tradingagents

name	tradingagents
description	Multi-agent financial trading analysis. Deploys 5 specialized teams (analysts, researchers, trader, risk manager, portfolio manager) on local Qwen3.6-35B to produce BUY/HOLD/SELL decisions. Agent-agnostic — all analysis via HTTP REST API at :8003.
tags	["trading","finance","stocks","market-analysis","multi-agent","llm","portfolio"]
version	1.0.0
source_url	https://github.com/tauricresearch/tradingagents

Skill: tradingagents

Description

Multi-agent financial trading framework adapted from https://github.com/tauricresearch/tradingagents. Runs entirely on local Qwen3.6-35B via llama.cpp — no remote API keys required (yfinance for data). Agent-agnostic: all 5-team analysis accessible via HTTP REST — callable by any agent/model/runtime.

When to Use

• Analyzing a stock for a trading decision
• Getting technical indicators, fundamentals, news, sentiment for any ticker
• Running a structured bull/bear research debate
• Risk-adjusting a proposed trade
• Any task involving: trade, trading, stock, ticker, market analysis, portfolio, BUY/HOLD/SELL, financial, earnings, fundamentals, technical indicators, bull/bear, equity, OHLCV, MACD, RSI

HTTP Invocation (any agent/model)

# Discover financial data tools
GET http://127.0.0.1:8003/trading/tools

# Quick market signal (market analyst only, fast)
GET http://127.0.0.1:8003/trading/forecast?ticker=AAPL&date=2026-05-01

# Full 5-team analysis
GET http://127.0.0.1:8003/trading/analyze?ticker=AAPL&date=2026-05-01

# Partial analysis (select analysts)
GET http://127.0.0.1:8003/trading/analyze?ticker=AAPL&date=2026-05-01&analysts=market,fundamentals

# Researcher debate round
POST http://127.0.0.1:8003/trading/debate
Body: {"ticker":"AAPL","trade_date":"2026-05-01","market_report":"..."}

# Past decisions
GET http://127.0.0.1:8003/trading/history?ticker=AAPL&limit=10

# Auto-select tools for a trading task
GET http://127.0.0.1:8003/tools/auto-select?task=analyze+AAPL+stock+trading+decision

CLI

aq-trading AAPL 2026-05-01                        # full pipeline
aq-trading AAPL 2026-05-01 --quick                # market signal only
aq-trading AAPL --analysts market,fundamentals    # select analysts
aq-trading --tools                                 # list financial tools
aq-trading AAPL --history                          # past decisions

5-Team Pipeline

Phase 1 — Analysts (run in parallel by type)
  market_analyst      → OHLCV + SMA/EMA/MACD/RSI/Bollinger/ATR
  fundamentals_analyst → balance sheet, cash flow, income statement, P/E
  news_analyst        → recent headlines, insider transactions, macro events
  sentiment_analyst   → headline tone scoring, fear/greed signals

Phase 2 — Researcher Debate
  bull_researcher → growth case, competitive advantages, catalyst
  bear_researcher → downside risk, valuation concerns, sector headwinds
  (iterate for max_debate_rounds)

Phase 3 — Trader Synthesis
  trader_agent → FINAL TRANSACTION PROPOSAL: BUY | HOLD | SELL
                 with entry zone, stop loss, price target

Phase 4 — Risk Management
  risk_manager → volatility, liquidity, drawdown assessment
                 APPROVED | REDUCED | REJECTED

Phase 5 — Portfolio Approval
  portfolio_manager → PORTFOLIO DECISION: APPROVED | MODIFIED | REJECTED
                      with final position size (0.0 – 1.0)

Available Financial Data Tools

Tool	Description	Data Source
get_stock_data	OHLCV price history	yfinance / Alpha Vantage
get_indicators	SMA50/200, EMA10, MACD, RSI, Bollinger, ATR	yfinance
get_fundamentals	P/E, EPS, revenue, margins, debt/equity	yfinance
get_balance_sheet	Assets, liabilities, equity, cash	yfinance
get_cashflow	Operating/investing/financing flows	yfinance
get_income_statement	Revenue, gross profit, EBIT, net income	yfinance
get_news	Recent ticker headlines	yfinance
get_insider_transactions	Insider buy/sell activity	yfinance

Agent Execution Sequence

1. GET  /trading/tools                              # discover available tools
2. GET  /hints?q=trading+analysis+<TICKER>          # ranked hints
3. GET  /trading/forecast?ticker=X&date=Y           # quick signal
4. GET  /trading/analyze?ticker=X&date=Y            # full pipeline
5. POST /memory/recall {query:"<TICKER> decisions"} # prior decisions
6. POST /query {query:"<TICKER>", project:"trading-knowledge"} # knowledge
7. POST /feedback {result, rating}                  # capture outcome

Configuration (env vars — never hardcoded)

• LLAMA_BASE_URL — llama.cpp endpoint (default: http://127.0.0.1:8080)
• TRADING_DATA_VENDOR — yfinance (default) or alpha_vantage
• ALPHA_VANTAGE_API_KEY — optional; enables alpha_vantage vendor
• TRADING_MAX_DEBATE_ROUNDS — researcher debate iterations (default: 1)
• TRADING_ANALYST_TYPES — analyst selection (default: all 4)

AIDB Knowledge Base

Trading knowledge stored in project trading-knowledge.

# Ingest (run once)
scripts/ai/ingest-trading-knowledge.sh

# Query from any agent
curl -s http://127.0.0.1:8002/query \
  -H "X-API-Key: <aidb_key>" \
  -d '{"query":"<topic>","project":"trading-knowledge","limit":3}'

Notes

• No Claude Code dependency — purely HTTP REST
• Runs on local Qwen3.6-35B; ~10-30 min for full pipeline at current hardware
• Skill registered in AIDB and discoverable via GET /control/ai-coordinator/skills
• Module: ai-stack/trading-agents/ (Python package)