Skip to main content
Junkie supports multiple LLM providers through an OpenAI-compatible interface. Different agents can use different models optimized for their specific tasks.

​
Provider Configuration

​
Groq (Default)

Groq is the default provider, offering fast inference for open-source models: 
CUSTOM_PROVIDER=groq
CUSTOM_MODEL=openai/gpt-oss-120b
GROQ_API_KEY=gsk_...
Model creation (agent/agent_factory.py:106-114): 
if PROVIDER == "groq":
    return OpenAILike(
        id=MODEL_NAME,
        max_tokens=4096,
        temperature=MODEL_TEMPERATURE,
        top_p=MODEL_TOP_P,
        base_url="https://api.groq.com/openai/v1",
        api_key=GROQ_API_KEY,
    )

​
Custom OpenAI-Compatible Provider

Use any OpenAI-compatible API endpoint: 
CUSTOM_PROVIDER=https://api.openai.com/v1
CUSTOM_MODEL=gpt-5
CUSTOM_PROVIDER_API_KEY=sk-...
Model creation (agent/agent_factory.py:116-124): 
return OpenAILike(
    id=MODEL_NAME,
    max_tokens=4096,
    temperature=MODEL_TEMPERATURE,
    top_p=MODEL_TOP_P,
    base_url=PROVIDER,
    api_key=CUSTOM_PROVIDER_API_KEY,
)

​
Environment Variables

​
CUSTOM_PROVIDER
string
default:"groq"
Provider identifier or base URL:
  • groq - Use Groq’s API
  • Any OpenAI-compatible base URL (e.g., https://api.openai.com/v1)
​
CUSTOM_MODEL
string
default:"openai/gpt-oss-120b"
Model identifier for the main team leader agent.Groq models:
  • openai/gpt-oss-120b
  • llama-3.1-70b-versatile
  • mixtral-8x7b-32768
OpenAI models:
  • gpt-4-turbo
  • gpt-5
  • gpt-4o
​
CUSTOM_PROVIDER_API_KEY
string
required
API key for custom OpenAI-compatible provider.
​
GROQ_API_KEY
string
required
Groq API key. Required even when using custom providers, as some agents always use Groq.
​
MODEL_TEMPERATURE
float
default:"0.3"
Sampling temperature (0.0 - 2.0). Lower values are more deterministic.
​
MODEL_TOP_P
float
default:"0.9"
Nucleus sampling threshold (0.0 - 1.0).

​
Agent-Specific Models

Junkie uses different models optimized for different agent roles:

​
Team Leader

Uses the main model configured via CUSTOM_PROVIDER and CUSTOM_MODEL. 
model = create_model(user_id)
team = Team(
    name="Hero Team",
    model=model,  # Main model
    # ...
)
Reference: agent/agent_factory.py:310-327

​
Code Agent

Hardcoded to use gpt-5 for advanced code generation: 
code_agent = Agent(
    id="code-agent",
    name="Code Agent",
    model=OpenAILike(
        id="gpt-5",
        base_url=PROVIDER,
        api_key=CUSTOM_PROVIDER_API_KEY,
    ),
    # ...
)
Reference: agent/agent_factory.py:196-204

​
Perplexity Sonar Agent

Uses Perplexity’s sonar-pro for real-time web research: 
perplexity_agent = Agent(
    id="pplx-agent",
    name="Perplexity Sonar Pro",
    model=OpenAILike(
        id="sonar-pro",
        base_url=PROVIDER,
        api_key=CUSTOM_PROVIDER_API_KEY
    ),
    # ...
)
Reference: agent/agent_factory.py:234-245

​
Groq Compound Agent

Always uses Groq’s compound model for fast execution: 
compound_agent = Agent(
    id="groq-compound",
    name="Groq Compound",
    model=OpenAILike(
        id="groq/compound",
        max_tokens=8000,
        base_url="https://api.groq.com/openai/v1",
        api_key=GROQ_API_KEY
    ),
    # ...
)
Reference: agent/agent_factory.py:248-260

​
Context Q&A Agent

Uses a long-context model configured separately: 
CONTEXT_AGENT_MODEL=gemini-2.5-flash-lite
CONTEXT_AGENT_MAX_MESSAGES=50000

context_qna_agent = Agent(
    id="context-qna-agent",
    model=OpenAILike(
        id=CONTEXT_AGENT_MODEL,
        max_tokens=8000,
        temperature=0.3,
        base_url=PROVIDER,
        api_key=CUSTOM_PROVIDER_API_KEY,
    ),
    # ...
)
Reference: agent/agent_factory.py:264-290

​
Memory Manager Model

Always uses Groq for fast memory processing: 
memory_model = OpenAILike(
    id="openai/gpt-oss-120b",
    base_url="https://api.groq.com/openai/v1",
    api_key=GROQ_API_KEY,
)
memory_manager = MemoryManager(
    model=memory_model,
    db=db,
)
Reference: agent/agent_factory.py:153-161

​
System Prompt Management

System prompts can be managed through Phoenix: 
def get_prompt() -> str:
    """Return system prompt from Phoenix or fallback."""
    prompt_name = "herocomp"
    
    try:
        fetched = client.prompts.get(
            prompt_identifier=prompt_name,
            tag="production"
        )
        return fetched.messages[0].get("content")
    except Exception as e:
        return get_system_prompt()  # Fallback to local
Reference: agent/agent_factory.py:126-147

​
Example Configurations

CUSTOM_PROVIDER=groq
CUSTOM_MODEL=openai/gpt-oss-120b
GROQ_API_KEY=gsk_your_key_here
MODEL_TEMPERATURE=0.3
MODEL_TOP_P=0.9

​
API Key Requirements

GROQ_API_KEY is always required, even when using a custom provider, because:
  • Memory manager uses Groq’s openai/gpt-oss-120b
  • Groq Compound agent uses groq/compound
  • Both agents are essential for Junkie’s functionality

​
Supported Providers

Any provider compatible with OpenAI’s API format will work:
  • Groq - Fast open-source models
  • OpenAI - GPT-4, GPT-5 models
  • Together AI - Various open models
  • Perplexity - Sonar models with web search
  • Anthropic (via proxy) - Claude models
  • Local providers - Ollama, LM Studio, etc.
Make sure your provider supports the OpenAI API format with /v1/chat/completions endpoint.

Build docs developers (and LLMs) love

Get started for freeTalk to us