@stackmemoryai/stackmemory 0.5.59 → 0.5.61

package/dist/core/context/model-aware-compaction.js

@@ -1,623 +0,0 @@
-import { fileURLToPath as __fileURLToPath } from 'url';
-import { dirname as __pathDirname } from 'path';
-const __filename = __fileURLToPath(import.meta.url);
-const __dirname = __pathDirname(__filename);
-import { logger } from "../monitoring/logger.js";
-class ModelAwareCompactionHandler {
-  static MODEL_PROFILES = {
-    // === ANTHROPIC CLAUDE MODELS ===
-    "claude-3-haiku": {
-      name: "Claude 3 Haiku",
-      contextWindow: 2e5,
-      outputLimit: 4096,
-      warningThreshold: 15e4,
-      // 75%
-      criticalThreshold: 17e4,
-      // 85%
-      compactionZone: 18e4,
-      // 90%
-      characteristics: {
-        supportsLongContext: true,
-        costTier: "low"
-      }
-    },
-    "claude-3-5-haiku": {
-      name: "Claude 3.5 Haiku",
-      contextWindow: 2e5,
-      outputLimit: 8192,
-      warningThreshold: 15e4,
-      criticalThreshold: 17e4,
-      compactionZone: 18e4,
-      characteristics: {
-        supportsLongContext: true,
-        costTier: "low"
-      }
-    },
-    "claude-3-sonnet": {
-      name: "Claude 3 Sonnet",
-      contextWindow: 2e5,
-      outputLimit: 4096,
-      warningThreshold: 15e4,
-      criticalThreshold: 17e4,
-      compactionZone: 18e4,
-      characteristics: {
-        supportsLongContext: true,
-        costTier: "medium"
-      }
-    },
-    "claude-3-5-sonnet": {
-      name: "Claude 3.5 Sonnet",
-      contextWindow: 2e5,
-      outputLimit: 8192,
-      // With beta header
-      warningThreshold: 15e4,
-      criticalThreshold: 17e4,
-      compactionZone: 18e4,
-      characteristics: {
-        supportsLongContext: true,
-        betaFeatures: ["max-tokens-3-5-sonnet-2024-07-15"],
-        costTier: "medium"
-      }
-    },
-    "claude-3-opus": {
-      name: "Claude 3 Opus",
-      contextWindow: 2e5,
-      // Standard, can be 1M in beta
-      outputLimit: 4096,
-      warningThreshold: 15e4,
-      criticalThreshold: 17e4,
-      compactionZone: 18e4,
-      characteristics: {
-        supportsLongContext: true,
-        costTier: "premium"
-      }
-    },
-    "claude-3-opus-beta-1m": {
-      name: "Claude 3 Opus (1M Beta)",
-      contextWindow: 1e6,
-      outputLimit: 4096,
-      warningThreshold: 75e4,
-      // 75%
-      criticalThreshold: 85e4,
-      // 85%
-      compactionZone: 9e5,
-      // 90%
-      characteristics: {
-        supportsLongContext: true,
-        betaFeatures: ["1m-context-window"],
-        costTier: "premium"
-      }
-    },
-    "claude-2": {
-      name: "Claude 2",
-      contextWindow: 1e5,
-      outputLimit: 4096,
-      warningThreshold: 75e3,
-      // 75%
-      criticalThreshold: 85e3,
-      // 85%
-      compactionZone: 9e4,
-      // 90%
-      characteristics: {
-        supportsLongContext: true,
-        costTier: "medium"
-      }
-    },
-    "claude-instant": {
-      name: "Claude Instant",
-      contextWindow: 1e5,
-      outputLimit: 4096,
-      warningThreshold: 75e3,
-      criticalThreshold: 85e3,
-      compactionZone: 9e4,
-      characteristics: {
-        supportsLongContext: true,
-        costTier: "low"
-      }
-    },
-    // === OPENAI GPT MODELS ===
-    "gpt-4o": {
-      name: "GPT-4o",
-      contextWindow: 128e3,
-      outputLimit: 16384,
-      // Can be 4096-16384
-      warningThreshold: 96e3,
-      // 75%
-      criticalThreshold: 108800,
-      // 85%
-      compactionZone: 115200,
-      // 90%
-      characteristics: {
-        supportsLongContext: true,
-        costTier: "high"
-      }
-    },
-    "gpt-4-turbo": {
-      name: "GPT-4 Turbo",
-      contextWindow: 128e3,
-      outputLimit: 4096,
-      warningThreshold: 96e3,
-      criticalThreshold: 108800,
-      compactionZone: 115200,
-      characteristics: {
-        supportsLongContext: true,
-        costTier: "high"
-      }
-    },
-    "gpt-4": {
-      name: "GPT-4",
-      contextWindow: 8192,
-      // Standard GPT-4
-      outputLimit: 4096,
-      warningThreshold: 6144,
-      // 75%
-      criticalThreshold: 6963,
-      // 85%
-      compactionZone: 7373,
-      // 90%
-      characteristics: {
-        supportsLongContext: false,
-        costTier: "high"
-      }
-    },
-    "gpt-4-32k": {
-      name: "GPT-4 32K",
-      contextWindow: 32768,
-      outputLimit: 4096,
-      warningThreshold: 24576,
-      // 75%
-      criticalThreshold: 27853,
-      // 85%
-      compactionZone: 29491,
-      // 90%
-      characteristics: {
-        supportsLongContext: true,
-        costTier: "premium"
-      }
-    },
-    "gpt-3.5-turbo": {
-      name: "GPT-3.5 Turbo",
-      contextWindow: 16385,
-      outputLimit: 4096,
-      warningThreshold: 12289,
-      // 75%
-      criticalThreshold: 13927,
-      // 85%
-      compactionZone: 14747,
-      // 90%
-      characteristics: {
-        supportsLongContext: false,
-        costTier: "low"
-      }
-    },
-    "gpt-3.5-turbo-16k": {
-      name: "GPT-3.5 Turbo 16K",
-      contextWindow: 16385,
-      outputLimit: 4096,
-      warningThreshold: 12289,
-      criticalThreshold: 13927,
-      compactionZone: 14747,
-      characteristics: {
-        supportsLongContext: false,
-        costTier: "low"
-      }
-    },
-    // === GOOGLE GEMINI MODELS ===
-    "gemini-1.5-pro": {
-      name: "Gemini 1.5 Pro",
-      contextWindow: 128e3,
-      // Standard, can be up to 2M
-      outputLimit: 32768,
-      // 32K output
-      warningThreshold: 96e3,
-      criticalThreshold: 108800,
-      compactionZone: 115200,
-      characteristics: {
-        supportsLongContext: true,
-        costTier: "medium"
-      }
-    },
-    "gemini-1.5-pro-1m": {
-      name: "Gemini 1.5 Pro (1M)",
-      contextWindow: 1e6,
-      outputLimit: 32768,
-      warningThreshold: 75e4,
-      // 75%
-      criticalThreshold: 85e4,
-      // 85%
-      compactionZone: 9e5,
-      // 90%
-      characteristics: {
-        supportsLongContext: true,
-        betaFeatures: ["1m-context"],
-        costTier: "high"
-      }
-    },
-    "gemini-1.5-pro-2m": {
-      name: "Gemini 1.5 Pro (2M)",
-      contextWindow: 2e6,
-      outputLimit: 32768,
-      warningThreshold: 15e5,
-      // 75%
-      criticalThreshold: 17e5,
-      // 85%
-      compactionZone: 18e5,
-      // 90%
-      characteristics: {
-        supportsLongContext: true,
-        betaFeatures: ["2m-context"],
-        costTier: "premium"
-      }
-    },
-    "gemini-1.5-flash": {
-      name: "Gemini 1.5 Flash",
-      contextWindow: 1e6,
-      outputLimit: 32768,
-      warningThreshold: 75e4,
-      criticalThreshold: 85e4,
-      compactionZone: 9e5,
-      characteristics: {
-        supportsLongContext: true,
-        costTier: "low"
-      }
-    },
-    "gemini-2.0-flash": {
-      name: "Gemini 2.0 Flash",
-      contextWindow: 1e6,
-      outputLimit: 32768,
-      warningThreshold: 75e4,
-      criticalThreshold: 85e4,
-      compactionZone: 9e5,
-      characteristics: {
-        supportsLongContext: true,
-        betaFeatures: ["native-tools"],
-        costTier: "low"
-      }
-    },
-    "gemini-1.0-pro": {
-      name: "Gemini 1.0 Pro",
-      contextWindow: 32768,
-      outputLimit: 8192,
-      warningThreshold: 24576,
-      criticalThreshold: 27853,
-      compactionZone: 29491,
-      characteristics: {
-        supportsLongContext: false,
-        costTier: "medium"
-      }
-    },
-    // === MISTRAL MODELS ===
-    "mistral-large": {
-      name: "Mistral Large",
-      contextWindow: 128e3,
-      outputLimit: 4096,
-      warningThreshold: 96e3,
-      criticalThreshold: 108800,
-      compactionZone: 115200,
-      characteristics: {
-        supportsLongContext: true,
-        costTier: "medium"
-      }
-    },
-    "mistral-medium": {
-      name: "Mistral Medium",
-      contextWindow: 32768,
-      outputLimit: 4096,
-      warningThreshold: 24576,
-      criticalThreshold: 27853,
-      compactionZone: 29491,
-      characteristics: {
-        supportsLongContext: false,
-        costTier: "low"
-      }
-    }
-  };
-  currentModel;
-  detectionHistory = [];
-  tokenEstimate = 0;
-  constructor() {
-    this.currentModel = ModelAwareCompactionHandler.MODEL_PROFILES["claude-3-5-sonnet"];
-  }
-  /**
-   * Detect model from various signals
-   */
-  detectModel(signals) {
-    let detectedModel = "claude-3-5-sonnet";
-    let confidence = 0.5;
-    let source = "default";
-    if (signals.headers) {
-      const modelHeader = signals.headers["x-anthropic-model"] || signals.headers["anthropic-model"] || signals.headers["model"];
-      if (modelHeader) {
-        const normalized = this.normalizeModelName(modelHeader);
-        if (ModelAwareCompactionHandler.MODEL_PROFILES[normalized]) {
-          detectedModel = normalized;
-          confidence = 0.95;
-          source = "header";
-        }
-      }
-      const betaHeader = signals.headers["anthropic-beta"];
-      if (betaHeader) {
-        if (betaHeader.includes("max-tokens-3-5-sonnet")) {
-          detectedModel = "claude-3-5-sonnet";
-          confidence = Math.max(confidence, 0.9);
-        } else if (betaHeader.includes("1m-context")) {
-          detectedModel = "claude-3-opus-beta-1m";
-          confidence = 0.95;
-          source = "header";
-        }
-      }
-    }
-    if (signals.outputLength && confidence < 0.9) {
-      if (signals.outputLength > 4096 && signals.outputLength <= 8192) {
-        detectedModel = detectedModel.includes("haiku") ? "claude-3-5-haiku" : "claude-3-5-sonnet";
-        confidence = Math.max(confidence, 0.7);
-        source = source === "default" ? "behavior" : source;
-      }
-    }
-    if (signals.tokenCount && signals.tokenCount > 2e5) {
-      detectedModel = "claude-3-opus-beta-1m";
-      confidence = 0.8;
-      source = "behavior";
-    }
-    if (signals.responsePatterns && confidence < 0.8) {
-      const patterns = signals.responsePatterns.toLowerCase();
-      if (patterns.includes("claude 3.5 sonnet")) {
-        detectedModel = "claude-3-5-sonnet";
-        confidence = 0.85;
-        source = "behavior";
-      } else if (patterns.includes("claude 3 opus")) {
-        detectedModel = "claude-3-opus";
-        confidence = 0.85;
-        source = "behavior";
-      } else if (patterns.includes("claude 3.5 haiku")) {
-        detectedModel = "claude-3-5-haiku";
-        confidence = 0.85;
-        source = "behavior";
-      }
-    }
-    const profile = ModelAwareCompactionHandler.MODEL_PROFILES[detectedModel];
-    const result = {
-      detectedModel,
-      confidence,
-      profile,
-      source
-    };
-    this.detectionHistory.push(result);
-    this.currentModel = profile;
-    logger.info(
-      `Model detected: ${profile.name} (confidence: ${confidence}, source: ${source})`
-    );
-    return result;
-  }
-  /**
-   * Normalize model name to match our profiles
-   */
-  normalizeModelName(modelName) {
-    const normalized = modelName.toLowerCase().replace(/[^a-z0-9-]/g, "-").replace(/-+/g, "-").replace(/^-|-$/g, "");
-    const mappings = {
-      // Claude models
-      "claude-3-haiku": "claude-3-haiku",
-      "claude-3-5-haiku": "claude-3-5-haiku",
-      "claude-haiku": "claude-3-haiku",
-      "claude-3-sonnet": "claude-3-sonnet",
-      "claude-3-5-sonnet": "claude-3-5-sonnet",
-      "claude-sonnet": "claude-3-sonnet",
-      "claude-3-opus": "claude-3-opus",
-      "claude-opus": "claude-3-opus",
-      "claude-2": "claude-2",
-      "claude-instant": "claude-instant",
-      "claude-instant-1": "claude-instant",
-      // OpenAI models
-      "gpt-4o": "gpt-4o",
-      "gpt-4o-mini": "gpt-4o",
-      "gpt-4-turbo": "gpt-4-turbo",
-      "gpt-4-turbo-preview": "gpt-4-turbo",
-      "gpt-4": "gpt-4",
-      "gpt-4-32k": "gpt-4-32k",
-      "gpt-3-5-turbo": "gpt-3.5-turbo",
-      "gpt-35-turbo": "gpt-3.5-turbo",
-      "gpt-3-5-turbo-16k": "gpt-3.5-turbo-16k",
-      // Google models
-      "gemini-pro": "gemini-1.0-pro",
-      "gemini-1-5-pro": "gemini-1.5-pro",
-      "gemini-1-5-flash": "gemini-1.5-flash",
-      "gemini-2-0-flash": "gemini-2.0-flash",
-      "gemini-pro-1-5": "gemini-1.5-pro",
-      "gemini-flash": "gemini-1.5-flash",
-      // Mistral models
-      "mistral-large": "mistral-large",
-      "mistral-large-latest": "mistral-large",
-      "mistral-medium": "mistral-medium",
-      "mistral-medium-latest": "mistral-medium"
-    };
-    return mappings[normalized] || normalized;
-  }
-  /**
-   * Get adaptive thresholds based on current model
-   */
-  getAdaptiveThresholds() {
-    return {
-      warning: this.currentModel.warningThreshold,
-      critical: this.currentModel.criticalThreshold,
-      compaction: this.currentModel.compactionZone
-    };
-  }
-  /**
-   * Calculate preservation strategy based on model
-   */
-  getPreservationStrategy() {
-    const costTier = this.currentModel.characteristics.costTier;
-    const contextSize = this.currentModel.contextWindow;
-    if (costTier === "premium") {
-      return {
-        aggressiveness: "aggressive",
-        preserveRatio: 0.3,
-        // Preserve 30% of context
-        compressionLevel: "light"
-      };
-    } else if (costTier === "medium") {
-      return {
-        aggressiveness: "balanced",
-        preserveRatio: 0.2,
-        // Preserve 20% of context
-        compressionLevel: "moderate"
-      };
-    } else {
-      return {
-        aggressiveness: "minimal",
-        preserveRatio: 0.1,
-        // Preserve 10% of context
-        compressionLevel: "heavy"
-      };
-    }
-  }
-  /**
-   * Estimate tokens more accurately based on model
-   */
-  estimateTokens(text) {
-    let baseEstimate = Math.ceil(text.length / 3.5);
-    const codeIndicators = [
-      "{",
-      "}",
-      "(",
-      ")",
-      ";",
-      "function",
-      "const",
-      "let"
-    ];
-    const codeScore = codeIndicators.reduce(
-      (score, indicator) => score + (text.split(indicator).length - 1),
-      0
-    );
-    if (codeScore > 50) {
-      baseEstimate *= 1.2;
-    }
-    const avgWordLength = text.split(/\s+/).reduce((sum, word) => sum + word.length, 0) / Math.max(1, text.split(/\s+/).length);
-    if (avgWordLength > 6) {
-      baseEstimate *= 0.9;
-    }
-    this.tokenEstimate += baseEstimate;
-    return Math.round(baseEstimate);
-  }
-  /**
-   * Check if approaching any threshold
-   */
-  checkThresholds(currentTokens) {
-    const thresholds = this.getAdaptiveThresholds();
-    const percentage = currentTokens / this.currentModel.contextWindow * 100;
-    const tokensRemaining = this.currentModel.contextWindow - currentTokens;
-    let status;
-    let recommendation;
-    if (currentTokens >= thresholds.compaction) {
-      status = "compaction";
-      recommendation = "Compaction imminent or occurred. Restore critical context immediately.";
-    } else if (currentTokens >= thresholds.critical) {
-      status = "critical";
-      recommendation = "Create full context preservation. Prepare for compaction.";
-    } else if (currentTokens >= thresholds.warning) {
-      status = "warning";
-      recommendation = "Begin selective context preservation. Monitor token usage closely.";
-    } else {
-      status = "safe";
-      recommendation = "Token usage within safe limits. Continue normal operation.";
-    }
-    return {
-      status,
-      percentage,
-      tokensRemaining,
-      recommendation
-    };
-  }
-  /**
-   * Get model-specific compaction hints
-   */
-  getCompactionHints() {
-    const hints = [];
-    if (this.currentModel.contextWindow >= 1e6) {
-      hints.push(
-        "1M context model detected - compaction less likely but still possible"
-      );
-      hints.push(
-        "Consider chunking very large operations to stay under 900K tokens"
-      );
-    } else if (this.currentModel.contextWindow >= 2e5) {
-      hints.push(
-        "Standard 200K context - expect compaction around 180K tokens"
-      );
-      hints.push("Preserve critical context every 50K tokens for safety");
-    } else {
-      hints.push("Limited context model - be aggressive with preservation");
-      hints.push("Consider breaking work into smaller sessions");
-    }
-    if (this.currentModel.characteristics.betaFeatures) {
-      hints.push(
-        `Beta features available: ${this.currentModel.characteristics.betaFeatures.join(", ")}`
-      );
-    }
-    if (this.currentModel.characteristics.costTier === "premium") {
-      hints.push(
-        "Premium model - maximize context preservation to avoid re-processing"
-      );
-    } else if (this.currentModel.characteristics.costTier === "low") {
-      hints.push(
-        "Cost-effective model - can be more aggressive with resets if needed"
-      );
-    }
-    return hints;
-  }
-  /**
-   * Suggest optimal preservation points based on model
-   */
-  getSuggestedPreservationPoints() {
-    const contextWindow = this.currentModel.contextWindow;
-    const points = [];
-    const interval = contextWindow >= 5e5 ? 1e5 : contextWindow >= 2e5 ? 5e4 : 25e3;
-    for (let point = interval; point < contextWindow * 0.9; point += interval) {
-      points.push(point);
-    }
-    points.push(this.currentModel.warningThreshold);
-    points.push(this.currentModel.criticalThreshold);
-    return [...new Set(points)].sort((a, b) => a - b);
-  }
-  /**
-   * Get current model profile
-   */
-  getCurrentModel() {
-    return this.currentModel;
-  }
-  /**
-   * Get detection history for diagnostics
-   */
-  getDetectionHistory() {
-    return this.detectionHistory;
-  }
-  /**
-   * Update model manually (for testing or override)
-   */
-  setModel(modelKey) {
-    const profile = ModelAwareCompactionHandler.MODEL_PROFILES[modelKey];
-    if (profile) {
-      this.currentModel = profile;
-      logger.info(`Model manually set to: ${profile.name}`);
-      return true;
-    }
-    return false;
-  }
-  /**
-   * Export model profiles for reference
-   */
-  static getAvailableModels() {
-    return Object.keys(ModelAwareCompactionHandler.MODEL_PROFILES);
-  }
-  /**
-   * Get specific model profile
-   */
-  static getModelProfile(modelKey) {
-    return ModelAwareCompactionHandler.MODEL_PROFILES[modelKey];
-  }
-}
-export {
-  ModelAwareCompactionHandler
-};
-//# sourceMappingURL=model-aware-compaction.js.map