-jina-staging, etc.
• The transform is automatic. Aurora checks the configured OPENAI_BASE_URL and applies StripJinaNamespace when it contains jina.ai. If you point OPENAI_BASE_URL somewhere else, the transform does not apply.
• Dimensions — jina-embeddings-v3 accepts a dimensions parameter (default 1024). Aurora forwards it.
• Task hint — jina-embeddings-v3 accepts a task parameter (retrieval.query, retrieval.passage, classification, etc.) for task-specific embeddings. Aurora forwards it.
• Pooling — Jina embeddings are best used in a provider pool for high-throughput retrieval. Pool the same Jina account across multiple keys, or pair with Cohere or vLLM for fallback.
• Rerank model in /v1/models — rerank models are listed alongside chat/embedding models. They are reachable only via /v1/rerank; calls to /v1/chat/completions against a rerank model will return a 400.
• Prompt caching — Jina does not implement prompt caching on the embedding/rerank APIs. Use Aurora's semantic cache if you need replay.

Sample workflow: RAG with Jina

# 1) Embed your query
curl http://localhost:8080/v1/embeddings \
  -H "Authorization: Bearer $AURORA_MASTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"openai/jina-embeddings-v3","input":"How does Aurora handle failover?"}'
 
# 2) Use your vector store to find top candidates, then rerank them
curl http://localhost:8080/v1/rerank \
  -H "Authorization: Bearer $AURORA_MASTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model":"openai/jina-reranker-v2-base-multilingual",
    "query":"How does Aurora handle failover?",
    "documents":["..."]
  }'

Related

• API: Rerank
• API: Embeddings
• Provider Pools — pool Jina across multiple keys
• Providers Overview — see the 14-provider matrix