Integrations / Provider Guides

OpenAI

Connect Aurora to OpenAI for GPT, o-series reasoning, and embedding models.

Browse docs

Use this provider to route to OpenAI's hosted models. Aurora speaks the OpenAI wire format natively, so requests to https://api.openai.com go through without translation.

Activate

text

OPENAI_API_KEY=sk-...

Restart Aurora. The openai provider appears at /v1/models with the current GPT, o-series, and embedding catalog.

Default models

Model	Use
`gpt-4o`	General chat, function calling, vision
`gpt-4o-mini`	Cheap general chat
`o3` / `o3-mini`	Reasoning models — Aurora translates `max_tokens` to `max_completion_tokens` for you
`text-embedding-3-small`	Embeddings, 1536 dimensions
`text-embedding-3-large`	Embeddings, 3072 dimensions

Override the catalog with OPENAI_MODELS=gpt-4o,gpt-4o-mini,o3-mini or pull the latest via aurora models sync.

Multiple instances

Run separate OpenAI providers for quota, region, or billing:

text

OPENAI_EAST_API_KEY=sk-...
OPENAI_WEST_API_KEY=sk-...

Aurora exposes them as openai-east and openai-west. Use them in aliases, pools, or workflows.

Custom base URL

text

OPENAI_BASE_URL=https://my-corp-openai-proxy.example.com/v1

Use this for corporate proxies, regional endpoints (e.g. Azure-hosted OpenAI), or any OpenAI-compatible gateway.

Gotchas

Reasoning models — o3 and o1 require max_completion_tokens rather than max_tokens. Aurora does this translation automatically. Streaming is supported; reasoning tokens are surfaced as reasoning content blocks in the Anthropic ingress and as reasoning_content deltas in the OpenAI format.
Embeddings — text-embedding-3-* accepts a dimensions parameter to truncate. Aurora forwards it.
Function calling — full support, including parallel tool calls. No provider-specific quirks.
Prompt caching — OpenAI's automatic prompt caching is enabled by default on supported models. Aurora forwards cache_control and surfaces cache-hit tokens in the response usage.

Sample request

text

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AURORA_MASTER_KEY" \
  -d '{
    "model": "openai/gpt-4o-mini",
    "messages": [{"role": "user", "content": "Hello from Aurora"}]
  }'

Providers Overview — full 14-provider matrix
Model Management — aliases, overrides, allowlists
Aliases — give your app a stable model name

Default models

Model	Use
`gpt-4o`	General chat, function calling, vision
`gpt-4o-mini`	Cheap general chat
`o3` / `o3-mini`	Reasoning models — Aurora translates `max_tokens` to `max_completion_tokens` for you
`text-embedding-3-small`	Embeddings, 1536 dimensions
`text-embedding-3-large`	Embeddings, 3072 dimensions

Override the catalog with OPENAI_MODELS=gpt-4o,gpt-4o-mini,o3-mini or pull the latest via aurora models sync.

Gotchas

Reasoning models — o3 and o1 require max_completion_tokens rather than max_tokens. Aurora does this translation automatically. Streaming is supported; reasoning tokens are surfaced as reasoning content blocks in the Anthropic ingress and as reasoning_content deltas in the OpenAI format.

Embeddings — text-embedding-3-* accepts a dimensions parameter to truncate. Aurora forwards it.

Function calling — full support, including parallel tool calls. No provider-specific quirks.

Prompt caching — OpenAI's automatic prompt caching is enabled by default on supported models. Aurora forwards cache_control and surfaces cache-hit tokens in the response usage.

curl http://localhost:8080/v1/chat/completions \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $AURORA_MASTER_KEY" \ -d '{ "model": "openai/gpt-4o-mini", "messages": [{"role": "user", "content": "Hello from Aurora"}] }'

OpenAI

Activate

Default models

Multiple instances

Custom base URL

Gotchas

Sample request

Related

OpenAI

Activate

Default models

Multiple instances

Custom base URL

Gotchas

Sample request

Related