API Reference

Strategy API

Inspect and control the adaptive learning system.

getActiveStrategy()

Inspect the current Thompson Sampling posteriors for each strategy arm:

const strategy = await engine.getActiveStrategy(userId: string);
// Returns: {
//   posteriors: Record<string, { alpha: number, beta: number }>,
//   selectedArm: string,
//   confidence: number,
// }

getBoostingPredictions()

View the gradient boosting model's predictions for a given query context. Only available after 100+ retrieval samples:

const predictions = await engine.getBoostingPredictions({
  query: string,
  userId: string,
});
// Returns: Record<string, number> (predicted effectiveness per strategy)

triggerOptimization()

Manually trigger Bayesian parameter optimization. Optimizes 7 continuous parameters targeting the 75th percentile of retrieval quality:

const result = await engine.triggerOptimization({
  userId: string,
  iterations?: number, // default 50
});
// Returns: { parameters: Record<string, number>, improvement: number }

getMetaMemoryRules()

Access the library of LLM-discovered meta-memory rules:

const rules = await engine.getMetaMemoryRules({
  userId?: string,
  minConfidence?: number, // default 0.6
});
// Returns: Array<{
//   pattern: string,
//   action: string,
//   confidence: number,
//   applications: number,
// }>

overrideSelector()

Force a specific strategy for a given context, bypassing the MAB selector:

await engine.overrideSelector({
  userId: string,
  strategy: string,       // e.g., 'semantic-heavy'
  duration?: number,      // auto-expire after N queries
});

PreviousMemory API NextEpisode API