Memory and ML Tools

Memory and ML tools give agents persistent memory, self-awareness of behavioral patterns, and predictive analytics on their own trading.

store_memory

Cost: $0.003 | Store a memory entry for later retrieval.

Agents can store observations, lessons, trade notes, or any structured data that they want to recall in future sessions. Memories are encrypted at rest using per-agent AES-256 keys.

result = client.call_tool(
    "store_memory",
    content="AAPL tends to gap down on earnings misses. Last 3 earnings: 2 gaps down, 1 flat.",
    category="observation",
    metadata={"symbol": "AAPL", "topic": "earnings"},
)

Returns: memory_id, stored_at, category.

search_memory

Cost: $0.003 | Search stored memories using semantic similarity.

result = client.call_tool(
    "search_memory",
    query="What do I know about AAPL earnings behavior?",
    category="observation",
    limit=5,
)

Returns: memories[] each with memory_id, content, category, similarity_score, stored_at.

get_trading_biases

Cost: $0.005 | Analyze the agent's trade history for behavioral biases.

Detects common biases: disposition effect (holding losers too long, cutting winners too short), recency bias, overtrading, loss aversion, anchoring, and confirmation bias.

result = client.call_tool(
    "get_trading_biases",
    trades=[
        {"r_multiple": "1.2", "hold_time_hours": "4", "direction": "long", "symbol": "AAPL"},
        {"r_multiple": "-0.5", "hold_time_hours": "48", "direction": "long", "symbol": "MSFT"},
        {"r_multiple": "0.3", "hold_time_hours": "2", "direction": "short", "symbol": "TSLA"},
    ],
)

Returns: biases[] each with bias_type, severity (LOW, MEDIUM, HIGH), evidence, recommendation.

get_behavioral_fingerprint

Cost: $0.005 | Generate a behavioral fingerprint of the agent's trading style.

Characterizes the agent across multiple dimensions: risk appetite, hold duration profile, directional preference, asset concentration, time-of-day patterns, and streak behavior.

result = client.call_tool(
    "get_behavioral_fingerprint",
    trades=recent_trades,
)

Returns: risk_profile (CONSERVATIVE, MODERATE, AGGRESSIVE), avg_hold_hours, directional_bias, concentration_score, streak_tendency, style_summary.

predict_trajectory

Cost: $0.008 | Predict the agent's performance trajectory based on current patterns.

Uses recent R-multiples and behavioral data to forecast likely outcomes over the next N trades.

result = client.call_tool(
    "predict_trajectory",
    r_multiples=["1.5", "-1.0", "2.3", "-0.5", "1.8", "-1.0", "3.2", "0.8", "1.1", "-0.7"],
    forecast_trades=50,
    starting_equity="100000",
)

Returns: projected_g, projected_equity, confidence_interval, trend_direction, risk_factors, recommendation.

detect_anomalies

Cost: $0.006 | Detect anomalous trades in the agent's history.

Identifies trades that deviate significantly from the agent's normal pattern. Useful for catching errors, unusual market conditions, or strategy drift.

result = client.call_tool(
    "detect_anomalies",
    r_multiples=["1.5", "-1.0", "2.3", "-0.5", "1.8", "-1.0", "3.2", "0.8", "-4.5", "1.1"],
    sensitivity="medium",
)

Returns: anomalies[] each with index, r_multiple, z_score, anomaly_type (EXTREME_WIN, EXTREME_LOSS, PATTERN_BREAK), severity.

cluster_trades

Cost: $0.006 | Cluster trades by outcome pattern to discover hidden groupings.

Groups trades into clusters based on R-multiple, hold time, and other features. Reveals whether the agent has distinct "modes" of trading (e.g., a high-conviction mode vs. a scalping mode).

result = client.call_tool(
    "cluster_trades",
    trades=[
        {"r_multiple": "1.5", "hold_time_hours": "4", "symbol": "AAPL", "direction": "long"},
        {"r_multiple": "-0.3", "hold_time_hours": "0.5", "symbol": "TSLA", "direction": "short"},
        {"r_multiple": "2.8", "hold_time_hours": "72", "symbol": "MSFT", "direction": "long"},
    ],
)

Returns: clusters[] each with cluster_id, trade_count, avg_r, avg_hold_hours, dominant_direction, dominant_symbols[], g_contribution.