Multi-Agent Stock Research with CrewAI + Chart Library

When you ask a single AI agent to research a stock, it tries to do everything at once: check the chart pattern, assess the market, evaluate sectors, and synthesize a view. The result is usually shallow.

Multi-agent systems fix this by splitting work across specialists -- exactly how institutional research desks operate. CrewAI lets you build that structure with AI agents and Chart Library's pattern intelligence API.

The Crew Design: Two Specialist Agents

Pattern Analyst -- specializes in individual stock chart analysis. Uses Chart Library's intelligence endpoint to find historically similar charts, reads forward return statistics, and runs scenario stress tests. Always cites specific numbers: "7 of 10 similar patterns went up over 5 days, averaging +2.3%."

Regime Analyst -- specializes in the market-wide environment. Checks SPY/QQQ regime status, sector rotation, and crowding risk. Frames everything as historical analogy: "similar conditions historically led to..." rather than predictions.

Agent backstories matter more than you'd expect. A backstory that says "you never make predictions, you present historical context" produces very different output than "you are a bold market forecaster."

Defining the Tools

Each agent gets curated tools from Chart Library's Python SDK:

pip install crewai crewai-tools chartlibrary

Pattern Analyst tools:

• chart_intelligence -- calls cl.intelligence(symbol, date) for 10 most similar historical patterns + forward returns
• scenario_stress_test -- calls cl.scenario(symbol, market_move_pct) for "what if SPY drops 5%?"
• daily_top_picks -- calls cl.discover(limit=10) for today's top patterns

Regime Analyst tools:

• market_regime -- calls cl.regime(compact=True) for SPY, QQQ, and 11 sector ETFs
• crowding_detector -- calls cl.crowding() for systematic risk signals
• sector_rotation -- calls cl.sector_rotation(lookback) for momentum rankings

Building the Tasks

The order matters: regime assessment first (establishes context), then stock analysis (uses that context), then synthesis (combines both).

Task 1 - Regime Assessment: Check SPY/QQQ, identify leading/lagging sectors, evaluate crowding risk.

Task 2 - Stock Analysis: For each symbol in your watchlist, report match count, quality, forward returns. Receives Task 1's output as context -- so findings can be framed like "NVDA's bullish pattern aligns with the current risk-on regime."

Task 3 - Synthesis: Produces a structured briefing: Market Environment, Stock Highlights, Risk Factors, Bottom Line.

Running the Crew

crew = Crew(
    agents=[pattern_analyst, regime_analyst],
    tasks=[regime_task, analysis_task, synthesis_task],
    process=Process.sequential,
    verbose=True
)

result = crew.kickoff()

Example Output

The Regime Analyst checks SPY: 7 of 10 historically similar regimes gained over 10 days. Tech (XLK) and Industrials (XLI) leading. Moderate crowding in large-cap tech.

The Pattern Analyst analyzes each symbol:

• NVDA: 10 matches, top match 94% similarity, 8/10 positive over 5 days (+3.1% avg)
• AAPL: More mixed -- 6/10 positive, +0.8% average
• TSLA: Historically volatile, wide outcome range

The synthesis combines both: bullish regime supports NVDA, AAPL is neutral, TSLA's wide range means sizing matters more than direction.

Extending the Crew

• Risk Manager agent: Run stress tests for -3%, -5%, -10% market moves across the watchlist
• Portfolio Optimizer agent: Suggest position sizes based on conviction and correlation
• Hierarchical process: A Research Director that delegates dynamically instead of fixed sequence

Start with two agents and add complexity only when you hit a real limitation. Multi-agent systems are powerful but harder to debug.

The complete working example is at github.com/grahammccain/chart-library-mcp in examples/crewai_tutorial.py.

---

Get your free API key at chartlibrary.io/developers and build your first research crew today. 24M patterns. 10 years. One API call.