Integrations / Provider Guides

Groq

Connect Aurora to Groq's hosted Llama, Qwen, and Whisper models with very low latency.

Browse docs

Use this provider to route to Groq's LPU-hosted open models. Groq is a fast OpenAI-compatible host; Aurora talks to it through the standard OpenAI endpoint with no special handling.

Activate

text

GROQ_API_KEY=gsk_...

Restart Aurora. The groq provider appears at /v1/models with Groq's full catalog.

Default models

Model	Use
`llama-3.3-70b-versatile`	General chat, function calling
`llama-3.1-8b-instant`	Cheapest, fastest Llama
`qwen-qwq-32b`	Qwen reasoning model
`whisper-large-v3-turbo`	Audio transcription

Override with GROQ_MODELS=llama-3.3-70b-versatile,llama-3.1-8b-instant.

Multiple instances

text

GROQ_PROD_API_KEY=gsk_...
GROQ_STAGING_API_KEY=gsk_...

Aurora exposes them as groq-prod and groq-staging.

Custom base URL

text

GROQ_BASE_URL=https://api.groq.com/openai/v1

Almost never changed; useful for proxies.

Gotchas

Rate limits — Groq's free tier is aggressive. Use Managed API Keys and Workflows to share quota across your team fairly.
Reasoning models — qwen-qwq-32b returns reasoning content as separate content blocks. Aurora surfaces them in the OpenAI-shaped response.
Audio — Groq Whisper is exposed via the OpenAI /v1/audio namespace, which is not in Aurora's translated path. Use provider passthrough at /p/groq/v1/audio/* to call it natively.
Function calling — full support, no quirks.
Prompt caching — Groq does not implement server-side prompt caching. Use Aurora's semantic cache if you need replay.

Sample request

text

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AURORA_MASTER_KEY" \
  -d '{
    "model": "groq/llama-3.3-70b-versatile",
    "messages": [{"role": "user", "content": "Hello from Aurora"}]
  }'

Providers Overview
Failover — combine Groq with another provider for high availability

Model

Use

llama-3.3-70b-versatile

General chat, function calling

llama-3.1-8b-instant

Cheapest, fastest Llama

qwen-qwq-32b

Qwen reasoning model

whisper-large-v3-turbo

Audio transcription

Gotchas

Rate limits — Groq's free tier is aggressive. Use Managed API Keys and Workflows to share quota across your team fairly.

Reasoning models — qwen-qwq-32b returns reasoning content as separate content blocks. Aurora surfaces them in the OpenAI-shaped response.

Audio — Groq Whisper is exposed via the OpenAI /v1/audio namespace, which is not in Aurora's translated path. Use provider passthrough at /p/groq/v1/audio/* to call it natively.

Function calling — full support, no quirks.

Prompt caching — Groq does not implement server-side prompt caching. Use Aurora's semantic cache if you need replay.

curl http://localhost:8080/v1/chat/completions \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $AURORA_MASTER_KEY" \ -d '{ "model": "groq/llama-3.3-70b-versatile", "messages": [{"role": "user", "content": "Hello from Aurora"}] }'

Groq

Activate

Default models

Multiple instances

Custom base URL

Gotchas

Sample request

Related

Groq

Activate

Default models

Multiple instances

Custom base URL

Gotchas

Sample request

Related