LLM

Groq

Groq provides ultra-fast LLM inference via its custom hardware. Average latency: ~443ms — significantly faster than cloud LLM providers.

Requires: Groq API key in Settings → Credentials

Models

Model	Best For
`meta-llama/llama-4-scout-17b-16e-instruct`	Recommended — fast, capable
`llama-3.1-8b-instant`	Ultra-fast, compact
`openai/gpt-oss-20b`	Reasoning tasks (supports `reasoning_effort`)

Configuration


Code
{
  "brain": {
    "provider": "groq",
    "model": "meta-llama/llama-4-scout-17b-16e-instruct",
    "instructions": "You are a helpful assistant.",
    "system_timezone": "UTC",
    "temperature": 0.8,
    "max_completion_tokens": 4096,
    "parallel_tool_calls": true,
    "tool_choice": "auto",
    "max_tool_steps": 3
  }
}

Fields

Field	Type	Default	Description
`model`	string	`meta-llama/llama-4-scout-17b-16e-instruct`	Model to use
`instructions`	string	`"You are a helpful AI assistant."`	System prompt
`system_timezone`	string	`"UTC"`	Timezone context
`temperature`	float (0–2)	0.8	Response randomness
`max_completion_tokens`	int (1–32768)	4096	Max response length
`reasoning_effort`	none/low/medium/high	null	Reasoning depth (gpt-oss-20b only)
`parallel_tool_calls`	bool	true	Allow concurrent tool calls
`tool_choice`	auto/required/none	auto	Tool use control
`max_tool_steps`	int (1–20)	3	Max tool call iterations per turn

Reasoning Model

When using openai/gpt-oss-20b, set reasoning_effort to control how much the model "thinks" before responding:


Code
{
  "brain": {
    "provider": "groq",
    "model": "openai/gpt-oss-20b",
    "instructions": "Analyze the caller's issue carefully before responding.",
    "reasoning_effort": "medium"
  }
}

Note: temperature is not used when reasoning_effort is set.

OpenAI — cloud LLM alternative
Phantom Brain — proprietary, no external key

Last modified on November 7, 2026

OpenAI Phantom Brain

LLM

Groq

Groq provides ultra-fast LLM inference via its custom hardware. Average latency: ~443ms — significantly faster than cloud LLM providers.

Requires: Groq API key in Settings → Credentials

Models

Model	Best For
`meta-llama/llama-4-scout-17b-16e-instruct`	Recommended — fast, capable
`llama-3.1-8b-instant`	Ultra-fast, compact
`openai/gpt-oss-20b`	Reasoning tasks (supports `reasoning_effort`)

Configuration


Code
{
  "brain": {
    "provider": "groq",
    "model": "meta-llama/llama-4-scout-17b-16e-instruct",
    "instructions": "You are a helpful assistant.",
    "system_timezone": "UTC",
    "temperature": 0.8,
    "max_completion_tokens": 4096,
    "parallel_tool_calls": true,
    "tool_choice": "auto",
    "max_tool_steps": 3
  }
}

Fields

Field	Type	Default	Description
`model`	string	`meta-llama/llama-4-scout-17b-16e-instruct`	Model to use
`instructions`	string	`"You are a helpful AI assistant."`	System prompt
`system_timezone`	string	`"UTC"`	Timezone context
`temperature`	float (0–2)	0.8	Response randomness
`max_completion_tokens`	int (1–32768)	4096	Max response length
`reasoning_effort`	none/low/medium/high	null	Reasoning depth (gpt-oss-20b only)
`parallel_tool_calls`	bool	true	Allow concurrent tool calls
`tool_choice`	auto/required/none	auto	Tool use control
`max_tool_steps`	int (1–20)	3	Max tool call iterations per turn

Reasoning Model

When using openai/gpt-oss-20b, set reasoning_effort to control how much the model "thinks" before responding:


Code
{
  "brain": {
    "provider": "groq",
    "model": "openai/gpt-oss-20b",
    "instructions": "Analyze the caller's issue carefully before responding.",
    "reasoning_effort": "medium"
  }
}

Note: temperature is not used when reasoning_effort is set.

OpenAI — cloud LLM alternative
Phantom Brain — proprietary, no external key

Last modified on November 7, 2026

OpenAI Phantom Brain

Models

Configuration

Fields

Reasoning Model

Related

Models

Configuration

Fields

Reasoning Model

Related