`groq://`¶

Groq — extremely fast LPU-hosted inference for open-source models (Llama 3.3, Mixtral, Whisper, Gemma) via an OpenAI-compatible API.

Built on pkg/component/llmutil/ — the same chat/embeddings shape as openai://. This page covers only the Groq-specific defaults.

URI shape¶

groq://chat?model=<id>&apiKey=<key>&temperature=<float>&maxTokens=<int>&baseURL=<url>

Parameters¶

Param	Type	Default	Producer/Consumer/Both	Description
path	enum	`chat`	Producer	`chat` (embeddings not currently offered by Groq).
`model`	string	`llama-3.3-70b-versatile`	Producer	Groq model id. Common: `llama-3.3-70b-versatile`, `llama-3.1-8b-instant`, `mixtral-8x7b-32768`, `gemma2-9b-it`.
`apiKey`	string	(from constructor)	Producer	Groq API key. Sent as `Authorization: Bearer <key>`.
`baseURL`	string	`https://api.groq.com/openai/v1`	Producer	Override for proxies.
`temperature`	float	`0.7`	Producer	Sampling temperature.
`maxTokens`	int	`4096`	Producer	Output cap.

Producer-only. Default timeout 120 s.

Headers¶

Same as openai:// but under groq.* (groq.model, groq.id, groq.finishReason, groq.usage.*). Input header systemPrompt works the same way.

Examples¶

- to: groq://chat?model=llama-3.3-70b-versatile&apiKey=${GROQ_API_KEY}

Smaller, faster model for routing decisions:

- to: groq://chat?model=llama-3.1-8b-instant&apiKey=${GROQ_API_KEY}&maxTokens=128&temperature=0

Behavior notes¶

Groq is very fast — sub-second first-token latency is common — so the 120 s timeout is generally overkill but harmless.
The Groq model catalogue moves often; check https://console.groq.com/docs/models for what's currently available. Older Llama 3.x ids get retired.
All other behaviour matches openai://.

Source¶

pkg/component/groq/groq.go — eight-line wrapper around llmutil.NewComponent.
pkg/component/llmutil/llmutil.go.

groq://¶