LLM providers

Configure OpenAI, Anthropic, or Qwen as the LLM backend for QuantAgent’s multi-agent analysis pipeline.

QuantAgent supports three LLM providers. Each analysis run uses two separate models: an agent_llm for the Indicator, Pattern, and Trend agents, and a graph_llm for the Decision agent that synthesises the final LONG/SHORT trade directive.

The Pattern and Trend agents pass chart images to the model. Your chosen provider and model must support vision (image) input. All default models listed below satisfy this requirement.

Supported providers

OpenAI

Default provider. Uses the OpenAI Chat Completions API.

Role	Default model
Agent LLM	gpt-4o-mini
Graph LLM	gpt-4o

Set the API key

Environment variable (recommended)

export OPENAI_API_KEY="sk-..."

Config dict

config = {
    "agent_llm_provider": "openai",
    "graph_llm_provider": "openai",
    "agent_llm_model": "gpt-4o-mini",
    "graph_llm_model": "gpt-4o",
    "api_key": "sk-...",
}

Web UI

Open Settings, select OpenAI, paste your key, and click Save. The key is applied immediately via POST /api/update-api-key.

Obtain a key at platform.openai.com/api-keys.

Anthropic

Uses the Anthropic Messages API via langchain-anthropic.

Role	Default model
Agent LLM	claude-haiku-4-5-20251001
Graph LLM	claude-haiku-4-5-20251001

Set the API key

Environment variable (recommended)

export ANTHROPIC_API_KEY="sk-ant-..."

Config dict

config = {
    "agent_llm_provider": "anthropic",
    "graph_llm_provider": "anthropic",
    "agent_llm_model": "claude-haiku-4-5-20251001",
    "graph_llm_model": "claude-haiku-4-5-20251001",
    "anthropic_api_key": "sk-ant-...",
}

Web UI

Open Settings, select Anthropic, paste your key, and click Save.

Obtain a key at console.anthropic.com.

Qwen

Uses Alibaba Cloud’s DashScope API via langchain-qwq. The graph model (qwen3-vl-plus) is a vision-language model required for chart analysis.

Role	Default model
Agent LLM	qwen3-max
Graph LLM	qwen3-vl-plus

DashScope inference endpoints are hosted in Singapore. Depending on your location, you may experience higher latency or occasional timeouts compared to OpenAI or Anthropic. The TradingGraph client is configured with max_retries=4 to handle transient failures automatically.

Set the API key

Environment variable (recommended)

export DASHSCOPE_API_KEY="sk-..."

Config dict

config = {
    "agent_llm_provider": "qwen",
    "graph_llm_provider": "qwen",
    "agent_llm_model": "qwen3-max",
    "graph_llm_model": "qwen3-vl-plus",
    "qwen_api_key": "sk-...",
}

Web UI

Open Settings, select Qwen, paste your DashScope key, and click Save.

Obtain a key at dashscope.console.aliyun.com.

Switching providers

Web UI

Open Settings, choose the provider from the dropdown, and click Apply. This sends a POST /api/update-provider request that updates both agent_llm_provider and graph_llm_provider simultaneously and swaps the model names to the defaults for that provider.

Programmatically

Update the config and call refresh_llms() to rebuild the agent and graph LLMs without recreating the full TradingGraph object:

from trading_graph import TradingGraph

tg = TradingGraph()

# Switch to Anthropic
tg.config["agent_llm_provider"] = "anthropic"
tg.config["graph_llm_provider"] = "anthropic"
tg.config["agent_llm_model"] = "claude-haiku-4-5-20251001"
tg.config["graph_llm_model"] = "claude-haiku-4-5-20251001"
tg.config["anthropic_api_key"] = "sk-ant-..."
tg.refresh_llms()

Or use update_api_key() which calls refresh_llms() for you:

tg.update_api_key("sk-ant-...", provider="anthropic")

Configuration reference

All provider settings live in default_config.py:

DEFAULT_CONFIG = {
    "agent_llm_model": "gpt-4o-mini",
    "graph_llm_model": "gpt-4o",
    "agent_llm_provider": "openai",   # "openai", "anthropic", or "qwen"
    "graph_llm_provider": "openai",
    "agent_llm_temperature": 0.1,
    "graph_llm_temperature": 0.1,
    "api_key": "sk-",                 # OpenAI key
    "anthropic_api_key": "sk-",       # Anthropic key
    "qwen_api_key": "sk-",            # Qwen / DashScope key
}

You can pass a full or partial override dict to TradingGraph(config=...). Any key not present in your dict falls back to the default.