agentic-flow 2.0.1-alpha.4 → 2.0.1-alpha.8

package/dist/reasoningbank/index.js

@@ -8,14 +8,17 @@
 import '../utils/agentdb-runtime-patch.js';
 // New hybrid backend (recommended for new code)
 export { HybridReasoningBank } from './HybridBackend.js';
+export { HybridReasoningBank as ReasoningBank } from './HybridBackend.js'; // Backwards compatibility alias
 export { AdvancedMemorySystem } from './AdvancedMemory.js';
-// Re-export AgentDB controllers for advanced usage
-export { ReflexionMemory } from 'agentdb/controllers/ReflexionMemory';
-export { SkillLibrary } from 'agentdb/controllers/SkillLibrary';
-export { CausalMemoryGraph } from 'agentdb/controllers/CausalMemoryGraph';
-export { CausalRecall } from 'agentdb/controllers/CausalRecall';
-export { NightlyLearner } from 'agentdb/controllers/NightlyLearner';
-export { EmbeddingService } from 'agentdb/controllers/EmbeddingService';
+// AgentDB controllers - these are available in agentdb v3.x
+// For now, use HybridReasoningBank which provides equivalent functionality
+// TODO: Re-enable when agentdb v3.x controllers are available
+// export { ReflexionMemory } from 'agentdb/controllers/ReflexionMemory';
+// export { SkillLibrary } from 'agentdb/controllers/SkillLibrary';
+// export { CausalMemoryGraph } from 'agentdb/controllers/CausalMemoryGraph';
+// export { CausalRecall } from 'agentdb/controllers/CausalRecall';
+// export { NightlyLearner } from 'agentdb/controllers/NightlyLearner';
+// export { EmbeddingService } from 'agentdb/controllers/EmbeddingService';
 // Original ReasoningBank implementations (backwards compatibility)
 export { retrieveMemories, formatMemoriesForPrompt } from './core/retrieve.js';
 export { judgeTrajectory } from './core/judge.js';

package/dist/reasoningbank/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/reasoningbank/index.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,yDAAyD;AACzD,OAAO,mCAAmC,CAAC;AAE3C,gDAAgD;AAChD,OAAO,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AACzD,OAAO,EAAE,oBAAoB,EAAE,MAAM,qBAAqB,CAAC;AAI3D,mDAAmD;AACnD,OAAO,EAAE,eAAe,EAAE,MAAM,qCAAqC,CAAC;AACtE,OAAO,EAAE,YAAY,EAAE,MAAM,kCAAkC,CAAC;AAChE,OAAO,EAAE,iBAAiB,EAAE,MAAM,uCAAuC,CAAC;AAC1E,OAAO,EAAE,YAAY,EAAE,MAAM,kCAAkC,CAAC;AAChE,OAAO,EAAE,cAAc,EAAE,MAAM,oCAAoC,CAAC;AACpE,OAAO,EAAE,gBAAgB,EAAE,MAAM,sCAAsC,CAAC;AAExE,mEAAmE;AACnE,OAAO,EAAE,gBAAgB,EAAE,uBAAuB,EAAE,MAAM,oBAAoB,CAAC;AAG/E,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAGlD,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AAGpD,OAAO,EAAE,WAAW,EAAE,iBAAiB,EAAE,MAAM,uBAAuB,CAAC;AAGvE,OAAO,EAAE,aAAa,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAGjE,OAAO,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,MAAM,uBAAuB,CAAC;AAC9E,OAAO,EAAE,YAAY,EAAE,gBAAgB,EAAE,MAAM,gBAAgB,CAAC;AAChE,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,WAAW,EAAE,MAAM,yBAAyB,CAAC;AAC7E,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C,+BAA+B;AAC/B,OAAO,KAAK,EAAE,MAAM,iBAAiB,CAAC;AACtC,OAAO,EAAE,EAAE,EAAE,CAAC;AAYd,+CAA+C;AAC/C,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AACtD,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAClD,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,iBAAiB,IAAI,UAAU,EAAE,WAAW,IAAI,IAAI,EAAE,MAAM,uBAAuB,CAAC;AAE7F,MAAM,CAAC,KAAK,UAAU,UAAU;IAC9B,MAAM,MAAM,GAAG,UAAU,EAAE,CAAC;IAC5B,OAAO,CAAC,GAAG,CAAC,iCAAiC,CAAC,CAAC;IAC/C,OAAO,CAAC,GAAG,CAAC,iDAAiD,CAAC,CAAC;IAC/D,OAAO,CAAC,GAAG,CAAC,6BAA6B,OAAO,CAAC,GAAG,CAAC,mBAAmB,IAAI,kBAAkB,EAAE,CAAC,CAAC;IAClG,OAAO,CAAC,GAAG,CAAC,+BAA+B,MAAM,CAAC,UAAU,CAAC,QAAQ,EAAE,CAAC,CAAC;IACzE,OAAO,CAAC,GAAG,CAAC,gCAAgC,MAAM,CAAC,QAAQ,CAAC,CAAC,EAAE,CAAC,CAAC;IAEjE,IAAI,CAAC;QACH,MAAM,EAAE,CAAC,aAAa,EAAE,CAAC;QACzB,OAAO,CAAC,GAAG,CAAC,gDAAgD,CAAC,CAAC;IAChE,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,OAAO,CAAC,KAAK,CAAC,kCAAkC,EAAE,KAAK,CAAC,CAAC;QACzD,MAAM,IAAI,KAAK,CAAC,+DAA+D,CAAC,CAAC;IACnF,CAAC;IAED,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,EAAE,CAAC,KAAK,EAAE,CAAC;QAC1B,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC,4EAA4E,CAAC,CAAC,GAAG,EAAE,CAAC;QAClH,OAAO,CAAC,GAAG,CAAC,gCAAgC,MAAM,CAAC,MAAM,eAAe,CAAC,CAAC;IAC5E,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,OAAO,CAAC,KAAK,CAAC,iCAAiC,EAAE,KAAK,CAAC,CAAC;QACxD,MAAM,IAAI,KAAK,CAAC,8DAA8D,CAAC,CAAC;IAClF,CAAC;IAED,OAAO,CAAC,GAAG,CAAC,yCAAyC,CAAC,CAAC;AACzD,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,OAAO,CAAC,OAM7B;IAMC,OAAO,CAAC,GAAG,CAAC,iCAAiC,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;IAC/D,MAAM,QAAQ,GAAG,MAAM,gBAAgB,CAAC,OAAO,CAAC,KAAK,EAAE;QACrD,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,KAAK,EAAE,OAAO,CAAC,OAAO;KACvB,CAAC,CAAC;IACH,OAAO,CAAC,GAAG,CAAC,6BAA6B,QAAQ,CAAC,MAAM,WAAW,CAAC,CAAC;IACrE,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC;IACrD,MAAM,OAAO,GAAG,MAAM,eAAe,CAAC,UAAU,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC;IACjE,OAAO,CAAC,GAAG,CAAC,4BAA4B,OAAO,CAAC,KAAK,KAAK,OAAO,CAAC,UAAU,GAAG,CAAC,CAAC;IACjF,MAAM,WAAW,GAAG,MAAM,eAAe,CAAC,UAAU,EAAE,OAAO,EAAE,OAAO,CAAC,KAAK,EAAE;QAC5E,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,OAAO,EAAE,OAAO,CAAC,OAAO;QACxB,MAAM,EAAE,OAAO,CAAC,MAAM;KACvB,CAAC,CAAC;IACH,OAAO,CAAC,GAAG,CAAC,6BAA6B,WAAW,CAAC,MAAM,eAAe,CAAC,CAAC;IAC5E,IAAI,YAAY,GAAG,KAAK,CAAC;IACzB,IAAI,UAAU,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,GAAG,CAAC,0CAA0C,CAAC,CAAC;QACxD,MAAM,IAAI,EAAE,CAAC;QACb,YAAY,GAAG,IAAI,CAAC;IACtB,CAAC;IACD,OAAO,EAAE,OAAO,EAAE,YAAY,EAAE,QAAQ,EAAE,WAAW,EAAE,YAAY,EAAE,CAAC;AACxE,CAAC;AAED,MAAM,CAAC,MAAM,OAAO,GAAG,OAAO,CAAC;AAC/B,MAAM,CAAC,MAAM,SAAS,GAAG,qCAAqC,CAAC","sourcesContent":["/**\n * ReasoningBank - Closed-loop memory system for AI agents\n * Based on arXiv:2509.25140 (Google DeepMind)\n *\n * @since v1.7.0 - Integrated AgentDB for optimal performance\n */\n\n// Apply AgentDB runtime patch before any AgentDB imports\nimport '../utils/agentdb-runtime-patch.js';\n\n// New hybrid backend (recommended for new code)\nexport { HybridReasoningBank } from './HybridBackend.js';\nexport { AdvancedMemorySystem } from './AdvancedMemory.js';\nexport type { PatternData, RetrievalOptions, CausalInsight } from './HybridBackend.js';\nexport type { FailureAnalysis, SkillComposition } from './AdvancedMemory.js';\n\n// Re-export AgentDB controllers for advanced usage\nexport { ReflexionMemory } from 'agentdb/controllers/ReflexionMemory';\nexport { SkillLibrary } from 'agentdb/controllers/SkillLibrary';\nexport { CausalMemoryGraph } from 'agentdb/controllers/CausalMemoryGraph';\nexport { CausalRecall } from 'agentdb/controllers/CausalRecall';\nexport { NightlyLearner } from 'agentdb/controllers/NightlyLearner';\nexport { EmbeddingService } from 'agentdb/controllers/EmbeddingService';\n\n// Original ReasoningBank implementations (backwards compatibility)\nexport { retrieveMemories, formatMemoriesForPrompt } from './core/retrieve.js';\nexport type { RetrievedMemory } from './core/retrieve.js';\n\nexport { judgeTrajectory } from './core/judge.js';\nexport type { Verdict } from './core/judge.js';\n\nexport { distillMemories } from './core/distill.js';\nexport type { DistilledMemory } from './core/distill.js';\n\nexport { consolidate, shouldConsolidate } from './core/consolidate.js';\nexport type { ConsolidationResult } from './core/consolidate.js';\n\nexport { mattsParallel, mattsSequential } from './core/matts.js';\nexport type { MattsResult } from './core/matts.js';\n\nexport { computeEmbedding, clearEmbeddingCache } from './utils/embeddings.js';\nexport { mmrSelection, cosineSimilarity } from './utils/mmr.js';\nexport { scrubPII, containsPII, scrubMemory } from './utils/pii-scrubber.js';\nexport { loadConfig } from './utils/config.js';\n\n// Re-export database utilities\nimport * as db from './db/queries.js';\nexport { db };\nexport type {\n  ReasoningMemory,\n  PatternEmbedding,\n  PatternLink,\n  TaskTrajectory,\n  MattsRun,\n  ConsolidationRun,\n  Trajectory,\n  TrajectoryStep\n} from './db/schema.js';\n\n// Original functions (backwards compatibility)\nimport { loadConfig } from './utils/config.js';\nimport { retrieveMemories } from './core/retrieve.js';\nimport { judgeTrajectory } from './core/judge.js';\nimport { distillMemories } from './core/distill.js';\nimport { shouldConsolidate as shouldCons, consolidate as cons } from './core/consolidate.js';\n\nexport async function initialize(): Promise<void> {\n  const config = loadConfig();\n  console.log('[ReasoningBank] Initializing...');\n  console.log('[ReasoningBank] Enabled: true (initializing...)');\n  console.log(`[ReasoningBank] Database: ${process.env.CLAUDE_FLOW_DB_PATH || '.swarm/memory.db'}`);\n  console.log(`[ReasoningBank] Embeddings: ${config.embeddings.provider}`);\n  console.log(`[ReasoningBank] Retrieval k: ${config.retrieve.k}`);\n\n  try {\n    await db.runMigrations();\n    console.log(`[ReasoningBank] Database migrated successfully`);\n  } catch (error) {\n    console.error('[ReasoningBank] Migration error:', error);\n    throw new Error('ReasoningBank initialization failed: could not run migrations');\n  }\n\n  try {\n    const dbConn = db.getDb();\n    const tables = dbConn.prepare(\"SELECT name FROM sqlite_master WHERE type='table' AND name LIKE 'pattern%'\").all();\n    console.log(`[ReasoningBank] Database OK: ${tables.length} tables found`);\n  } catch (error) {\n    console.error('[ReasoningBank] Database error:', error);\n    throw new Error('ReasoningBank initialization failed: database not accessible');\n  }\n\n  console.log('[ReasoningBank] Initialization complete');\n}\n\nexport async function runTask(options: {\n  taskId: string;\n  agentId: string;\n  query: string;\n  domain?: string;\n  executeFn: (memories: any[]) => Promise<any>;\n}): Promise<{\n  verdict: any;\n  usedMemories: any[];\n  newMemories: string[];\n  consolidated: boolean;\n}> {\n  console.log(`[ReasoningBank] Running task: ${options.taskId}`);\n  const memories = await retrieveMemories(options.query, {\n    domain: options.domain,\n    agent: options.agentId\n  });\n  console.log(`[ReasoningBank] Retrieved ${memories.length} memories`);\n  const trajectory = await options.executeFn(memories);\n  const verdict = await judgeTrajectory(trajectory, options.query);\n  console.log(`[ReasoningBank] Verdict: ${verdict.label} (${verdict.confidence})`);\n  const newMemories = await distillMemories(trajectory, verdict, options.query, {\n    taskId: options.taskId,\n    agentId: options.agentId,\n    domain: options.domain\n  });\n  console.log(`[ReasoningBank] Distilled ${newMemories.length} new memories`);\n  let consolidated = false;\n  if (shouldCons()) {\n    console.log('[ReasoningBank] Running consolidation...');\n    await cons();\n    consolidated = true;\n  }\n  return { verdict, usedMemories: memories, newMemories, consolidated };\n}\n\nexport const VERSION = '1.7.1';\nexport const PAPER_URL = 'https://arxiv.org/html/2509.25140v1';\n"]}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/reasoningbank/index.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,yDAAyD;AACzD,OAAO,mCAAmC,CAAC;AAE3C,gDAAgD;AAChD,OAAO,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AACzD,OAAO,EAAE,mBAAmB,IAAI,aAAa,EAAE,MAAM,oBAAoB,CAAC,CAAC,gCAAgC;AAC3G,OAAO,EAAE,oBAAoB,EAAE,MAAM,qBAAqB,CAAC;AAI3D,4DAA4D;AAC5D,2EAA2E;AAC3E,8DAA8D;AAC9D,yEAAyE;AACzE,mEAAmE;AACnE,6EAA6E;AAC7E,mEAAmE;AACnE,uEAAuE;AACvE,2EAA2E;AAE3E,mEAAmE;AACnE,OAAO,EAAE,gBAAgB,EAAE,uBAAuB,EAAE,MAAM,oBAAoB,CAAC;AAG/E,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAGlD,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AAGpD,OAAO,EAAE,WAAW,EAAE,iBAAiB,EAAE,MAAM,uBAAuB,CAAC;AAGvE,OAAO,EAAE,aAAa,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAGjE,OAAO,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,MAAM,uBAAuB,CAAC;AAC9E,OAAO,EAAE,YAAY,EAAE,gBAAgB,EAAE,MAAM,gBAAgB,CAAC;AAChE,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,WAAW,EAAE,MAAM,yBAAyB,CAAC;AAC7E,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C,+BAA+B;AAC/B,OAAO,KAAK,EAAE,MAAM,iBAAiB,CAAC;AACtC,OAAO,EAAE,EAAE,EAAE,CAAC;AAYd,+CAA+C;AAC/C,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AACtD,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAClD,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,iBAAiB,IAAI,UAAU,EAAE,WAAW,IAAI,IAAI,EAAE,MAAM,uBAAuB,CAAC;AAE7F,MAAM,CAAC,KAAK,UAAU,UAAU;IAC9B,MAAM,MAAM,GAAG,UAAU,EAAE,CAAC;IAC5B,OAAO,CAAC,GAAG,CAAC,iCAAiC,CAAC,CAAC;IAC/C,OAAO,CAAC,GAAG,CAAC,iDAAiD,CAAC,CAAC;IAC/D,OAAO,CAAC,GAAG,CAAC,6BAA6B,OAAO,CAAC,GAAG,CAAC,mBAAmB,IAAI,kBAAkB,EAAE,CAAC,CAAC;IAClG,OAAO,CAAC,GAAG,CAAC,+BAA+B,MAAM,CAAC,UAAU,CAAC,QAAQ,EAAE,CAAC,CAAC;IACzE,OAAO,CAAC,GAAG,CAAC,gCAAgC,MAAM,CAAC,QAAQ,CAAC,CAAC,EAAE,CAAC,CAAC;IAEjE,IAAI,CAAC;QACH,MAAM,EAAE,CAAC,aAAa,EAAE,CAAC;QACzB,OAAO,CAAC,GAAG,CAAC,gDAAgD,CAAC,CAAC;IAChE,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,OAAO,CAAC,KAAK,CAAC,kCAAkC,EAAE,KAAK,CAAC,CAAC;QACzD,MAAM,IAAI,KAAK,CAAC,+DAA+D,CAAC,CAAC;IACnF,CAAC;IAED,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,EAAE,CAAC,KAAK,EAAE,CAAC;QAC1B,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC,4EAA4E,CAAC,CAAC,GAAG,EAAE,CAAC;QAClH,OAAO,CAAC,GAAG,CAAC,gCAAgC,MAAM,CAAC,MAAM,eAAe,CAAC,CAAC;IAC5E,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,OAAO,CAAC,KAAK,CAAC,iCAAiC,EAAE,KAAK,CAAC,CAAC;QACxD,MAAM,IAAI,KAAK,CAAC,8DAA8D,CAAC,CAAC;IAClF,CAAC;IAED,OAAO,CAAC,GAAG,CAAC,yCAAyC,CAAC,CAAC;AACzD,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,OAAO,CAAC,OAM7B;IAMC,OAAO,CAAC,GAAG,CAAC,iCAAiC,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;IAC/D,MAAM,QAAQ,GAAG,MAAM,gBAAgB,CAAC,OAAO,CAAC,KAAK,EAAE;QACrD,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,KAAK,EAAE,OAAO,CAAC,OAAO;KACvB,CAAC,CAAC;IACH,OAAO,CAAC,GAAG,CAAC,6BAA6B,QAAQ,CAAC,MAAM,WAAW,CAAC,CAAC;IACrE,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC;IACrD,MAAM,OAAO,GAAG,MAAM,eAAe,CAAC,UAAU,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC;IACjE,OAAO,CAAC,GAAG,CAAC,4BAA4B,OAAO,CAAC,KAAK,KAAK,OAAO,CAAC,UAAU,GAAG,CAAC,CAAC;IACjF,MAAM,WAAW,GAAG,MAAM,eAAe,CAAC,UAAU,EAAE,OAAO,EAAE,OAAO,CAAC,KAAK,EAAE;QAC5E,MAAM,EAAE,OAAO,CAAC,MAAM;QACtB,OAAO,EAAE,OAAO,CAAC,OAAO;QACxB,MAAM,EAAE,OAAO,CAAC,MAAM;KACvB,CAAC,CAAC;IACH,OAAO,CAAC,GAAG,CAAC,6BAA6B,WAAW,CAAC,MAAM,eAAe,CAAC,CAAC;IAC5E,IAAI,YAAY,GAAG,KAAK,CAAC;IACzB,IAAI,UAAU,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,GAAG,CAAC,0CAA0C,CAAC,CAAC;QACxD,MAAM,IAAI,EAAE,CAAC;QACb,YAAY,GAAG,IAAI,CAAC;IACtB,CAAC;IACD,OAAO,EAAE,OAAO,EAAE,YAAY,EAAE,QAAQ,EAAE,WAAW,EAAE,YAAY,EAAE,CAAC;AACxE,CAAC;AAED,MAAM,CAAC,MAAM,OAAO,GAAG,OAAO,CAAC;AAC/B,MAAM,CAAC,MAAM,SAAS,GAAG,qCAAqC,CAAC","sourcesContent":["/**\n * ReasoningBank - Closed-loop memory system for AI agents\n * Based on arXiv:2509.25140 (Google DeepMind)\n *\n * @since v1.7.0 - Integrated AgentDB for optimal performance\n */\n\n// Apply AgentDB runtime patch before any AgentDB imports\nimport '../utils/agentdb-runtime-patch.js';\n\n// New hybrid backend (recommended for new code)\nexport { HybridReasoningBank } from './HybridBackend.js';\nexport { HybridReasoningBank as ReasoningBank } from './HybridBackend.js'; // Backwards compatibility alias\nexport { AdvancedMemorySystem } from './AdvancedMemory.js';\nexport type { PatternData, RetrievalOptions, CausalInsight } from './HybridBackend.js';\nexport type { FailureAnalysis, SkillComposition } from './AdvancedMemory.js';\n\n// AgentDB controllers - these are available in agentdb v3.x\n// For now, use HybridReasoningBank which provides equivalent functionality\n// TODO: Re-enable when agentdb v3.x controllers are available\n// export { ReflexionMemory } from 'agentdb/controllers/ReflexionMemory';\n// export { SkillLibrary } from 'agentdb/controllers/SkillLibrary';\n// export { CausalMemoryGraph } from 'agentdb/controllers/CausalMemoryGraph';\n// export { CausalRecall } from 'agentdb/controllers/CausalRecall';\n// export { NightlyLearner } from 'agentdb/controllers/NightlyLearner';\n// export { EmbeddingService } from 'agentdb/controllers/EmbeddingService';\n\n// Original ReasoningBank implementations (backwards compatibility)\nexport { retrieveMemories, formatMemoriesForPrompt } from './core/retrieve.js';\nexport type { RetrievedMemory } from './core/retrieve.js';\n\nexport { judgeTrajectory } from './core/judge.js';\nexport type { Verdict } from './core/judge.js';\n\nexport { distillMemories } from './core/distill.js';\nexport type { DistilledMemory } from './core/distill.js';\n\nexport { consolidate, shouldConsolidate } from './core/consolidate.js';\nexport type { ConsolidationResult } from './core/consolidate.js';\n\nexport { mattsParallel, mattsSequential } from './core/matts.js';\nexport type { MattsResult } from './core/matts.js';\n\nexport { computeEmbedding, clearEmbeddingCache } from './utils/embeddings.js';\nexport { mmrSelection, cosineSimilarity } from './utils/mmr.js';\nexport { scrubPII, containsPII, scrubMemory } from './utils/pii-scrubber.js';\nexport { loadConfig } from './utils/config.js';\n\n// Re-export database utilities\nimport * as db from './db/queries.js';\nexport { db };\nexport type {\n  ReasoningMemory,\n  PatternEmbedding,\n  PatternLink,\n  TaskTrajectory,\n  MattsRun,\n  ConsolidationRun,\n  Trajectory,\n  TrajectoryStep\n} from './db/schema.js';\n\n// Original functions (backwards compatibility)\nimport { loadConfig } from './utils/config.js';\nimport { retrieveMemories } from './core/retrieve.js';\nimport { judgeTrajectory } from './core/judge.js';\nimport { distillMemories } from './core/distill.js';\nimport { shouldConsolidate as shouldCons, consolidate as cons } from './core/consolidate.js';\n\nexport async function initialize(): Promise<void> {\n  const config = loadConfig();\n  console.log('[ReasoningBank] Initializing...');\n  console.log('[ReasoningBank] Enabled: true (initializing...)');\n  console.log(`[ReasoningBank] Database: ${process.env.CLAUDE_FLOW_DB_PATH || '.swarm/memory.db'}`);\n  console.log(`[ReasoningBank] Embeddings: ${config.embeddings.provider}`);\n  console.log(`[ReasoningBank] Retrieval k: ${config.retrieve.k}`);\n\n  try {\n    await db.runMigrations();\n    console.log(`[ReasoningBank] Database migrated successfully`);\n  } catch (error) {\n    console.error('[ReasoningBank] Migration error:', error);\n    throw new Error('ReasoningBank initialization failed: could not run migrations');\n  }\n\n  try {\n    const dbConn = db.getDb();\n    const tables = dbConn.prepare(\"SELECT name FROM sqlite_master WHERE type='table' AND name LIKE 'pattern%'\").all();\n    console.log(`[ReasoningBank] Database OK: ${tables.length} tables found`);\n  } catch (error) {\n    console.error('[ReasoningBank] Database error:', error);\n    throw new Error('ReasoningBank initialization failed: database not accessible');\n  }\n\n  console.log('[ReasoningBank] Initialization complete');\n}\n\nexport async function runTask(options: {\n  taskId: string;\n  agentId: string;\n  query: string;\n  domain?: string;\n  executeFn: (memories: any[]) => Promise<any>;\n}): Promise<{\n  verdict: any;\n  usedMemories: any[];\n  newMemories: string[];\n  consolidated: boolean;\n}> {\n  console.log(`[ReasoningBank] Running task: ${options.taskId}`);\n  const memories = await retrieveMemories(options.query, {\n    domain: options.domain,\n    agent: options.agentId\n  });\n  console.log(`[ReasoningBank] Retrieved ${memories.length} memories`);\n  const trajectory = await options.executeFn(memories);\n  const verdict = await judgeTrajectory(trajectory, options.query);\n  console.log(`[ReasoningBank] Verdict: ${verdict.label} (${verdict.confidence})`);\n  const newMemories = await distillMemories(trajectory, verdict, options.query, {\n    taskId: options.taskId,\n    agentId: options.agentId,\n    domain: options.domain\n  });\n  console.log(`[ReasoningBank] Distilled ${newMemories.length} new memories`);\n  let consolidated = false;\n  if (shouldCons()) {\n    console.log('[ReasoningBank] Running consolidation...');\n    await cons();\n    consolidated = true;\n  }\n  return { verdict, usedMemories: memories, newMemories, consolidated };\n}\n\nexport const VERSION = '1.7.1';\nexport const PAPER_URL = 'https://arxiv.org/html/2509.25140v1';\n"]}

package/dist/router/providers/onnx-local.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"onnx-local.d.ts","sourceRoot":"","sources":["../../../src/router/providers/onnx-local.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAOH,OAAO,KAAK,EACV,WAAW,EACX,UAAU,EACV,YAAY,EACZ,WAAW,EAIZ,MAAM,aAAa,CAAC;AAErB,MAAM,WAAW,eAAe;IAC9B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,qBAAa,iBAAkB,YAAW,WAAW;IACnD,IAAI,SAAgB;IACpB,IAAI,EAAG,QAAQ,CAAU;IACzB,iBAAiB,UAAS;IAC1B,aAAa,UAAS;IACtB,WAAW,UAAS;IAEpB,OAAO,CAAC,OAAO,CAAqC;IACpD,OAAO,CAAC,MAAM,CAA4B;IAC1C,OAAO,CAAC,SAAS,CAAa;IAC9B,OAAO,CAAC,QAAQ,CAAa;gBAEjB,MAAM,GAAE,eAAoB;IASxC;;OAEG;YACW,aAAa;IAc3B;;OAEG;IACH,OAAO,CAAC,MAAM;IAId;;OAEG;IACH,OAAO,CAAC,MAAM;IAgBd;;OAEG;YACW,iBAAiB;IA6C/B;;OAEG;IACH,OAAO,CAAC,cAAc;IAqBtB;;;OAGG;IACH,OAAO,CAAC,iBAAiB;IA2BzB;;OAEG;IACG,IAAI,CAAC,MAAM,EAAE,UAAU,GAAG,OAAO,CAAC,YAAY,CAAC;IA2IrD;;OAEG;IACI,MAAM,CAAC,MAAM,EAAE,UAAU,GAAG,cAAc,CAAC,WAAW,CAAC;IAI9D;;OAEG;IACH,oBAAoB,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,OAAO;IAKjD;;OAEG;IACH,YAAY;;;;;;IASZ;;OAEG;IACG,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;CAU/B"}
+{"version":3,"file":"onnx-local.d.ts","sourceRoot":"","sources":["../../../src/router/providers/onnx-local.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAOH,OAAO,KAAK,EACV,WAAW,EACX,UAAU,EACV,YAAY,EACZ,WAAW,EAIZ,MAAM,aAAa,CAAC;AAErB,MAAM,WAAW,eAAe;IAC9B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,qBAAa,iBAAkB,YAAW,WAAW;IACnD,IAAI,SAAgB;IACpB,IAAI,EAAG,QAAQ,CAAU;IACzB,iBAAiB,UAAS;IAC1B,aAAa,UAAS;IACtB,WAAW,UAAS;IAEpB,OAAO,CAAC,OAAO,CAAqC;IACpD,OAAO,CAAC,MAAM,CAA4B;IAC1C,OAAO,CAAC,SAAS,CAAa;IAC9B,OAAO,CAAC,QAAQ,CAAa;gBAEjB,MAAM,GAAE,eAAoB;IASxC;;OAEG;YACW,aAAa;IAc3B;;OAEG;IACH,OAAO,CAAC,MAAM;IAId;;OAEG;IACH,OAAO,CAAC,MAAM;IAgBd;;OAEG;YACW,iBAAiB;IA6C/B;;OAEG;IACH,OAAO,CAAC,cAAc;IAqBtB;;;OAGG;IACH,OAAO,CAAC,iBAAiB;IA6BzB;;OAEG;IACG,IAAI,CAAC,MAAM,EAAE,UAAU,GAAG,OAAO,CAAC,YAAY,CAAC;IA2IrD;;OAEG;IACI,MAAM,CAAC,MAAM,EAAE,UAAU,GAAG,cAAc,CAAC,WAAW,CAAC;IAI9D;;OAEG;IACH,oBAAoB,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,OAAO;IAKjD;;OAEG;IACH,YAAY;;;;;;IASZ;;OAEG;IACG,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;CAU/B"}

package/dist/router/providers/onnx-local.js

@@ -134,11 +134,13 @@ export class ONNXLocalProvider {
         const numLayers = 32;
         const numKVHeads = 8;
         const headDim = 128; // 3072 / 24 = 128
+        // Use any type for KV cache since ort.Tensor is a value, not a type
         const kvCache = {};
         // Initialize empty cache for each layer (key and value)
         for (let i = 0; i < numLayers; i++) {
             // Empty cache: [batch_size, num_kv_heads, 0, head_dim]
             const emptyCache = new Float32Array(0);
+            // Cast to any to avoid TS issues with onnxruntime-node Tensor constructor
             kvCache[`past_key_values.${i}.key`] = new ort.Tensor('float32', emptyCache, [batchSize, numKVHeads, 0, headDim]);
             kvCache[`past_key_values.${i}.value`] = new ort.Tensor('float32', emptyCache, [batchSize, numKVHeads, 0, headDim]);
         }
@@ -168,7 +170,7 @@ export class ONNXLocalProvider {
                 // For first step, use all input tokens; for subsequent steps, use only last token
                 const currentInputIds = step === 0 ? inputIds : [outputIds[outputIds.length - 1]];
                 const currentSeqLen = currentInputIds.length;
-                // Create input tensor for current step
+                // Create input tensor for current step (cast to any for TS compatibility)
                 const inputTensor = new ort.Tensor('int64', BigInt64Array.from(currentInputIds.map(BigInt)), [1, currentSeqLen]);
                 // Create attention mask for current step
                 const totalSeqLen = allTokenIds.length;

package/dist/router/providers/onnx-local.js.map

@@ -1 +1 @@
-{"version":3,"file":"onnx-local.js","sourceRoot":"","sources":["../../../src/router/providers/onnx-local.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,GAAG,MAAM,kBAAkB,CAAC;AAGxC,OAAO,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AACxC,OAAO,EAAE,eAAe,EAAE,eAAe,EAAE,MAAM,iCAAiC,CAAC;AAkBnF,MAAM,OAAO,iBAAiB;IAC5B,IAAI,GAAG,YAAY,CAAC;IACpB,IAAI,GAAG,QAAiB,CAAC;IACzB,iBAAiB,GAAG,KAAK,CAAC,CAAC,mDAAmD;IAC9E,aAAa,GAAG,KAAK,CAAC;IACtB,WAAW,GAAG,KAAK,CAAC;IAEZ,OAAO,GAAgC,IAAI,CAAC;IAC5C,MAAM,CAA4B;IAClC,SAAS,GAAQ,IAAI,CAAC;IACtB,QAAQ,GAAQ,IAAI,CAAC;IAE7B,YAAY,SAA0B,EAAE;QACtC,IAAI,CAAC,MAAM,GAAG;YACZ,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,4EAA4E;YAC3G,kBAAkB,EAAE,MAAM,CAAC,kBAAkB,IAAI,CAAC,KAAK,CAAC;YACxD,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,GAAG;YAClC,WAAW,EAAE,MAAM,CAAC,WAAW,IAAI,GAAG;SACvC,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,aAAa;QACzB,IAAI,IAAI,CAAC,QAAQ;YAAE,OAAO;QAE1B,IAAI,CAAC;YACH,qDAAqD;YACrD,IAAI,CAAC,QAAQ,GAAG,YAAY,CAAC,aAAa,CAAC,CAAC;YAE5C,OAAO,CAAC,GAAG,CAAC,2CAA2C,CAAC,CAAC;QAC3D,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,CAAC,KAAK,CAAC,4BAA4B,EAAE,KAAK,CAAC,CAAC;YACnD,MAAM,IAAI,KAAK,CAAC,6BAA6B,KAAK,EAAE,CAAC,CAAC;QACxD,CAAC;IACH,CAAC;IAED;;OAEG;IACK,MAAM,CAAC,IAAY;QACzB,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;IAChD,CAAC;IAED;;OAEG;IACK,MAAM,CAAC,GAAa;QAC1B,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC;YAC3D,6CAA6C;YAC7C,IAAI,OAAO,OAAO,KAAK,QAAQ,EAAE,CAAC;gBAChC,OAAO,OAAO,CAAC;YACjB,CAAC;iBAAM,IAAI,OAAO,YAAY,UAAU,IAAI,OAAO,YAAY,MAAM,EAAE,CAAC;gBACtE,OAAO,IAAI,WAAW,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YAC3C,CAAC;YACD,OAAO,MAAM,CAAC,OAAO,CAAC,CAAC;QACzB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,CAAC,IAAI,CAAC,kCAAkC,EAAE,KAAK,CAAC,CAAC;YACxD,OAAO,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACvB,CAAC;IACH,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,iBAAiB;QAC7B,IAAI,IAAI,CAAC,OAAO;YAAE,OAAO;QAEzB,IAAI,CAAC;YACH,6BAA6B;YAC7B,OAAO,CAAC,GAAG,CAAC,qCAAqC,CAAC,CAAC;YAEnD,MAAM,SAAS,GAAG,MAAM,eAAe,CAAC,CAAC,QAAQ,EAAE,EAAE;gBACnD,IAAI,QAAQ,CAAC,UAAU,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC,iBAAiB;oBACnD,OAAO,CAAC,GAAG,CAAC,sBAAsB,eAAe,CAAC,cAAc,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;gBAChF,CAAC;YACH,CAAC,CAAC,CAAC;YAEH,uCAAuC;YACvC,IAAI,CAAC,MAAM,CAAC,SAAS,GAAG,SAAS,CAAC;YAElC,OAAO,CAAC,GAAG,CAAC,0BAA0B,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC;YAE/D,IAAI,CAAC,OAAO,GAAG,MAAM,GAAG,CAAC,gBAAgB,CAAC,MAAM,CAC9C,IAAI,CAAC,MAAM,CAAC,SAAS,EACrB;gBACE,kBAAkB,EAAE,IAAI,CAAC,MAAM,CAAC,kBAAyB;gBACzD,sBAAsB,EAAE,KAAK;gBAC7B,iBAAiB,EAAE,IAAI;gBACvB,gBAAgB,EAAE,IAAI;aACvB,CACF,CAAC;YAEF,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC,CAAC;YACnC,OAAO,CAAC,GAAG,CAAC,2BAA2B,IAAI,CAAC,MAAM,CAAC,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YAEpF,iBAAiB;YACjB,MAAM,IAAI,CAAC,aAAa,EAAE,CAAC;QAE7B,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,aAAa,GAAkB;gBACnC,IAAI,EAAE,eAAe;gBACrB,OAAO,EAAE,oCAAoC,KAAK,EAAE;gBACpD,QAAQ,EAAE,YAAY;gBACtB,SAAS,EAAE,KAAK;aACjB,CAAC;YACF,MAAM,aAAa,CAAC;QACtB,CAAC;IACH,CAAC;IAED;;OAEG;IACK,cAAc,CAAC,QAAmB;QACxC,IAAI,MAAM,GAAG,EAAE,CAAC;QAEhB,KAAK,MAAM,GAAG,IAAI,QAAQ,EAAE,CAAC;YAC3B,MAAM,OAAO,GAAG,OAAO,GAAG,CAAC,OAAO,KAAK,QAAQ;gBAC7C,CAAC,CAAC,GAAG,CAAC,OAAO;gBACb,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YAEnE,IAAI,GAAG,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;gBAC1B,MAAM,IAAI,eAAe,OAAO,WAAW,CAAC;YAC9C,CAAC;iBAAM,IAAI,GAAG,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBAC/B,MAAM,IAAI,aAAa,OAAO,WAAW,CAAC;YAC5C,CAAC;iBAAM,IAAI,GAAG,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;gBACpC,MAAM,IAAI,kBAAkB,OAAO,WAAW,CAAC;YACjD,CAAC;QACH,CAAC;QAED,MAAM,IAAI,iBAAiB,CAAC;QAC5B,OAAO,MAAM,CAAC;IAChB,CAAC;IAED;;;OAGG;IACK,iBAAiB,CAAC,SAAiB,EAAE,cAAsB;QACjE,MAAM,SAAS,GAAG,EAAE,CAAC;QACrB,MAAM,UAAU,GAAG,CAAC,CAAC;QACrB,MAAM,OAAO,GAAG,GAAG,CAAC,CAAC,kBAAkB;QACvC,MAAM,OAAO,GAA+B,EAAE,CAAC;QAE/C,wDAAwD;QACxD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,EAAE,CAAC,EAAE,EAAE,CAAC;YACnC,uDAAuD;YACvD,MAAM,UAAU,GAAG,IAAI,YAAY,CAAC,CAAC,CAAC,CAAC;YAEvC,OAAO,CAAC,mBAAmB,CAAC,MAAM,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,CAClD,SAAS,EACT,UAAU,EACV,CAAC,SAAS,EAAE,UAAU,EAAE,CAAC,EAAE,OAAO,CAAC,CACpC,CAAC;YAEF,OAAO,CAAC,mBAAmB,CAAC,QAAQ,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,CACpD,SAAS,EACT,UAAU,EACV,CAAC,SAAS,EAAE,UAAU,EAAE,CAAC,EAAE,OAAO,CAAC,CACpC,CAAC;QACJ,CAAC;QAED,OAAO,OAAO,CAAC;IACjB,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,IAAI,CAAC,MAAkB;QAC3B,MAAM,IAAI,CAAC,iBAAiB,EAAE,CAAC;QAE/B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,MAAM,MAAM,GAAG,IAAI,CAAC,cAAc,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QAEpD,IAAI,CAAC;YACH,0CAA0C;YAC1C,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;YACrC,OAAO,CAAC,GAAG,CAAC,oBAAoB,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;YAEnD,2CAA2C;YAC3C,IAAI,WAAW,GAAG,IAAI,CAAC,iBAAiB,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAE/C,6BAA6B;YAC7B,MAAM,WAAW,GAAG,CAAC,GAAG,QAAQ,CAAC,CAAC;YAClC,MAAM,SAAS,GAAa,EAAE,CAAC;YAE/B,8CAA8C;YAC9C,MAAM,SAAS,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,MAAM,CAAC,SAAS,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;YAEhF,iCAAiC;YACjC,MAAM,YAAY,GAAG,MAAM,CAAC,SAAS,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC;YAE/D,KAAK,IAAI,IAAI,GAAG,CAAC,EAAE,IAAI,GAAG,YAAY,EAAE,IAAI,EAAE,EAAE,CAAC;gBAC/C,kFAAkF;gBAClF,MAAM,eAAe,GAAG,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC;gBAClF,MAAM,aAAa,GAAG,eAAe,CAAC,MAAM,CAAC;gBAE7C,uCAAuC;gBACvC,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC,MAAM,CAChC,OAAO,EACP,aAAa,CAAC,IAAI,CAAC,eAAe,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,EAC/C,CAAC,CAAC,EAAE,aAAa,CAAC,CACnB,CAAC;gBAEF,yCAAyC;gBACzC,MAAM,WAAW,GAAG,WAAW,CAAC,MAAM,CAAC;gBACvC,MAAM,aAAa,GAAG,IAAI,GAAG,CAAC,MAAM,CAClC,OAAO,EACP,aAAa,CAAC,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,EAC/C,CAAC,CAAC,EAAE,WAAW,CAAC,CACjB,CAAC;gBAEF,uDAAuD;gBACvD,MAAM,KAAK,GAA+B;oBACxC,SAAS,EAAE,WAAW;oBACtB,cAAc,EAAE,aAAa;oBAC7B,GAAG,WAAW;iBACf,CAAC;gBAEF,gBAAgB;gBAChB,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;gBAE/C,4CAA4C;gBAC5C,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,IAAoB,CAAC;gBACnD,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;gBAEtE,gCAAgC;gBAChC,MAAM,qBAAqB,GAAG,CAAC,aAAa,GAAG,CAAC,CAAC,GAAG,SAAS,CAAC;gBAE9D,uCAAuC;gBACvC,IAAI,SAAS,GAAG,CAAC,CAAC;gBAClB,IAAI,MAAM,GAAG,CAAC,QAAQ,CAAC;gBAEvB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,EAAE,CAAC,EAAE,EAAE,CAAC;oBACnC,MAAM,KAAK,GAAG,MAAM,CAAC,qBAAqB,GAAG,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,WAAW,IAAI,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;oBAClG,IAAI,KAAK,GAAG,MAAM,EAAE,CAAC;wBACnB,MAAM,GAAG,KAAK,CAAC;wBACf,SAAS,GAAG,CAAC,CAAC;oBAChB,CAAC;gBACH,CAAC;gBAED,gBAAgB;gBAChB,SAAS,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;gBAC1B,WAAW,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;gBAE5B,wDAAwD;gBACxD,IAAI,SAAS,KAAK,CAAC,IAAI,SAAS,KAAK,CAAC,EAAE,CAAC;oBACvC,OAAO,CAAC,GAAG,CAAC,2BAA2B,SAAS,EAAE,CAAC,CAAC;oBACpD,MAAM;gBACR,CAAC;gBAED,kDAAkD;gBAClD,WAAW,GAAG,EAAE,CAAC;gBACjB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;oBAC5B,WAAW,CAAC,mBAAmB,CAAC,MAAM,CAAC,GAAG,OAAO,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC;oBACtE,WAAW,CAAC,mBAAmB,CAAC,QAAQ,CAAC,GAAG,OAAO,CAAC,WAAW,CAAC,QAAQ,CAAC,CAAC;gBAC5E,CAAC;gBAED,qBAAqB;gBACrB,IAAI,CAAC,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,CAAC;oBAC1B,OAAO,CAAC,GAAG,CAAC,gBAAgB,IAAI,GAAG,CAAC,YAAY,CAAC,CAAC;gBACpD,CAAC;YACH,CAAC;YAED,yCAAyC;YACzC,MAAM,aAAa,GAAG,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;YAC7C,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;YACvC,MAAM,eAAe,GAAG,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,OAAO,GAAG,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YAEzE,OAAO,CAAC,GAAG,CAAC,gBAAgB,aAAa,EAAE,CAAC,CAAC;YAC7C,OAAO,CAAC,GAAG,CAAC,gBAAgB,OAAO,OAAO,eAAe,cAAc,CAAC,CAAC;YAEzE,MAAM,OAAO,GAAmB,CAAC;oBAC/B,IAAI,EAAE,MAAM;oBACZ,IAAI,EAAE,aAAa,CAAC,IAAI,EAAE;iBAC3B,CAAC,CAAC;YAEH,OAAO;gBACL,EAAE,EAAE,cAAc,IAAI,CAAC,GAAG,EAAE,EAAE;gBAC9B,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,SAAS;gBAC5B,OAAO;gBACP,UAAU,EAAE,UAAU;gBACtB,KAAK,EAAE;oBACL,WAAW,EAAE,QAAQ,CAAC,MAAM;oBAC5B,YAAY,EAAE,SAAS,CAAC,MAAM;iBAC/B;gBACD,QAAQ,EAAE;oBACR,QAAQ,EAAE,YAAY;oBACtB,KAAK,EAAE,0BAA0B;oBACjC,OAAO;oBACP,IAAI,EAAE,CAAC,EAAE,0BAA0B;oBACnC,kBAAkB,EAAE,IAAI,CAAC,MAAM,CAAC,kBAAkB;oBAClD,eAAe,EAAE,UAAU,CAAC,eAAe,CAAC;iBAC7C;aACF,CAAC;QAEJ,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,aAAa,GAAkB;gBACnC,IAAI,EAAE,oBAAoB;gBAC1B,OAAO,EAAE,0BAA0B,KAAK,EAAE;gBAC1C,QAAQ,EAAE,YAAY;gBACtB,SAAS,EAAE,IAAI;aAChB,CAAC;YACF,MAAM,aAAa,CAAC;QACtB,CAAC;IACH,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,CAAC,MAAM,CAAC,MAAkB;QAC9B,MAAM,IAAI,KAAK,CAAC,wDAAwD,CAAC,CAAC;IAC5E,CAAC;IAED;;OAEG;IACH,oBAAoB,CAAC,QAAkB;QACrC,MAAM,SAAS,GAAG,CAAC,MAAM,CAAC,CAAC;QAC3B,OAAO,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IACpD,CAAC;IAED;;OAEG;IACH,YAAY;QACV,OAAO;YACL,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,SAAS;YAChC,kBAAkB,EAAE,IAAI,CAAC,MAAM,CAAC,kBAAkB;YAClD,WAAW,EAAE,IAAI,CAAC,OAAO,KAAK,IAAI;YAClC,eAAe,EAAE,IAAI,CAAC,QAAQ,KAAK,IAAI;SACxC,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,OAAO;QACX,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YACjB,gEAAgE;YAChE,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC;QACtB,CAAC;QACD,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YAClB,IAAI,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;YACrB,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC;QACvB,CAAC;IACH,CAAC;CACF","sourcesContent":["/**\n * ONNX Runtime Local Inference Provider for Phi-4\n *\n * Uses onnxruntime-node for true local CPU/GPU inference\n */\n\nimport * as ort from 'onnxruntime-node';\nimport * as fs from 'fs';\nimport * as path from 'path';\nimport { get_encoding } from 'tiktoken';\nimport { ensurePhi4Model, ModelDownloader } from '../../utils/model-downloader.js';\nimport type {\n  LLMProvider,\n  ChatParams,\n  ChatResponse,\n  StreamChunk,\n  ProviderError,\n  Message,\n  ContentBlock\n} from '../types.js';\n\nexport interface ONNXLocalConfig {\n  modelPath?: string;\n  executionProviders?: string[];\n  maxTokens?: number;\n  temperature?: number;\n}\n\nexport class ONNXLocalProvider implements LLMProvider {\n  name = 'onnx-local';\n  type = 'custom' as const;\n  supportsStreaming = false; // Streaming requires complex token generation loop\n  supportsTools = false;\n  supportsMCP = false;\n\n  private session: ort.InferenceSession | null = null;\n  private config: Required<ONNXLocalConfig>;\n  private tokenizer: any = null;\n  private tiktoken: any = null;\n\n  constructor(config: ONNXLocalConfig = {}) {\n    this.config = {\n      modelPath: config.modelPath || './models/phi-4/cpu_and_mobile/cpu-int4-rtn-block-32-acc-level-4/model.onnx',\n      executionProviders: config.executionProviders || ['cpu'],\n      maxTokens: config.maxTokens || 100,\n      temperature: config.temperature || 0.7\n    };\n  }\n\n  /**\n   * Load optimized tiktoken tokenizer (cl100k_base for Phi-4)\n   */\n  private async loadTokenizer(): Promise<void> {\n    if (this.tiktoken) return;\n\n    try {\n      // Use cl100k_base encoding (GPT-4, similar to Phi-4)\n      this.tiktoken = get_encoding('cl100k_base');\n\n      console.log('✅ Tokenizer loaded (tiktoken cl100k_base)');\n    } catch (error) {\n      console.error('❌ Failed to load tiktoken:', error);\n      throw new Error(`Tokenizer loading failed: ${error}`);\n    }\n  }\n\n  /**\n   * Encode text using tiktoken (fast BPE)\n   */\n  private encode(text: string): number[] {\n    return Array.from(this.tiktoken.encode(text));\n  }\n\n  /**\n   * Decode tokens using tiktoken\n   */\n  private decode(ids: number[]): string {\n    try {\n      const decoded = this.tiktoken.decode(new Uint32Array(ids));\n      // tiktoken returns buffer, convert to string\n      if (typeof decoded === 'string') {\n        return decoded;\n      } else if (decoded instanceof Uint8Array || decoded instanceof Buffer) {\n        return new TextDecoder().decode(decoded);\n      }\n      return String(decoded);\n    } catch (error) {\n      console.warn('Decode error, returning raw IDs:', error);\n      return ids.join(',');\n    }\n  }\n\n  /**\n   * Initialize ONNX session (with automatic model download)\n   */\n  private async initializeSession(): Promise<void> {\n    if (this.session) return;\n\n    try {\n      // Ensure model is downloaded\n      console.log(`🔍 Checking for Phi-4 ONNX model...`);\n\n      const modelPath = await ensurePhi4Model((progress) => {\n        if (progress.percentage % 10 < 1) { // Log every ~10%\n          console.log(`   📥 Downloading: ${ModelDownloader.formatProgress(progress)}`);\n        }\n      });\n\n      // Update config with actual model path\n      this.config.modelPath = modelPath;\n\n      console.log(`📦 Loading ONNX model: ${this.config.modelPath}`);\n\n      this.session = await ort.InferenceSession.create(\n        this.config.modelPath,\n        {\n          executionProviders: this.config.executionProviders as any,\n          graphOptimizationLevel: 'all',\n          enableCpuMemArena: true,\n          enableMemPattern: true\n        }\n      );\n\n      console.log(`✅ ONNX model loaded`);\n      console.log(`🔧 Execution providers: ${this.config.executionProviders.join(', ')}`);\n\n      // Load tokenizer\n      await this.loadTokenizer();\n\n    } catch (error) {\n      const providerError: ProviderError = {\n        name: 'ONNXInitError',\n        message: `Failed to initialize ONNX model: ${error}`,\n        provider: 'onnx-local',\n        retryable: false\n      };\n      throw providerError;\n    }\n  }\n\n  /**\n   * Format messages for Phi-4 chat template\n   */\n  private formatMessages(messages: Message[]): string {\n    let prompt = '';\n\n    for (const msg of messages) {\n      const content = typeof msg.content === 'string'\n        ? msg.content\n        : msg.content.map(c => c.type === 'text' ? c.text : '').join('');\n\n      if (msg.role === 'system') {\n        prompt += `<|system|>\\n${content}<|end|>\\n`;\n      } else if (msg.role === 'user') {\n        prompt += `<|user|>\\n${content}<|end|>\\n`;\n      } else if (msg.role === 'assistant') {\n        prompt += `<|assistant|>\\n${content}<|end|>\\n`;\n      }\n    }\n\n    prompt += '<|assistant|>\\n';\n    return prompt;\n  }\n\n  /**\n   * Initialize KV cache tensors for all 32 layers\n   * Phi-4 architecture: 32 layers, 8 KV heads, 128 head_dim\n   */\n  private initializeKVCache(batchSize: number, sequenceLength: number) {\n    const numLayers = 32;\n    const numKVHeads = 8;\n    const headDim = 128; // 3072 / 24 = 128\n    const kvCache: Record<string, ort.Tensor> = {};\n\n    // Initialize empty cache for each layer (key and value)\n    for (let i = 0; i < numLayers; i++) {\n      // Empty cache: [batch_size, num_kv_heads, 0, head_dim]\n      const emptyCache = new Float32Array(0);\n\n      kvCache[`past_key_values.${i}.key`] = new ort.Tensor(\n        'float32',\n        emptyCache,\n        [batchSize, numKVHeads, 0, headDim]\n      );\n\n      kvCache[`past_key_values.${i}.value`] = new ort.Tensor(\n        'float32',\n        emptyCache,\n        [batchSize, numKVHeads, 0, headDim]\n      );\n    }\n\n    return kvCache;\n  }\n\n  /**\n   * Chat completion using ONNX with KV cache\n   */\n  async chat(params: ChatParams): Promise<ChatResponse> {\n    await this.initializeSession();\n\n    const startTime = Date.now();\n    const prompt = this.formatMessages(params.messages);\n\n    try {\n      // Tokenize input using optimized tiktoken\n      const inputIds = this.encode(prompt);\n      console.log(`📝 Input tokens: ${inputIds.length}`);\n\n      // Initialize KV cache (reusable for batch)\n      let pastKVCache = this.initializeKVCache(1, 0);\n\n      // Track all generated tokens\n      const allTokenIds = [...inputIds];\n      const outputIds: number[] = [];\n\n      // Pre-allocate tensor buffers for performance\n      const maxSeqLen = inputIds.length + (params.maxTokens || this.config.maxTokens);\n\n      // Autoregressive generation loop\n      const maxNewTokens = params.maxTokens || this.config.maxTokens;\n\n      for (let step = 0; step < maxNewTokens; step++) {\n        // For first step, use all input tokens; for subsequent steps, use only last token\n        const currentInputIds = step === 0 ? inputIds : [outputIds[outputIds.length - 1]];\n        const currentSeqLen = currentInputIds.length;\n\n        // Create input tensor for current step\n        const inputTensor = new ort.Tensor(\n          'int64',\n          BigInt64Array.from(currentInputIds.map(BigInt)),\n          [1, currentSeqLen]\n        );\n\n        // Create attention mask for current step\n        const totalSeqLen = allTokenIds.length;\n        const attentionMask = new ort.Tensor(\n          'int64',\n          BigInt64Array.from(Array(totalSeqLen).fill(1n)),\n          [1, totalSeqLen]\n        );\n\n        // Build feeds with input, attention mask, and KV cache\n        const feeds: Record<string, ort.Tensor> = {\n          input_ids: inputTensor,\n          attention_mask: attentionMask,\n          ...pastKVCache\n        };\n\n        // Run inference\n        const results = await this.session!.run(feeds);\n\n        // Get logits for next token (last position)\n        const logits = results.logits.data as Float32Array;\n        const vocabSize = results.logits.dims[results.logits.dims.length - 1];\n\n        // Extract logits for last token\n        const lastTokenLogitsOffset = (currentSeqLen - 1) * vocabSize;\n\n        // Apply temperature and get next token\n        let nextToken = 0;\n        let maxVal = -Infinity;\n\n        for (let i = 0; i < vocabSize; i++) {\n          const logit = logits[lastTokenLogitsOffset + i] / (params.temperature || this.config.temperature);\n          if (logit > maxVal) {\n            maxVal = logit;\n            nextToken = i;\n          }\n        }\n\n        // Add to output\n        outputIds.push(nextToken);\n        allTokenIds.push(nextToken);\n\n        // Check for end token (2 is typical EOS for Phi models)\n        if (nextToken === 2 || nextToken === 0) {\n          console.log(`🛑 Stop token detected: ${nextToken}`);\n          break;\n        }\n\n        // Update KV cache from outputs for next iteration\n        pastKVCache = {};\n        for (let i = 0; i < 32; i++) {\n          pastKVCache[`past_key_values.${i}.key`] = results[`present.${i}.key`];\n          pastKVCache[`past_key_values.${i}.value`] = results[`present.${i}.value`];\n        }\n\n        // Progress indicator\n        if ((step + 1) % 10 === 0) {\n          console.log(`🔄 Generated ${step + 1} tokens...`);\n        }\n      }\n\n      // Decode output using optimized tiktoken\n      const generatedText = this.decode(outputIds);\n      const latency = Date.now() - startTime;\n      const tokensPerSecond = (outputIds.length / (latency / 1000)).toFixed(1);\n\n      console.log(`✅ Generated: ${generatedText}`);\n      console.log(`⏱️  Latency: ${latency}ms (${tokensPerSecond} tokens/sec)`);\n\n      const content: ContentBlock[] = [{\n        type: 'text',\n        text: generatedText.trim()\n      }];\n\n      return {\n        id: `onnx-local-${Date.now()}`,\n        model: this.config.modelPath,\n        content,\n        stopReason: 'end_turn',\n        usage: {\n          inputTokens: inputIds.length,\n          outputTokens: outputIds.length\n        },\n        metadata: {\n          provider: 'onnx-local',\n          model: 'Phi-4-mini-instruct-onnx',\n          latency,\n          cost: 0, // Local inference is free\n          executionProviders: this.config.executionProviders,\n          tokensPerSecond: parseFloat(tokensPerSecond)\n        }\n      };\n\n    } catch (error) {\n      const providerError: ProviderError = {\n        name: 'ONNXInferenceError',\n        message: `ONNX inference failed: ${error}`,\n        provider: 'onnx-local',\n        retryable: true\n      };\n      throw providerError;\n    }\n  }\n\n  /**\n   * Streaming not implemented (requires complex generation loop)\n   */\n  async *stream(params: ChatParams): AsyncGenerator<StreamChunk> {\n    throw new Error('Streaming not yet implemented for ONNX local inference');\n  }\n\n  /**\n   * Validate capabilities\n   */\n  validateCapabilities(features: string[]): boolean {\n    const supported = ['chat'];\n    return features.every(f => supported.includes(f));\n  }\n\n  /**\n   * Get model info\n   */\n  getModelInfo() {\n    return {\n      modelPath: this.config.modelPath,\n      executionProviders: this.config.executionProviders,\n      initialized: this.session !== null,\n      tokenizerLoaded: this.tiktoken !== null\n    };\n  }\n\n  /**\n   * Cleanup resources\n   */\n  async dispose(): Promise<void> {\n    if (this.session) {\n      // ONNX Runtime sessions don't have explicit disposal in Node.js\n      this.session = null;\n    }\n    if (this.tiktoken) {\n      this.tiktoken.free();\n      this.tiktoken = null;\n    }\n  }\n}\n"]}
+{"version":3,"file":"onnx-local.js","sourceRoot":"","sources":["../../../src/router/providers/onnx-local.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,GAAG,MAAM,kBAAkB,CAAC;AAGxC,OAAO,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AACxC,OAAO,EAAE,eAAe,EAAE,eAAe,EAAE,MAAM,iCAAiC,CAAC;AAkBnF,MAAM,OAAO,iBAAiB;IAC5B,IAAI,GAAG,YAAY,CAAC;IACpB,IAAI,GAAG,QAAiB,CAAC;IACzB,iBAAiB,GAAG,KAAK,CAAC,CAAC,mDAAmD;IAC9E,aAAa,GAAG,KAAK,CAAC;IACtB,WAAW,GAAG,KAAK,CAAC;IAEZ,OAAO,GAAgC,IAAI,CAAC;IAC5C,MAAM,CAA4B;IAClC,SAAS,GAAQ,IAAI,CAAC;IACtB,QAAQ,GAAQ,IAAI,CAAC;IAE7B,YAAY,SAA0B,EAAE;QACtC,IAAI,CAAC,MAAM,GAAG;YACZ,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,4EAA4E;YAC3G,kBAAkB,EAAE,MAAM,CAAC,kBAAkB,IAAI,CAAC,KAAK,CAAC;YACxD,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,GAAG;YAClC,WAAW,EAAE,MAAM,CAAC,WAAW,IAAI,GAAG;SACvC,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,aAAa;QACzB,IAAI,IAAI,CAAC,QAAQ;YAAE,OAAO;QAE1B,IAAI,CAAC;YACH,qDAAqD;YACrD,IAAI,CAAC,QAAQ,GAAG,YAAY,CAAC,aAAa,CAAC,CAAC;YAE5C,OAAO,CAAC,GAAG,CAAC,2CAA2C,CAAC,CAAC;QAC3D,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,CAAC,KAAK,CAAC,4BAA4B,EAAE,KAAK,CAAC,CAAC;YACnD,MAAM,IAAI,KAAK,CAAC,6BAA6B,KAAK,EAAE,CAAC,CAAC;QACxD,CAAC;IACH,CAAC;IAED;;OAEG;IACK,MAAM,CAAC,IAAY;QACzB,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;IAChD,CAAC;IAED;;OAEG;IACK,MAAM,CAAC,GAAa;QAC1B,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC;YAC3D,6CAA6C;YAC7C,IAAI,OAAO,OAAO,KAAK,QAAQ,EAAE,CAAC;gBAChC,OAAO,OAAO,CAAC;YACjB,CAAC;iBAAM,IAAI,OAAO,YAAY,UAAU,IAAI,OAAO,YAAY,MAAM,EAAE,CAAC;gBACtE,OAAO,IAAI,WAAW,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YAC3C,CAAC;YACD,OAAO,MAAM,CAAC,OAAO,CAAC,CAAC;QACzB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,CAAC,IAAI,CAAC,kCAAkC,EAAE,KAAK,CAAC,CAAC;YACxD,OAAO,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACvB,CAAC;IACH,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,iBAAiB;QAC7B,IAAI,IAAI,CAAC,OAAO;YAAE,OAAO;QAEzB,IAAI,CAAC;YACH,6BAA6B;YAC7B,OAAO,CAAC,GAAG,CAAC,qCAAqC,CAAC,CAAC;YAEnD,MAAM,SAAS,GAAG,MAAM,eAAe,CAAC,CAAC,QAAQ,EAAE,EAAE;gBACnD,IAAI,QAAQ,CAAC,UAAU,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC,iBAAiB;oBACnD,OAAO,CAAC,GAAG,CAAC,sBAAsB,eAAe,CAAC,cAAc,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;gBAChF,CAAC;YACH,CAAC,CAAC,CAAC;YAEH,uCAAuC;YACvC,IAAI,CAAC,MAAM,CAAC,SAAS,GAAG,SAAS,CAAC;YAElC,OAAO,CAAC,GAAG,CAAC,0BAA0B,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC;YAE/D,IAAI,CAAC,OAAO,GAAG,MAAM,GAAG,CAAC,gBAAgB,CAAC,MAAM,CAC9C,IAAI,CAAC,MAAM,CAAC,SAAS,EACrB;gBACE,kBAAkB,EAAE,IAAI,CAAC,MAAM,CAAC,kBAAyB;gBACzD,sBAAsB,EAAE,KAAK;gBAC7B,iBAAiB,EAAE,IAAI;gBACvB,gBAAgB,EAAE,IAAI;aACvB,CACF,CAAC;YAEF,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC,CAAC;YACnC,OAAO,CAAC,GAAG,CAAC,2BAA2B,IAAI,CAAC,MAAM,CAAC,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YAEpF,iBAAiB;YACjB,MAAM,IAAI,CAAC,aAAa,EAAE,CAAC;QAE7B,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,aAAa,GAAkB;gBACnC,IAAI,EAAE,eAAe;gBACrB,OAAO,EAAE,oCAAoC,KAAK,EAAE;gBACpD,QAAQ,EAAE,YAAY;gBACtB,SAAS,EAAE,KAAK;aACjB,CAAC;YACF,MAAM,aAAa,CAAC;QACtB,CAAC;IACH,CAAC;IAED;;OAEG;IACK,cAAc,CAAC,QAAmB;QACxC,IAAI,MAAM,GAAG,EAAE,CAAC;QAEhB,KAAK,MAAM,GAAG,IAAI,QAAQ,EAAE,CAAC;YAC3B,MAAM,OAAO,GAAG,OAAO,GAAG,CAAC,OAAO,KAAK,QAAQ;gBAC7C,CAAC,CAAC,GAAG,CAAC,OAAO;gBACb,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YAEnE,IAAI,GAAG,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;gBAC1B,MAAM,IAAI,eAAe,OAAO,WAAW,CAAC;YAC9C,CAAC;iBAAM,IAAI,GAAG,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBAC/B,MAAM,IAAI,aAAa,OAAO,WAAW,CAAC;YAC5C,CAAC;iBAAM,IAAI,GAAG,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;gBACpC,MAAM,IAAI,kBAAkB,OAAO,WAAW,CAAC;YACjD,CAAC;QACH,CAAC;QAED,MAAM,IAAI,iBAAiB,CAAC;QAC5B,OAAO,MAAM,CAAC;IAChB,CAAC;IAED;;;OAGG;IACK,iBAAiB,CAAC,SAAiB,EAAE,cAAsB;QACjE,MAAM,SAAS,GAAG,EAAE,CAAC;QACrB,MAAM,UAAU,GAAG,CAAC,CAAC;QACrB,MAAM,OAAO,GAAG,GAAG,CAAC,CAAC,kBAAkB;QACvC,oEAAoE;QACpE,MAAM,OAAO,GAAwB,EAAE,CAAC;QAExC,wDAAwD;QACxD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,EAAE,CAAC,EAAE,EAAE,CAAC;YACnC,uDAAuD;YACvD,MAAM,UAAU,GAAG,IAAI,YAAY,CAAC,CAAC,CAAC,CAAC;YAEvC,0EAA0E;YAC1E,OAAO,CAAC,mBAAmB,CAAC,MAAM,CAAC,GAAG,IAAK,GAAG,CAAC,MAAc,CAC3D,SAAS,EACT,UAAU,EACV,CAAC,SAAS,EAAE,UAAU,EAAE,CAAC,EAAE,OAAO,CAAC,CACpC,CAAC;YAEF,OAAO,CAAC,mBAAmB,CAAC,QAAQ,CAAC,GAAG,IAAK,GAAG,CAAC,MAAc,CAC7D,SAAS,EACT,UAAU,EACV,CAAC,SAAS,EAAE,UAAU,EAAE,CAAC,EAAE,OAAO,CAAC,CACpC,CAAC;QACJ,CAAC;QAED,OAAO,OAAO,CAAC;IACjB,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,IAAI,CAAC,MAAkB;QAC3B,MAAM,IAAI,CAAC,iBAAiB,EAAE,CAAC;QAE/B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,MAAM,MAAM,GAAG,IAAI,CAAC,cAAc,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QAEpD,IAAI,CAAC;YACH,0CAA0C;YAC1C,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;YACrC,OAAO,CAAC,GAAG,CAAC,oBAAoB,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;YAEnD,2CAA2C;YAC3C,IAAI,WAAW,GAAG,IAAI,CAAC,iBAAiB,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAE/C,6BAA6B;YAC7B,MAAM,WAAW,GAAG,CAAC,GAAG,QAAQ,CAAC,CAAC;YAClC,MAAM,SAAS,GAAa,EAAE,CAAC;YAE/B,8CAA8C;YAC9C,MAAM,SAAS,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,MAAM,CAAC,SAAS,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;YAEhF,iCAAiC;YACjC,MAAM,YAAY,GAAG,MAAM,CAAC,SAAS,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC;YAE/D,KAAK,IAAI,IAAI,GAAG,CAAC,EAAE,IAAI,GAAG,YAAY,EAAE,IAAI,EAAE,EAAE,CAAC;gBAC/C,kFAAkF;gBAClF,MAAM,eAAe,GAAG,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC;gBAClF,MAAM,aAAa,GAAG,eAAe,CAAC,MAAM,CAAC;gBAE7C,0EAA0E;gBAC1E,MAAM,WAAW,GAAG,IAAK,GAAG,CAAC,MAAc,CACzC,OAAO,EACP,aAAa,CAAC,IAAI,CAAC,eAAe,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,EAC/C,CAAC,CAAC,EAAE,aAAa,CAAC,CACnB,CAAC;gBAEF,yCAAyC;gBACzC,MAAM,WAAW,GAAG,WAAW,CAAC,MAAM,CAAC;gBACvC,MAAM,aAAa,GAAG,IAAK,GAAG,CAAC,MAAc,CAC3C,OAAO,EACP,aAAa,CAAC,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,EAC/C,CAAC,CAAC,EAAE,WAAW,CAAC,CACjB,CAAC;gBAEF,uDAAuD;gBACvD,MAAM,KAAK,GAAwB;oBACjC,SAAS,EAAE,WAAW;oBACtB,cAAc,EAAE,aAAa;oBAC7B,GAAG,WAAW;iBACf,CAAC;gBAEF,gBAAgB;gBAChB,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;gBAE/C,4CAA4C;gBAC5C,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,IAAoB,CAAC;gBACnD,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;gBAEtE,gCAAgC;gBAChC,MAAM,qBAAqB,GAAG,CAAC,aAAa,GAAG,CAAC,CAAC,GAAG,SAAS,CAAC;gBAE9D,uCAAuC;gBACvC,IAAI,SAAS,GAAG,CAAC,CAAC;gBAClB,IAAI,MAAM,GAAG,CAAC,QAAQ,CAAC;gBAEvB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,EAAE,CAAC,EAAE,EAAE,CAAC;oBACnC,MAAM,KAAK,GAAG,MAAM,CAAC,qBAAqB,GAAG,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,WAAW,IAAI,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;oBAClG,IAAI,KAAK,GAAG,MAAM,EAAE,CAAC;wBACnB,MAAM,GAAG,KAAK,CAAC;wBACf,SAAS,GAAG,CAAC,CAAC;oBAChB,CAAC;gBACH,CAAC;gBAED,gBAAgB;gBAChB,SAAS,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;gBAC1B,WAAW,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;gBAE5B,wDAAwD;gBACxD,IAAI,SAAS,KAAK,CAAC,IAAI,SAAS,KAAK,CAAC,EAAE,CAAC;oBACvC,OAAO,CAAC,GAAG,CAAC,2BAA2B,SAAS,EAAE,CAAC,CAAC;oBACpD,MAAM;gBACR,CAAC;gBAED,kDAAkD;gBAClD,WAAW,GAAG,EAAE,CAAC;gBACjB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;oBAC5B,WAAW,CAAC,mBAAmB,CAAC,MAAM,CAAC,GAAG,OAAO,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC;oBACtE,WAAW,CAAC,mBAAmB,CAAC,QAAQ,CAAC,GAAG,OAAO,CAAC,WAAW,CAAC,QAAQ,CAAC,CAAC;gBAC5E,CAAC;gBAED,qBAAqB;gBACrB,IAAI,CAAC,IAAI,GAAG,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,CAAC;oBAC1B,OAAO,CAAC,GAAG,CAAC,gBAAgB,IAAI,GAAG,CAAC,YAAY,CAAC,CAAC;gBACpD,CAAC;YACH,CAAC;YAED,yCAAyC;YACzC,MAAM,aAAa,GAAG,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;YAC7C,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;YACvC,MAAM,eAAe,GAAG,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,OAAO,GAAG,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YAEzE,OAAO,CAAC,GAAG,CAAC,gBAAgB,aAAa,EAAE,CAAC,CAAC;YAC7C,OAAO,CAAC,GAAG,CAAC,gBAAgB,OAAO,OAAO,eAAe,cAAc,CAAC,CAAC;YAEzE,MAAM,OAAO,GAAmB,CAAC;oBAC/B,IAAI,EAAE,MAAM;oBACZ,IAAI,EAAE,aAAa,CAAC,IAAI,EAAE;iBAC3B,CAAC,CAAC;YAEH,OAAO;gBACL,EAAE,EAAE,cAAc,IAAI,CAAC,GAAG,EAAE,EAAE;gBAC9B,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,SAAS;gBAC5B,OAAO;gBACP,UAAU,EAAE,UAAU;gBACtB,KAAK,EAAE;oBACL,WAAW,EAAE,QAAQ,CAAC,MAAM;oBAC5B,YAAY,EAAE,SAAS,CAAC,MAAM;iBAC/B;gBACD,QAAQ,EAAE;oBACR,QAAQ,EAAE,YAAY;oBACtB,KAAK,EAAE,0BAA0B;oBACjC,OAAO;oBACP,IAAI,EAAE,CAAC,EAAE,0BAA0B;oBACnC,kBAAkB,EAAE,IAAI,CAAC,MAAM,CAAC,kBAAkB;oBAClD,eAAe,EAAE,UAAU,CAAC,eAAe,CAAC;iBAC7C;aACF,CAAC;QAEJ,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,aAAa,GAAkB;gBACnC,IAAI,EAAE,oBAAoB;gBAC1B,OAAO,EAAE,0BAA0B,KAAK,EAAE;gBAC1C,QAAQ,EAAE,YAAY;gBACtB,SAAS,EAAE,IAAI;aAChB,CAAC;YACF,MAAM,aAAa,CAAC;QACtB,CAAC;IACH,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,CAAC,MAAM,CAAC,MAAkB;QAC9B,MAAM,IAAI,KAAK,CAAC,wDAAwD,CAAC,CAAC;IAC5E,CAAC;IAED;;OAEG;IACH,oBAAoB,CAAC,QAAkB;QACrC,MAAM,SAAS,GAAG,CAAC,MAAM,CAAC,CAAC;QAC3B,OAAO,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IACpD,CAAC;IAED;;OAEG;IACH,YAAY;QACV,OAAO;YACL,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,SAAS;YAChC,kBAAkB,EAAE,IAAI,CAAC,MAAM,CAAC,kBAAkB;YAClD,WAAW,EAAE,IAAI,CAAC,OAAO,KAAK,IAAI;YAClC,eAAe,EAAE,IAAI,CAAC,QAAQ,KAAK,IAAI;SACxC,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,OAAO;QACX,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YACjB,gEAAgE;YAChE,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC;QACtB,CAAC;QACD,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YAClB,IAAI,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;YACrB,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC;QACvB,CAAC;IACH,CAAC;CACF","sourcesContent":["/**\n * ONNX Runtime Local Inference Provider for Phi-4\n *\n * Uses onnxruntime-node for true local CPU/GPU inference\n */\n\nimport * as ort from 'onnxruntime-node';\nimport * as fs from 'fs';\nimport * as path from 'path';\nimport { get_encoding } from 'tiktoken';\nimport { ensurePhi4Model, ModelDownloader } from '../../utils/model-downloader.js';\nimport type {\n  LLMProvider,\n  ChatParams,\n  ChatResponse,\n  StreamChunk,\n  ProviderError,\n  Message,\n  ContentBlock\n} from '../types.js';\n\nexport interface ONNXLocalConfig {\n  modelPath?: string;\n  executionProviders?: string[];\n  maxTokens?: number;\n  temperature?: number;\n}\n\nexport class ONNXLocalProvider implements LLMProvider {\n  name = 'onnx-local';\n  type = 'custom' as const;\n  supportsStreaming = false; // Streaming requires complex token generation loop\n  supportsTools = false;\n  supportsMCP = false;\n\n  private session: ort.InferenceSession | null = null;\n  private config: Required<ONNXLocalConfig>;\n  private tokenizer: any = null;\n  private tiktoken: any = null;\n\n  constructor(config: ONNXLocalConfig = {}) {\n    this.config = {\n      modelPath: config.modelPath || './models/phi-4/cpu_and_mobile/cpu-int4-rtn-block-32-acc-level-4/model.onnx',\n      executionProviders: config.executionProviders || ['cpu'],\n      maxTokens: config.maxTokens || 100,\n      temperature: config.temperature || 0.7\n    };\n  }\n\n  /**\n   * Load optimized tiktoken tokenizer (cl100k_base for Phi-4)\n   */\n  private async loadTokenizer(): Promise<void> {\n    if (this.tiktoken) return;\n\n    try {\n      // Use cl100k_base encoding (GPT-4, similar to Phi-4)\n      this.tiktoken = get_encoding('cl100k_base');\n\n      console.log('✅ Tokenizer loaded (tiktoken cl100k_base)');\n    } catch (error) {\n      console.error('❌ Failed to load tiktoken:', error);\n      throw new Error(`Tokenizer loading failed: ${error}`);\n    }\n  }\n\n  /**\n   * Encode text using tiktoken (fast BPE)\n   */\n  private encode(text: string): number[] {\n    return Array.from(this.tiktoken.encode(text));\n  }\n\n  /**\n   * Decode tokens using tiktoken\n   */\n  private decode(ids: number[]): string {\n    try {\n      const decoded = this.tiktoken.decode(new Uint32Array(ids));\n      // tiktoken returns buffer, convert to string\n      if (typeof decoded === 'string') {\n        return decoded;\n      } else if (decoded instanceof Uint8Array || decoded instanceof Buffer) {\n        return new TextDecoder().decode(decoded);\n      }\n      return String(decoded);\n    } catch (error) {\n      console.warn('Decode error, returning raw IDs:', error);\n      return ids.join(',');\n    }\n  }\n\n  /**\n   * Initialize ONNX session (with automatic model download)\n   */\n  private async initializeSession(): Promise<void> {\n    if (this.session) return;\n\n    try {\n      // Ensure model is downloaded\n      console.log(`🔍 Checking for Phi-4 ONNX model...`);\n\n      const modelPath = await ensurePhi4Model((progress) => {\n        if (progress.percentage % 10 < 1) { // Log every ~10%\n          console.log(`   📥 Downloading: ${ModelDownloader.formatProgress(progress)}`);\n        }\n      });\n\n      // Update config with actual model path\n      this.config.modelPath = modelPath;\n\n      console.log(`📦 Loading ONNX model: ${this.config.modelPath}`);\n\n      this.session = await ort.InferenceSession.create(\n        this.config.modelPath,\n        {\n          executionProviders: this.config.executionProviders as any,\n          graphOptimizationLevel: 'all',\n          enableCpuMemArena: true,\n          enableMemPattern: true\n        }\n      );\n\n      console.log(`✅ ONNX model loaded`);\n      console.log(`🔧 Execution providers: ${this.config.executionProviders.join(', ')}`);\n\n      // Load tokenizer\n      await this.loadTokenizer();\n\n    } catch (error) {\n      const providerError: ProviderError = {\n        name: 'ONNXInitError',\n        message: `Failed to initialize ONNX model: ${error}`,\n        provider: 'onnx-local',\n        retryable: false\n      };\n      throw providerError;\n    }\n  }\n\n  /**\n   * Format messages for Phi-4 chat template\n   */\n  private formatMessages(messages: Message[]): string {\n    let prompt = '';\n\n    for (const msg of messages) {\n      const content = typeof msg.content === 'string'\n        ? msg.content\n        : msg.content.map(c => c.type === 'text' ? c.text : '').join('');\n\n      if (msg.role === 'system') {\n        prompt += `<|system|>\\n${content}<|end|>\\n`;\n      } else if (msg.role === 'user') {\n        prompt += `<|user|>\\n${content}<|end|>\\n`;\n      } else if (msg.role === 'assistant') {\n        prompt += `<|assistant|>\\n${content}<|end|>\\n`;\n      }\n    }\n\n    prompt += '<|assistant|>\\n';\n    return prompt;\n  }\n\n  /**\n   * Initialize KV cache tensors for all 32 layers\n   * Phi-4 architecture: 32 layers, 8 KV heads, 128 head_dim\n   */\n  private initializeKVCache(batchSize: number, sequenceLength: number) {\n    const numLayers = 32;\n    const numKVHeads = 8;\n    const headDim = 128; // 3072 / 24 = 128\n    // Use any type for KV cache since ort.Tensor is a value, not a type\n    const kvCache: Record<string, any> = {};\n\n    // Initialize empty cache for each layer (key and value)\n    for (let i = 0; i < numLayers; i++) {\n      // Empty cache: [batch_size, num_kv_heads, 0, head_dim]\n      const emptyCache = new Float32Array(0);\n\n      // Cast to any to avoid TS issues with onnxruntime-node Tensor constructor\n      kvCache[`past_key_values.${i}.key`] = new (ort.Tensor as any)(\n        'float32',\n        emptyCache,\n        [batchSize, numKVHeads, 0, headDim]\n      );\n\n      kvCache[`past_key_values.${i}.value`] = new (ort.Tensor as any)(\n        'float32',\n        emptyCache,\n        [batchSize, numKVHeads, 0, headDim]\n      );\n    }\n\n    return kvCache;\n  }\n\n  /**\n   * Chat completion using ONNX with KV cache\n   */\n  async chat(params: ChatParams): Promise<ChatResponse> {\n    await this.initializeSession();\n\n    const startTime = Date.now();\n    const prompt = this.formatMessages(params.messages);\n\n    try {\n      // Tokenize input using optimized tiktoken\n      const inputIds = this.encode(prompt);\n      console.log(`📝 Input tokens: ${inputIds.length}`);\n\n      // Initialize KV cache (reusable for batch)\n      let pastKVCache = this.initializeKVCache(1, 0);\n\n      // Track all generated tokens\n      const allTokenIds = [...inputIds];\n      const outputIds: number[] = [];\n\n      // Pre-allocate tensor buffers for performance\n      const maxSeqLen = inputIds.length + (params.maxTokens || this.config.maxTokens);\n\n      // Autoregressive generation loop\n      const maxNewTokens = params.maxTokens || this.config.maxTokens;\n\n      for (let step = 0; step < maxNewTokens; step++) {\n        // For first step, use all input tokens; for subsequent steps, use only last token\n        const currentInputIds = step === 0 ? inputIds : [outputIds[outputIds.length - 1]];\n        const currentSeqLen = currentInputIds.length;\n\n        // Create input tensor for current step (cast to any for TS compatibility)\n        const inputTensor = new (ort.Tensor as any)(\n          'int64',\n          BigInt64Array.from(currentInputIds.map(BigInt)),\n          [1, currentSeqLen]\n        );\n\n        // Create attention mask for current step\n        const totalSeqLen = allTokenIds.length;\n        const attentionMask = new (ort.Tensor as any)(\n          'int64',\n          BigInt64Array.from(Array(totalSeqLen).fill(1n)),\n          [1, totalSeqLen]\n        );\n\n        // Build feeds with input, attention mask, and KV cache\n        const feeds: Record<string, any> = {\n          input_ids: inputTensor,\n          attention_mask: attentionMask,\n          ...pastKVCache\n        };\n\n        // Run inference\n        const results = await this.session!.run(feeds);\n\n        // Get logits for next token (last position)\n        const logits = results.logits.data as Float32Array;\n        const vocabSize = results.logits.dims[results.logits.dims.length - 1];\n\n        // Extract logits for last token\n        const lastTokenLogitsOffset = (currentSeqLen - 1) * vocabSize;\n\n        // Apply temperature and get next token\n        let nextToken = 0;\n        let maxVal = -Infinity;\n\n        for (let i = 0; i < vocabSize; i++) {\n          const logit = logits[lastTokenLogitsOffset + i] / (params.temperature || this.config.temperature);\n          if (logit > maxVal) {\n            maxVal = logit;\n            nextToken = i;\n          }\n        }\n\n        // Add to output\n        outputIds.push(nextToken);\n        allTokenIds.push(nextToken);\n\n        // Check for end token (2 is typical EOS for Phi models)\n        if (nextToken === 2 || nextToken === 0) {\n          console.log(`🛑 Stop token detected: ${nextToken}`);\n          break;\n        }\n\n        // Update KV cache from outputs for next iteration\n        pastKVCache = {};\n        for (let i = 0; i < 32; i++) {\n          pastKVCache[`past_key_values.${i}.key`] = results[`present.${i}.key`];\n          pastKVCache[`past_key_values.${i}.value`] = results[`present.${i}.value`];\n        }\n\n        // Progress indicator\n        if ((step + 1) % 10 === 0) {\n          console.log(`🔄 Generated ${step + 1} tokens...`);\n        }\n      }\n\n      // Decode output using optimized tiktoken\n      const generatedText = this.decode(outputIds);\n      const latency = Date.now() - startTime;\n      const tokensPerSecond = (outputIds.length / (latency / 1000)).toFixed(1);\n\n      console.log(`✅ Generated: ${generatedText}`);\n      console.log(`⏱️  Latency: ${latency}ms (${tokensPerSecond} tokens/sec)`);\n\n      const content: ContentBlock[] = [{\n        type: 'text',\n        text: generatedText.trim()\n      }];\n\n      return {\n        id: `onnx-local-${Date.now()}`,\n        model: this.config.modelPath,\n        content,\n        stopReason: 'end_turn',\n        usage: {\n          inputTokens: inputIds.length,\n          outputTokens: outputIds.length\n        },\n        metadata: {\n          provider: 'onnx-local',\n          model: 'Phi-4-mini-instruct-onnx',\n          latency,\n          cost: 0, // Local inference is free\n          executionProviders: this.config.executionProviders,\n          tokensPerSecond: parseFloat(tokensPerSecond)\n        }\n      };\n\n    } catch (error) {\n      const providerError: ProviderError = {\n        name: 'ONNXInferenceError',\n        message: `ONNX inference failed: ${error}`,\n        provider: 'onnx-local',\n        retryable: true\n      };\n      throw providerError;\n    }\n  }\n\n  /**\n   * Streaming not implemented (requires complex generation loop)\n   */\n  async *stream(params: ChatParams): AsyncGenerator<StreamChunk> {\n    throw new Error('Streaming not yet implemented for ONNX local inference');\n  }\n\n  /**\n   * Validate capabilities\n   */\n  validateCapabilities(features: string[]): boolean {\n    const supported = ['chat'];\n    return features.every(f => supported.includes(f));\n  }\n\n  /**\n   * Get model info\n   */\n  getModelInfo() {\n    return {\n      modelPath: this.config.modelPath,\n      executionProviders: this.config.executionProviders,\n      initialized: this.session !== null,\n      tokenizerLoaded: this.tiktoken !== null\n    };\n  }\n\n  /**\n   * Cleanup resources\n   */\n  async dispose(): Promise<void> {\n    if (this.session) {\n      // ONNX Runtime sessions don't have explicit disposal in Node.js\n      this.session = null;\n    }\n    if (this.tiktoken) {\n      this.tiktoken.free();\n      this.tiktoken = null;\n    }\n  }\n}\n"]}

package/dist/routing/CircuitBreakerRouter.d.ts

@@ -0,0 +1,187 @@
+/**
+ * Circuit Breaker Router - Fault-Tolerant Agent Routing
+ *
+ * Integrates @ruvector/tiny-dancer circuit breaker pattern for 99.9% uptime.
+ *
+ * Features:
+ * - Circuit breaker states: CLOSED, OPEN, HALF_OPEN
+ * - Automatic failure detection and recovery
+ * - Fallback chains for degraded service
+ * - Hot-reload capability for configuration updates
+ * - Uncertainty estimation for routing decisions
+ * - Rate limiting for request protection
+ *
+ * Performance:
+ * - <5ms routing overhead
+ * - 99.9% uptime guarantee
+ * - Automatic failover in <100ms
+ */
+export interface CircuitBreakerConfig {
+    /** Failure threshold before opening circuit (default: 5) */
+    failureThreshold?: number;
+    /** Success threshold to close circuit from half-open (default: 3) */
+    successThreshold?: number;
+    /** Timeout before attempting recovery (ms, default: 30000) */
+    resetTimeout?: number;
+    /** Request timeout (ms, default: 5000) */
+    requestTimeout?: number;
+    /** Enable uncertainty estimation (default: true) */
+    enableUncertaintyEstimation?: boolean;
+}
+export declare enum CircuitState {
+    CLOSED = "CLOSED",// Normal operation
+    OPEN = "OPEN",// Failures detected, routing blocked
+    HALF_OPEN = "HALF_OPEN"
+}
+export interface RouteRequest {
+    taskDescription: string;
+    preferredAgent?: string;
+    fallbackAgents?: string[];
+    timeout?: number;
+}
+export interface RouteResult {
+    selectedAgent: string;
+    confidence: number;
+    circuitState: CircuitState;
+    fallbackUsed: boolean;
+    uncertainty?: number;
+    metrics: {
+        routingTimeMs: number;
+        failureCount: number;
+        successCount: number;
+    };
+}
+export interface AgentHealth {
+    agent: string;
+    state: CircuitState;
+    failureCount: number;
+    successCount: number;
+    lastFailureTime?: number;
+    lastSuccessTime?: number;
+    availability: number;
+}
+/**
+ * Circuit Breaker Router
+ *
+ * Provides fault-tolerant routing with:
+ * 1. Circuit breaker pattern for each agent
+ * 2. Automatic fallback chains
+ * 3. Health monitoring and recovery
+ * 4. Uncertainty-aware decision making
+ */
+export declare class CircuitBreakerRouter {
+    private config;
+    private circuitStates;
+    private failureCounts;
+    private successCounts;
+    private lastFailureTimes;
+    private lastSuccessTimes;
+    private resetTimers;
+    private rateLimiter;
+    private routingMetrics;
+    constructor(config?: CircuitBreakerConfig);
+    /**
+     * Route request to agent with circuit breaker protection
+     *
+     * Process:
+     * 1. Check circuit state for preferred agent
+     * 2. If circuit OPEN, try fallback chain
+     * 3. If circuit HALF_OPEN, allow test request
+     * 4. Estimate uncertainty if enabled
+     *
+     * @param request - Route request with task and agent preferences
+     * @returns Route result with selected agent and metrics
+     */
+    route(request: RouteRequest): Promise<RouteResult>;
+    /**
+     * Record successful agent execution
+     *
+     * Updates circuit breaker state:
+     * - Increment success count
+     * - Reset failure count if threshold reached
+     * - Transition HALF_OPEN -> CLOSED if successful
+     *
+     * @param agent - Agent that succeeded
+     */
+    recordSuccess(agent: string): void;
+    /**
+     * Record failed agent execution
+     *
+     * Updates circuit breaker state:
+     * - Increment failure count
+     * - Transition CLOSED -> OPEN if threshold exceeded
+     * - Transition HALF_OPEN -> OPEN on failure
+     * - Schedule automatic reset
+     *
+     * @param agent - Agent that failed
+     */
+    recordFailure(agent: string): void;
+    /**
+     * Get circuit state for agent
+     *
+     * @param agent - Agent name
+     * @returns Current circuit state
+     */
+    getCircuitState(agent: string): CircuitState;
+    /**
+     * Get health status for all agents
+     *
+     * @returns Array of agent health metrics
+     */
+    getAgentHealth(): AgentHealth[];
+    /**
+     * Get routing metrics
+     *
+     * @returns Cumulative routing statistics
+     */
+    getMetrics(): typeof this.routingMetrics;
+    /**
+     * Manually reset circuit for agent
+     *
+     * @param agent - Agent to reset
+     */
+    resetCircuit(agent: string): void;
+    /**
+     * Hot-reload configuration
+     *
+     * Allows updating circuit breaker parameters without restart.
+     *
+     * @param config - New configuration
+     */
+    updateConfig(config: Partial<CircuitBreakerConfig>): void;
+    /**
+     * Open circuit and schedule automatic reset
+     */
+    private openCircuit;
+    /**
+     * Calculate confidence score
+     */
+    private calculateConfidence;
+    /**
+     * Estimate uncertainty for routing decision
+     *
+     * Uses failure history and circuit state to estimate decision uncertainty.
+     * Higher uncertainty indicates less reliable routing.
+     *
+     * @param agent - Agent to estimate uncertainty for
+     * @returns Uncertainty score (0-1, lower is better)
+     */
+    private estimateUncertainty;
+    /**
+     * Calculate recency factor based on time since last failure
+     */
+    private getRecencyFactor;
+    /**
+     * Update average routing time (exponential moving average)
+     */
+    private updateAvgRoutingTime;
+    /**
+     * Simple string hash function for rate limiting
+     */
+    private hashString;
+    /**
+     * Cleanup resources on shutdown
+     */
+    destroy(): void;
+}
+//# sourceMappingURL=CircuitBreakerRouter.d.ts.map

package/dist/routing/CircuitBreakerRouter.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"CircuitBreakerRouter.d.ts","sourceRoot":"","sources":["../../src/routing/CircuitBreakerRouter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAMH,MAAM,WAAW,oBAAoB;IACnC,4DAA4D;IAC5D,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,qEAAqE;IACrE,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,8DAA8D;IAC9D,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,0CAA0C;IAC1C,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,oDAAoD;IACpD,2BAA2B,CAAC,EAAE,OAAO,CAAC;CACvC;AAED,oBAAY,YAAY;IACtB,MAAM,WAAW,CAAM,mBAAmB;IAC1C,IAAI,SAAS,CAAU,qCAAqC;IAC5D,SAAS,cAAc;CACxB;AAED,MAAM,WAAW,YAAY;IAC3B,eAAe,EAAE,MAAM,CAAC;IACxB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,WAAW;IAC1B,aAAa,EAAE,MAAM,CAAC;IACtB,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,YAAY,CAAC;IAC3B,YAAY,EAAE,OAAO,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE;QACP,aAAa,EAAE,MAAM,CAAC;QACtB,YAAY,EAAE,MAAM,CAAC;QACrB,YAAY,EAAE,MAAM,CAAC;KACtB,CAAC;CACH;AAED,MAAM,WAAW,WAAW;IAC1B,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,YAAY,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,YAAY,EAAE,MAAM,CAAC;CACtB;AAED;;;;;;;;GAQG;AACH,qBAAa,oBAAoB;IAC/B,OAAO,CAAC,MAAM,CAAiC;IAG/C,OAAO,CAAC,aAAa,CAA4B;IACjD,OAAO,CAAC,aAAa,CAAsB;IAC3C,OAAO,CAAC,aAAa,CAAsB;IAC3C,OAAO,CAAC,gBAAgB,CAAsB;IAC9C,OAAO,CAAC,gBAAgB,CAAsB;IAC9C,OAAO,CAAC,WAAW,CAA8B;IAGjD,OAAO,CAAC,WAAW,CAAc;IAGjC,OAAO,CAAC,cAAc,CAMpB;gBAEU,MAAM,CAAC,EAAE,oBAAoB;IAmEzC;;;;;;;;;;;OAWG;IACG,KAAK,CAAC,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,WAAW,CAAC;IAiHxD;;;;;;;;;OASG;IACH,aAAa,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI;IAmBlC;;;;;;;;;;OAUG;IACH,aAAa,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI;IAoBlC;;;;;OAKG;IACH,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,YAAY;IAI5C;;;;OAIG;IACH,cAAc,IAAI,WAAW,EAAE;IA0B/B;;;;OAIG;IACH,UAAU,IAAI,OAAO,IAAI,CAAC,cAAc;IAIxC;;;;OAIG;IACH,YAAY,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI;IAajC;;;;;;OAMG;IACH,YAAY,CAAC,MAAM,EAAE,OAAO,CAAC,oBAAoB,CAAC,GAAG,IAAI;IAQzD;;OAEG;IACH,OAAO,CAAC,WAAW;IAmBnB;;OAEG;IACH,OAAO,CAAC,mBAAmB;IA4B3B;;;;;;;;OAQG;IACH,OAAO,CAAC,mBAAmB;IAqB3B;;OAEG;IACH,OAAO,CAAC,gBAAgB;IAaxB;;OAEG;IACH,OAAO,CAAC,oBAAoB;IAM5B;;OAEG;IACH,OAAO,CAAC,UAAU;IAUlB;;OAEG;IACH,OAAO,IAAI,IAAI;CAUhB"}