Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Y-Research-SBU/QuantAgent/llms.txt

Use this file to discover all available pages before exploring further.

TradingGraph is the top-level entry point for running QuantAgent analysis. It initializes the language models, the TechnicalTools toolkit, and the compiled LangGraph StateGraph that wires together the Indicator, Pattern, Trend, and Decision agents.

​
Constructor

TradingGraph(config=None)
​
config
dict
default:"DEFAULT_CONFIG"
Optional configuration dictionary. When None, DEFAULT_CONFIG is used. Accepts the following keys:
KeyTypeDefaultDescription
agent_llm_providerstr"openai"Provider for per-agent LLMs ("openai", "anthropic", or "qwen").
agent_llm_modelstr"gpt-4o-mini"Model name for the agent LLM.
agent_llm_temperaturefloat0.1Sampling temperature for the agent LLM.
graph_llm_providerstr"openai"Provider for the vision/graph LLM.
graph_llm_modelstr"gpt-4o"Model name for the graph LLM. Must be vision-capable.
graph_llm_temperaturefloat0.1Sampling temperature for the graph LLM.
api_keystr—OpenAI API key. Falls back to OPENAI_API_KEY env var.
anthropic_api_keystr—Anthropic API key. Falls back to ANTHROPIC_API_KEY env var.
qwen_api_keystr—Qwen (DashScope) API key. Falls back to DASHSCOPE_API_KEY env var.
graph_llm_model must be a vision-capable model. The Pattern and Trend agents pass base64-encoded chart images to this LLM for visual analysis.

​
Attributes

AttributeTypeDescription
configdictActive configuration dictionary.
agent_llmBaseChatModelLLM instance used by the Pattern and Trend agents for the tool-dispatch step (chart generation).
graph_llmBaseChatModelPrimary LLM instance used by the Indicator agent (tool calling and report), the Pattern and Trend agents (vision analysis), and the Decision agent (trade synthesis). Must be vision-capable.
toolkitTechnicalToolsToolkit providing all indicator and charting tools.
graphCompiledStateGraphThe compiled LangGraph graph. Call .invoke() on this to run a full analysis.

​
graph.invoke(initial_state)

Runs all four agents in sequence (Indicator → Pattern → Trend → Decision) and returns the completed state dictionary.

​
Parameters

​
initial_state
dict
required
The starting state passed into the graph. Required keys:

​
Return value

Returns the final IndicatorAgentState dictionary. Key output fields:
​
final_trade_decision
str
JSON string containing the trade decision. Fields: forecast_horizon, decision (LONG or SHORT), justification, and risk_reward_ratio.
​
indicator_report
str
Narrative report from the Indicator agent summarizing RSI, MACD, Stochastic, ROC, and Williams %R values.
​
pattern_report
str
Narrative report from the Pattern agent identifying classical candlestick formations.
​
trend_report
str
Narrative report from the Trend agent describing support/resistance lines and directional bias.

​
update_api_key(api_key, provider='openai')

Updates the API key in both self.config and the corresponding environment variable, then calls refresh_llms() to rebuild all LLM objects.
​
api_key
str
required
The new API key value.
​
provider
str
default:"openai"
The provider whose key to update. One of "openai", "anthropic", or "qwen".

​
refresh_llms()

Recreates agent_llm, graph_llm, the SetGraph instance, and the compiled graph from the current self.config. Call this after manually modifying self.config to apply the changes.

​
Example

import yfinance as yf
from trading_graph import TradingGraph

# Download recent 4-hour BTC/USD data
ticker = yf.Ticker("BTC-USD")
df = ticker.history(period="7d", interval="1h").tail(30)
df.index.name = "Datetime"
df = df.reset_index()[["Datetime", "Open", "High", "Low", "Close"]]
df["Datetime"] = df["Datetime"].astype(str)
kline_data = df.to_dict(orient="list")

# Initialize with a custom config
config = {
    "agent_llm_provider": "openai",
    "agent_llm_model": "gpt-4o-mini",
    "graph_llm_provider": "openai",
    "graph_llm_model": "gpt-4o",
    "agent_llm_temperature": 0.1,
    "graph_llm_temperature": 0.1,
    "api_key": "sk-...",
}
tg = TradingGraph(config=config)

# Build the initial state
initial_state = {
    "kline_data": kline_data,
    "analysis_results": None,
    "messages": [],
    "time_frame": "1h",
    "stock_name": "BTC-USD",
}

# Run the full multi-agent pipeline
final_state = tg.graph.invoke(initial_state)

import json
decision = json.loads(final_state["final_trade_decision"])
print(decision["decision"])         # "LONG" or "SHORT"
print(decision["justification"])
print(final_state["indicator_report"])
print(final_state["pattern_report"])
print(final_state["trend_report"])

​
Updating the API key at runtime

# Switch to Anthropic mid-session
tg.update_api_key("sk-ant-...", provider="anthropic")
tg.config["agent_llm_model"] = "claude-haiku-4-5-20251001"
tg.config["graph_llm_model"] = "claude-haiku-4-5-20251001"
tg.refresh_llms()

Build docs developers (and LLMs) love

Get started for freeTalk to us