agentic-flow 2.0.13 → 2.1.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +87 -0
- package/README.md +376 -186
- package/dist/.tsbuildinfo +1 -1
- package/dist/agent-booster/index.d.ts +14 -0
- package/dist/agent-booster/index.d.ts.map +1 -0
- package/dist/agent-booster/index.js +25 -0
- package/dist/agent-booster/index.js.map +1 -0
- package/dist/agents/directApiAgent.d.ts.map +1 -1
- package/dist/agents/directApiAgent.js +37 -19
- package/dist/agents/directApiAgent.js.map +1 -1
- package/dist/cli/commands/hooks.d.ts.map +1 -1
- package/dist/cli/commands/hooks.js +4 -1
- package/dist/cli/commands/hooks.js.map +1 -1
- package/dist/cli/mcp-manager.js +13 -6
- package/dist/cli/mcp-manager.js.map +1 -1
- package/dist/harness/provenance.d.ts +72 -0
- package/dist/harness/provenance.d.ts.map +1 -0
- package/dist/harness/provenance.js +92 -0
- package/dist/harness/provenance.js.map +1 -0
- package/dist/intelligence/EmbeddingService.js +3 -3
- package/dist/intelligence/EmbeddingService.js.map +1 -1
- package/dist/intelligence/agent-booster-enhanced.d.ts.map +1 -1
- package/dist/intelligence/agent-booster-enhanced.js +5 -3
- package/dist/intelligence/agent-booster-enhanced.js.map +1 -1
- package/dist/mcp/claudeFlowSdkServer.d.ts.map +1 -1
- package/dist/mcp/claudeFlowSdkServer.js +76 -29
- package/dist/mcp/claudeFlowSdkServer.js.map +1 -1
- package/dist/mcp/fastmcp/servers/claude-flow-sdk.js +27 -22
- package/dist/mcp/fastmcp/servers/claude-flow-sdk.js.map +1 -1
- package/dist/mcp/fastmcp/servers/http-sse.js +31 -20
- package/dist/mcp/fastmcp/servers/http-sse.js.map +1 -1
- package/dist/mcp/fastmcp/servers/http-streaming-updated.js +33 -32
- package/dist/mcp/fastmcp/servers/http-streaming-updated.js.map +1 -1
- package/dist/mcp/fastmcp/servers/poc-stdio.js +10 -11
- package/dist/mcp/fastmcp/servers/poc-stdio.js.map +1 -1
- package/dist/mcp/fastmcp/servers/stdio-full.js +33 -32
- package/dist/mcp/fastmcp/servers/stdio-full.js.map +1 -1
- package/dist/mcp/fastmcp/tools/agent/execute.d.ts.map +1 -1
- package/dist/mcp/fastmcp/tools/agent/execute.js +8 -5
- package/dist/mcp/fastmcp/tools/agent/execute.js.map +1 -1
- package/dist/mcp/fastmcp/tools/agent/list.d.ts.map +1 -1
- package/dist/mcp/fastmcp/tools/agent/list.js +4 -2
- package/dist/mcp/fastmcp/tools/agent/list.js.map +1 -1
- package/dist/mcp/fastmcp/tools/agent/parallel.d.ts.map +1 -1
- package/dist/mcp/fastmcp/tools/agent/parallel.js +4 -2
- package/dist/mcp/fastmcp/tools/agent/parallel.js.map +1 -1
- package/dist/mcp/fastmcp/tools/hooks/pretrain.d.ts.map +1 -1
- package/dist/mcp/fastmcp/tools/hooks/pretrain.js +11 -3
- package/dist/mcp/fastmcp/tools/hooks/pretrain.js.map +1 -1
- package/dist/mcp/fastmcp/tools/swarm/orchestrate.d.ts.map +1 -1
- package/dist/mcp/fastmcp/tools/swarm/orchestrate.js +9 -5
- package/dist/mcp/fastmcp/tools/swarm/orchestrate.js.map +1 -1
- package/dist/mcp/standalone-stdio.js +76 -57
- package/dist/mcp/standalone-stdio.js.map +1 -1
- package/dist/mcp/tools/harness-tools.d.ts +28 -0
- package/dist/mcp/tools/harness-tools.d.ts.map +1 -0
- package/dist/mcp/tools/harness-tools.js +90 -0
- package/dist/mcp/tools/harness-tools.js.map +1 -0
- package/dist/optimizations/configuration-tuning.d.ts +1 -1
- package/dist/reasoningbank/utils/embeddings.d.ts +0 -6
- package/dist/reasoningbank/utils/embeddings.d.ts.map +1 -1
- package/dist/reasoningbank/utils/embeddings.js +9 -43
- package/dist/reasoningbank/utils/embeddings.js.map +1 -1
- package/dist/repair/cli.d.ts +24 -0
- package/dist/repair/cli.d.ts.map +1 -0
- package/dist/repair/cli.js +69 -0
- package/dist/repair/cli.js.map +1 -0
- package/dist/repair/darwin-repair.d.ts +80 -0
- package/dist/repair/darwin-repair.d.ts.map +1 -0
- package/dist/repair/darwin-repair.js +81 -0
- package/dist/repair/darwin-repair.js.map +1 -0
- package/dist/router/cost-optimal-router.d.ts +89 -0
- package/dist/router/cost-optimal-router.d.ts.map +1 -0
- package/dist/router/cost-optimal-router.js +94 -0
- package/dist/router/cost-optimal-router.js.map +1 -0
- package/dist/router/providers/onnx-local-optimized.d.ts +1 -10
- package/dist/router/providers/onnx-local-optimized.d.ts.map +1 -1
- package/dist/router/providers/onnx-local-optimized.js +9 -10
- package/dist/router/providers/onnx-local-optimized.js.map +1 -1
- package/dist/router/providers/onnx-local.d.ts +0 -9
- package/dist/router/providers/onnx-local.d.ts.map +1 -1
- package/dist/router/providers/onnx-local.js +7 -22
- package/dist/router/providers/onnx-local.js.map +1 -1
- package/dist/router/router.d.ts +29 -0
- package/dist/router/router.d.ts.map +1 -1
- package/dist/router/router.js +83 -0
- package/dist/router/router.js.map +1 -1
- package/dist/router/types.d.ts +1 -1
- package/dist/router/types.d.ts.map +1 -1
- package/dist/router/types.js.map +1 -1
- package/dist/sdk/plugins.d.ts +6 -6
- package/dist/sdk/plugins.d.ts.map +1 -1
- package/dist/sdk/plugins.js +37 -0
- package/dist/sdk/plugins.js.map +1 -1
- package/dist/transport/index.d.ts +1 -0
- package/dist/transport/index.d.ts.map +1 -1
- package/dist/transport/index.js +1 -0
- package/dist/transport/index.js.map +1 -1
- package/dist/transport/quic-loader.d.ts +250 -0
- package/dist/transport/quic-loader.d.ts.map +1 -0
- package/dist/transport/quic-loader.js +507 -0
- package/dist/transport/quic-loader.js.map +1 -0
- package/dist/utils/agentBoosterPreprocessor.js +16 -22
- package/dist/utils/agentBoosterPreprocessor.js.map +1 -1
- package/package.json +21 -6
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"onnx-local-optimized.js","sourceRoot":"","sources":["../../../src/router/providers/onnx-local-optimized.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAWH,OAAO,EAAE,iBAAiB,EAAmB,MAAM,iBAAiB,CAAC;AAYrE,MAAM,OAAO,qBAAsB,SAAQ,iBAAiB;IAClD,eAAe,CAAgC;IAC/C,WAAW,GAAqB,IAAI,GAAG,EAAE,CAAC;IAC1C,iBAAiB,GAAyD,IAAI,GAAG,EAAE,CAAC;IAE5F,YAAY,SAA8B,EAAE;QAC1C,KAAK,CAAC,MAAM,CAAC,CAAC;QAEd,IAAI,CAAC,eAAe,GAAG;YACrB,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,iFAAiF;YAChH,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,EAAG,sCAAsC;YAC/E,gBAAgB,EAAE,MAAM,CAAC,gBAAgB,IAAI,IAAI,EAAG,sBAAsB;YAC1E,aAAa,EAAE,MAAM,CAAC,aAAa,KAAK,KAAK,EAAG,eAAe;YAC/D,kBAAkB,EAAE,MAAM,CAAC,kBAAkB,KAAK,KAAK,EAAG,eAAe;YACzE,kBAAkB,EAAE,MAAM,CAAC,kBAAkB,KAAK,KAAK,EAAG,eAAe;YACzE,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,EAAE;YACvB,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,GAAG;YACxB,iBAAiB,EAAE,MAAM,CAAC,iBAAiB,IAAI,GAAG;SACnD,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,cAAc,CAAC,IAAY;QACjC,qDAAqD;QACrD,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACpC,CAAC;IAED;;OAEG;IACK,eAAe,CAAC,QAAmB;QACzC,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC,aAAa,EAAE,CAAC;YACxC,OAAO,QAAQ,CAAC;QAClB,CAAC;QAED,MAAM,SAAS,GAAG,IAAI,CAAC,eAAe,CAAC,gBAAgB,CAAC;QACxD,IAAI,WAAW,GAAG,CAAC,CAAC;QACpB,MAAM,SAAS,GAAc,EAAE,CAAC;QAEhC,wCAAwC;QACxC,MAAM,SAAS,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC;QAC1D,IAAI,SAAS,EAAE,CAAC;YACd,MAAM,OAAO,GAAG,OAAO,SAAS,CAAC,OAAO,KAAK,QAAQ;gBACnD,CAAC,CAAC,SAAS,CAAC,OAAO;gBACnB,CAAC,CAAC,SAAS,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;YAEzE,SAAS,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;YAC1B,WAAW,IAAI,IAAI,CAAC,cAAc,CAAC,OAAO,CAAC,CAAC;QAC9C,CAAC;QAED,+CAA+C;QAC/C,KAAK,IAAI,CAAC,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;YAC9C,MAAM,GAAG,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;YAExB,yCAAyC;YACzC,IAAI,GAAG,CAAC,IAAI,KAAK,QAAQ;gBAAE,SAAS;YAEpC,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,MAAM,MAAM,GAAG,IAAI,CAAC,cAAc,CAAC,OAAO,CAAC,CAAC;YAE5C,IAAI,WAAW,GAAG,MAAM,GAAG,SAAS,EAAE,CAAC;gBACrC,OAAO,CAAC,GAAG,CAAC,4BAA4B,QAAQ,CAAC,MAAM,GAAG,SAAS,CAAC,MAAM,eAAe,WAAW,cAAc,CAAC,CAAC;gBACpH,MAAM;YACR,CAAC;YAED,SAAS,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;YACvB,WAAW,IAAI,MAAM,CAAC;QACxB,CAAC;QAED,sCAAsC;QACtC,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,EAAE,CAAC;YACtE,MAAM,WAAW,GAAG,QAAQ,CAAC,KAAK,EAAE,CAAC,OAAO,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC;YAC5E,IAAI,WAAW;gBAAE,SAAS,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QAC/C,CAAC;QAED,OAAO,SAAS,CAAC;IACnB,CAAC;IAED;;OAEG;IACK,cAAc,CAAC,QAAmB;QACxC,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC,kBAAkB,EAAE,CAAC;YAC7C,OAAO,QAAQ,CAAC;QAClB,CAAC;QAED,MAAM,SAAS,GAAG,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE;YACnC,IAAI,GAAG,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBACxB,MAAM,OAAO,GAAG,OAAO,GAAG,CAAC,OAAO,KAAK,QAAQ;oBAC7C,CAAC,CAAC,GAAG,CAAC,OAAO;oBACb,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;gBAEnE,wCAAwC;gBACxC,MAAM,UAAU,GAAG,0DAA0D,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;gBAE5F,IAAI,UAAU,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC;oBAC5E,MAAM,eAAe,GAAG,GAAG,OAAO,kHAAkH,CAAC;oBAErJ,OAAO;wBACL,GAAG,GAAG;wBACN,OAAO,EAAE,eAAe;qBACzB,CAAC;gBACJ,CAAC;YACH,CAAC;YAED,OAAO,GAAG,CAAC;QACb,CAAC,CAAC,CAAC;QAEH,OAAO,SAAS,CAAC;IACnB,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,IAAI,CAAC,MAAkB;QAC3B,4CAA4C;QAC5C,IAAI,QAAQ,GAAG,IAAI,CAAC,eAAe,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QAErD,uCAAuC;QACvC,QAAQ,GAAG,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,CAAC;QAEzC,2DAA2D;QAC3D,MAAM,cAAc,GAAG;YACrB,GAAG,MAAM;YACT,QAAQ;YACR,WAAW,EAAE,MAAM,CAAC,WAAW,IAAI,IAAI,CAAC,eAAe,CAAC,WAAW;YACnE,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,IAAI,CAAC,eAAe,CAAC,SAAS;SAC9D,CAAC;QAEF,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;QAElD,4BAA4B;QAC5B,IAAI,QAAQ,CAAC,QAAQ,EAAE,CAAC;YACtB,QAAQ,CAAC,QAAQ,CAAC,aAAa,GAAG;gBAChC,cAAc,EAAE,IAAI,CAAC,eAAe,CAAC,aAAa;gBAClD,kBAAkB,EAAE,IAAI,CAAC,eAAe,CAAC,kBAAkB;gBAC3D,mBAAmB,EAAE,IAAI,CAAC,eAAe,CAAC,kBAAkB;gBAC5D,oBAAoB,EAAE,MAAM,CAAC,QAAQ,CAAC,MAAM;gBAC5C,qBAAqB,EAAE,QAAQ,CAAC,MAAM;aACvC,CAAC;QACJ,CAAC;QAED,OAAO,QAAQ,CAAC;IAClB,CAAC;IAED;;OAEG;IACH,mBAAmB;QACjB,OAAO;YACL,GAAG,KAAK,CAAC,YAAY,EAAE;YACvB,aAAa,EAAE;gBACb,gBAAgB,EAAE,IAAI,CAAC,eAAe,CAAC,gBAAgB;gBACvD,aAAa,EAAE,IAAI,CAAC,eAAe,CAAC,aAAa;gBACjD,kBAAkB,EAAE,IAAI,CAAC,eAAe,CAAC,kBAAkB;gBAC3D,kBAAkB,EAAE,IAAI,CAAC,eAAe,CAAC,kBAAkB;gBAC3D,WAAW,EAAE,IAAI,CAAC,eAAe,CAAC,WAAW;gBAC7C,IAAI,EAAE,IAAI,CAAC,eAAe,CAAC,IAAI;gBAC/B,IAAI,EAAE,IAAI,CAAC,eAAe,CAAC,IAAI;gBAC/B,iBAAiB,EAAE,IAAI,CAAC,eAAe,CAAC,iBAAiB;aAC1D;YACD,UAAU,EAAE;gBACV,eAAe,EAAE,IAAI,CAAC,WAAW,CAAC,IAAI;gBACtC,qBAAqB,EAAE,IAAI,CAAC,iBAAiB,CAAC,IAAI;aACnD;SACF,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,WAAW;QACT,IAAI,CAAC,WAAW,CAAC,KAAK,EAAE,CAAC;QACzB,IAAI,CAAC,iBAAiB,CAAC,KAAK,EAAE,CAAC;QAC/B,OAAO,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC;IACnC,CAAC;CACF","sourcesContent":["/**\n * Optimized ONNX Runtime Local Inference Provider\n *\n * Improvements over base implementation:\n * - Context pruning for 2-4x speed improvement\n * - Prompt optimization for 30-50% quality improvement\n * - KV cache pooling for 20-30% faster generation\n * - Better generation parameters for code tasks\n * - System prompt caching\n *\n * Note: onnxruntime-node is optional - will error if not installed.\n *\n * NOTE (ruvnet/ruflo#2048): the previous top-level `await import('onnxruntime-node')`\n * fired the native-binding load (`onnxruntime_binding.node`) at module\n * import time. On Windows this crashes with \"OS cannot run %1\" — and the\n * crash propagated to any consumer that transitively imports this file\n * (e.g. `agentic-flow/reasoningbank` via `core/distill → router/router`).\n * This file does not use `ort` directly — the base `ONNXLocalProvider`\n * it extends does, and that file now lazy-loads ort on first session\n * init. So we just drop the eager top-level load here.\n */\n\nimport { get_encoding } from 'tiktoken';\nimport { ensurePhi4Model, ModelDownloader } from '../../utils/model-downloader.js';\nimport type {\n ChatParams,\n ChatResponse,\n Message,\n ContentBlock,\n ProviderError\n} from '../types.js';\nimport { ONNXLocalProvider, ONNXLocalConfig } from './onnx-local.js';\n\nexport interface OptimizedONNXConfig extends ONNXLocalConfig {\n maxContextTokens?: number;\n slidingWindow?: boolean;\n cacheSystemPrompts?: boolean;\n promptOptimization?: boolean;\n topK?: number;\n topP?: number;\n repetitionPenalty?: number;\n}\n\nexport class OptimizedONNXProvider extends ONNXLocalProvider {\n private optimizedConfig: Required<OptimizedONNXConfig>;\n private kvCachePool: Map<string, any> = new Map();\n private systemPromptCache: Map<string, { tokens: number[]; timestamp: number }> = new Map();\n\n constructor(config: OptimizedONNXConfig = {}) {\n super(config);\n\n this.optimizedConfig = {\n modelPath: config.modelPath || './models/phi-4-mini/cpu_and_mobile/cpu-int4-rtn-block-32-acc-level-4/model.onnx',\n executionProviders: config.executionProviders || ['cpu'],\n maxTokens: config.maxTokens || 200,\n temperature: config.temperature || 0.3, // Lower for code (more deterministic)\n maxContextTokens: config.maxContextTokens || 2048, // Keep under 4K limit\n slidingWindow: config.slidingWindow !== false, // Default true\n cacheSystemPrompts: config.cacheSystemPrompts !== false, // Default true\n promptOptimization: config.promptOptimization !== false, // Default true\n topK: config.topK || 50,\n topP: config.topP || 0.9,\n repetitionPenalty: config.repetitionPenalty || 1.1\n };\n }\n\n /**\n * Estimate token count for a string\n */\n private estimateTokens(text: string): number {\n // Rough estimate: 1 token ≈ 4 characters for English\n return Math.ceil(text.length / 4);\n }\n\n /**\n * Optimize messages using sliding window context pruning\n */\n private optimizeContext(messages: Message[]): Message[] {\n if (!this.optimizedConfig.slidingWindow) {\n return messages;\n }\n\n const maxTokens = this.optimizedConfig.maxContextTokens;\n let totalTokens = 0;\n const optimized: Message[] = [];\n\n // Always keep system message if present\n const systemMsg = messages.find(m => m.role === 'system');\n if (systemMsg) {\n const content = typeof systemMsg.content === 'string'\n ? systemMsg.content\n : systemMsg.content.map(c => c.type === 'text' ? c.text : '').join('');\n\n optimized.push(systemMsg);\n totalTokens += this.estimateTokens(content);\n }\n\n // Add recent messages from end (most relevant)\n for (let i = messages.length - 1; i >= 0; i--) {\n const msg = messages[i];\n\n // Skip if already added (system message)\n if (msg.role === 'system') continue;\n\n const content = typeof msg.content === 'string'\n ? msg.content\n : msg.content.map(c => c.type === 'text' ? c.text : '').join('');\n\n const tokens = this.estimateTokens(content);\n\n if (totalTokens + tokens > maxTokens) {\n console.log(`📊 Context pruned: Saved ${messages.length - optimized.length} messages, ~${totalTokens} tokens kept`);\n break;\n }\n\n optimized.unshift(msg);\n totalTokens += tokens;\n }\n\n // Ensure at least user message exists\n if (optimized.length === 0 || !optimized.some(m => m.role === 'user')) {\n const lastUserMsg = messages.slice().reverse().find(m => m.role === 'user');\n if (lastUserMsg) optimized.push(lastUserMsg);\n }\n\n return optimized;\n }\n\n /**\n * Optimize prompt for better quality output\n */\n private optimizePrompt(messages: Message[]): Message[] {\n if (!this.optimizedConfig.promptOptimization) {\n return messages;\n }\n\n const optimized = messages.map(msg => {\n if (msg.role === 'user') {\n const content = typeof msg.content === 'string'\n ? msg.content\n : msg.content.map(c => c.type === 'text' ? c.text : '').join('');\n\n // Add quality indicators for code tasks\n const isCodeTask = /write|create|implement|generate|code|function|class|api/i.test(content);\n\n if (isCodeTask && !content.includes('include') && !content.includes('with')) {\n const enhancedContent = `${content}. Include: proper error handling, type hints/types, and edge case handling. Return clean, production-ready code.`;\n\n return {\n ...msg,\n content: enhancedContent\n };\n }\n }\n\n return msg;\n });\n\n return optimized;\n }\n\n /**\n * Enhanced chat with optimization\n */\n async chat(params: ChatParams): Promise<ChatResponse> {\n // Step 1: Optimize context (sliding window)\n let messages = this.optimizeContext(params.messages);\n\n // Step 2: Optimize prompts for quality\n messages = this.optimizePrompt(messages);\n\n // Step 3: Call base implementation with optimized messages\n const enhancedParams = {\n ...params,\n messages,\n temperature: params.temperature || this.optimizedConfig.temperature,\n maxTokens: params.maxTokens || this.optimizedConfig.maxTokens\n };\n\n const response = await super.chat(enhancedParams);\n\n // Add optimization metadata\n if (response.metadata) {\n response.metadata.optimizations = {\n contextPruning: this.optimizedConfig.slidingWindow,\n promptOptimization: this.optimizedConfig.promptOptimization,\n systemPromptCaching: this.optimizedConfig.cacheSystemPrompts,\n originalMessageCount: params.messages.length,\n optimizedMessageCount: messages.length\n };\n }\n\n return response;\n }\n\n /**\n * Get optimization info\n */\n getOptimizationInfo() {\n return {\n ...super.getModelInfo(),\n optimizations: {\n maxContextTokens: this.optimizedConfig.maxContextTokens,\n slidingWindow: this.optimizedConfig.slidingWindow,\n cacheSystemPrompts: this.optimizedConfig.cacheSystemPrompts,\n promptOptimization: this.optimizedConfig.promptOptimization,\n temperature: this.optimizedConfig.temperature,\n topK: this.optimizedConfig.topK,\n topP: this.optimizedConfig.topP,\n repetitionPenalty: this.optimizedConfig.repetitionPenalty\n },\n cacheStats: {\n kvCachePoolSize: this.kvCachePool.size,\n systemPromptCacheSize: this.systemPromptCache.size\n }\n };\n }\n\n /**\n * Clear caches\n */\n clearCaches() {\n this.kvCachePool.clear();\n this.systemPromptCache.clear();\n console.log('🧹 Caches cleared');\n }\n}\n"]}
|
|
1
|
+
{"version":3,"file":"onnx-local-optimized.js","sourceRoot":"","sources":["../../../src/router/providers/onnx-local-optimized.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,IAAI,GAAG,GAAQ,IAAI,CAAC;AAEpB,+CAA+C;AAC/C,IAAI,CAAC;IACH,GAAG,GAAG,MAAM,MAAM,CAAC,kBAAkB,CAAC,CAAC;AACzC,CAAC;AAAC,MAAM,CAAC;IACP,6BAA6B;AAC/B,CAAC;AAWD,OAAO,EAAE,iBAAiB,EAAmB,MAAM,iBAAiB,CAAC;AAYrE,MAAM,OAAO,qBAAsB,SAAQ,iBAAiB;IAClD,eAAe,CAAgC;IAC/C,WAAW,GAAqB,IAAI,GAAG,EAAE,CAAC;IAC1C,iBAAiB,GAAyD,IAAI,GAAG,EAAE,CAAC;IAE5F,YAAY,SAA8B,EAAE;QAC1C,KAAK,CAAC,MAAM,CAAC,CAAC;QAEd,IAAI,CAAC,eAAe,GAAG;YACrB,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,iFAAiF;YAChH,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,EAAG,sCAAsC;YAC/E,gBAAgB,EAAE,MAAM,CAAC,gBAAgB,IAAI,IAAI,EAAG,sBAAsB;YAC1E,aAAa,EAAE,MAAM,CAAC,aAAa,KAAK,KAAK,EAAG,eAAe;YAC/D,kBAAkB,EAAE,MAAM,CAAC,kBAAkB,KAAK,KAAK,EAAG,eAAe;YACzE,kBAAkB,EAAE,MAAM,CAAC,kBAAkB,KAAK,KAAK,EAAG,eAAe;YACzE,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,EAAE;YACvB,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,GAAG;YACxB,iBAAiB,EAAE,MAAM,CAAC,iBAAiB,IAAI,GAAG;SACnD,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,cAAc,CAAC,IAAY;QACjC,qDAAqD;QACrD,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACpC,CAAC;IAED;;OAEG;IACK,eAAe,CAAC,QAAmB;QACzC,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC,aAAa,EAAE,CAAC;YACxC,OAAO,QAAQ,CAAC;QAClB,CAAC;QAED,MAAM,SAAS,GAAG,IAAI,CAAC,eAAe,CAAC,gBAAgB,CAAC;QACxD,IAAI,WAAW,GAAG,CAAC,CAAC;QACpB,MAAM,SAAS,GAAc,EAAE,CAAC;QAEhC,wCAAwC;QACxC,MAAM,SAAS,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC;QAC1D,IAAI,SAAS,EAAE,CAAC;YACd,MAAM,OAAO,GAAG,OAAO,SAAS,CAAC,OAAO,KAAK,QAAQ;gBACnD,CAAC,CAAC,SAAS,CAAC,OAAO;gBACnB,CAAC,CAAC,SAAS,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;YAEzE,SAAS,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;YAC1B,WAAW,IAAI,IAAI,CAAC,cAAc,CAAC,OAAO,CAAC,CAAC;QAC9C,CAAC;QAED,+CAA+C;QAC/C,KAAK,IAAI,CAAC,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;YAC9C,MAAM,GAAG,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;YAExB,yCAAyC;YACzC,IAAI,GAAG,CAAC,IAAI,KAAK,QAAQ;gBAAE,SAAS;YAEpC,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,MAAM,MAAM,GAAG,IAAI,CAAC,cAAc,CAAC,OAAO,CAAC,CAAC;YAE5C,IAAI,WAAW,GAAG,MAAM,GAAG,SAAS,EAAE,CAAC;gBACrC,OAAO,CAAC,GAAG,CAAC,4BAA4B,QAAQ,CAAC,MAAM,GAAG,SAAS,CAAC,MAAM,eAAe,WAAW,cAAc,CAAC,CAAC;gBACpH,MAAM;YACR,CAAC;YAED,SAAS,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;YACvB,WAAW,IAAI,MAAM,CAAC;QACxB,CAAC;QAED,sCAAsC;QACtC,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,EAAE,CAAC;YACtE,MAAM,WAAW,GAAG,QAAQ,CAAC,KAAK,EAAE,CAAC,OAAO,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC;YAC5E,IAAI,WAAW;gBAAE,SAAS,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QAC/C,CAAC;QAED,OAAO,SAAS,CAAC;IACnB,CAAC;IAED;;OAEG;IACK,cAAc,CAAC,QAAmB;QACxC,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC,kBAAkB,EAAE,CAAC;YAC7C,OAAO,QAAQ,CAAC;QAClB,CAAC;QAED,MAAM,SAAS,GAAG,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE;YACnC,IAAI,GAAG,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBACxB,MAAM,OAAO,GAAG,OAAO,GAAG,CAAC,OAAO,KAAK,QAAQ;oBAC7C,CAAC,CAAC,GAAG,CAAC,OAAO;oBACb,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;gBAEnE,wCAAwC;gBACxC,MAAM,UAAU,GAAG,0DAA0D,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;gBAE5F,IAAI,UAAU,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC;oBAC5E,MAAM,eAAe,GAAG,GAAG,OAAO,kHAAkH,CAAC;oBAErJ,OAAO;wBACL,GAAG,GAAG;wBACN,OAAO,EAAE,eAAe;qBACzB,CAAC;gBACJ,CAAC;YACH,CAAC;YAED,OAAO,GAAG,CAAC;QACb,CAAC,CAAC,CAAC;QAEH,OAAO,SAAS,CAAC;IACnB,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,IAAI,CAAC,MAAkB;QAC3B,4CAA4C;QAC5C,IAAI,QAAQ,GAAG,IAAI,CAAC,eAAe,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QAErD,uCAAuC;QACvC,QAAQ,GAAG,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,CAAC;QAEzC,2DAA2D;QAC3D,MAAM,cAAc,GAAG;YACrB,GAAG,MAAM;YACT,QAAQ;YACR,WAAW,EAAE,MAAM,CAAC,WAAW,IAAI,IAAI,CAAC,eAAe,CAAC,WAAW;YACnE,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,IAAI,CAAC,eAAe,CAAC,SAAS;SAC9D,CAAC;QAEF,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;QAElD,4BAA4B;QAC5B,IAAI,QAAQ,CAAC,QAAQ,EAAE,CAAC;YACtB,QAAQ,CAAC,QAAQ,CAAC,aAAa,GAAG;gBAChC,cAAc,EAAE,IAAI,CAAC,eAAe,CAAC,aAAa;gBAClD,kBAAkB,EAAE,IAAI,CAAC,eAAe,CAAC,kBAAkB;gBAC3D,mBAAmB,EAAE,IAAI,CAAC,eAAe,CAAC,kBAAkB;gBAC5D,oBAAoB,EAAE,MAAM,CAAC,QAAQ,CAAC,MAAM;gBAC5C,qBAAqB,EAAE,QAAQ,CAAC,MAAM;aACvC,CAAC;QACJ,CAAC;QAED,OAAO,QAAQ,CAAC;IAClB,CAAC;IAED;;OAEG;IACH,mBAAmB;QACjB,OAAO;YACL,GAAG,KAAK,CAAC,YAAY,EAAE;YACvB,aAAa,EAAE;gBACb,gBAAgB,EAAE,IAAI,CAAC,eAAe,CAAC,gBAAgB;gBACvD,aAAa,EAAE,IAAI,CAAC,eAAe,CAAC,aAAa;gBACjD,kBAAkB,EAAE,IAAI,CAAC,eAAe,CAAC,kBAAkB;gBAC3D,kBAAkB,EAAE,IAAI,CAAC,eAAe,CAAC,kBAAkB;gBAC3D,WAAW,EAAE,IAAI,CAAC,eAAe,CAAC,WAAW;gBAC7C,IAAI,EAAE,IAAI,CAAC,eAAe,CAAC,IAAI;gBAC/B,IAAI,EAAE,IAAI,CAAC,eAAe,CAAC,IAAI;gBAC/B,iBAAiB,EAAE,IAAI,CAAC,eAAe,CAAC,iBAAiB;aAC1D;YACD,UAAU,EAAE;gBACV,eAAe,EAAE,IAAI,CAAC,WAAW,CAAC,IAAI;gBACtC,qBAAqB,EAAE,IAAI,CAAC,iBAAiB,CAAC,IAAI;aACnD;SACF,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,WAAW;QACT,IAAI,CAAC,WAAW,CAAC,KAAK,EAAE,CAAC;QACzB,IAAI,CAAC,iBAAiB,CAAC,KAAK,EAAE,CAAC;QAC/B,OAAO,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC;IACnC,CAAC;CACF","sourcesContent":["/**\n * Optimized ONNX Runtime Local Inference Provider\n *\n * Improvements over base implementation:\n * - Context pruning for 2-4x speed improvement\n * - Prompt optimization for 30-50% quality improvement\n * - KV cache pooling for 20-30% faster generation\n * - Better generation parameters for code tasks\n * - System prompt caching\n *\n * Note: onnxruntime-node is optional - will error if not installed\n */\n\nlet ort: any = null;\n\n// Dynamic import for optional onnxruntime-node\ntry {\n ort = await import('onnxruntime-node');\n} catch {\n // Will be handled at runtime\n}\n\nimport { get_encoding } from 'tiktoken';\nimport { ensurePhi4Model, ModelDownloader } from '../../utils/model-downloader.js';\nimport type {\n ChatParams,\n ChatResponse,\n Message,\n ContentBlock,\n ProviderError\n} from '../types.js';\nimport { ONNXLocalProvider, ONNXLocalConfig } from './onnx-local.js';\n\nexport interface OptimizedONNXConfig extends ONNXLocalConfig {\n maxContextTokens?: number;\n slidingWindow?: boolean;\n cacheSystemPrompts?: boolean;\n promptOptimization?: boolean;\n topK?: number;\n topP?: number;\n repetitionPenalty?: number;\n}\n\nexport class OptimizedONNXProvider extends ONNXLocalProvider {\n private optimizedConfig: Required<OptimizedONNXConfig>;\n private kvCachePool: Map<string, any> = new Map();\n private systemPromptCache: Map<string, { tokens: number[]; timestamp: number }> = new Map();\n\n constructor(config: OptimizedONNXConfig = {}) {\n super(config);\n\n this.optimizedConfig = {\n modelPath: config.modelPath || './models/phi-4-mini/cpu_and_mobile/cpu-int4-rtn-block-32-acc-level-4/model.onnx',\n executionProviders: config.executionProviders || ['cpu'],\n maxTokens: config.maxTokens || 200,\n temperature: config.temperature || 0.3, // Lower for code (more deterministic)\n maxContextTokens: config.maxContextTokens || 2048, // Keep under 4K limit\n slidingWindow: config.slidingWindow !== false, // Default true\n cacheSystemPrompts: config.cacheSystemPrompts !== false, // Default true\n promptOptimization: config.promptOptimization !== false, // Default true\n topK: config.topK || 50,\n topP: config.topP || 0.9,\n repetitionPenalty: config.repetitionPenalty || 1.1\n };\n }\n\n /**\n * Estimate token count for a string\n */\n private estimateTokens(text: string): number {\n // Rough estimate: 1 token ≈ 4 characters for English\n return Math.ceil(text.length / 4);\n }\n\n /**\n * Optimize messages using sliding window context pruning\n */\n private optimizeContext(messages: Message[]): Message[] {\n if (!this.optimizedConfig.slidingWindow) {\n return messages;\n }\n\n const maxTokens = this.optimizedConfig.maxContextTokens;\n let totalTokens = 0;\n const optimized: Message[] = [];\n\n // Always keep system message if present\n const systemMsg = messages.find(m => m.role === 'system');\n if (systemMsg) {\n const content = typeof systemMsg.content === 'string'\n ? systemMsg.content\n : systemMsg.content.map(c => c.type === 'text' ? c.text : '').join('');\n\n optimized.push(systemMsg);\n totalTokens += this.estimateTokens(content);\n }\n\n // Add recent messages from end (most relevant)\n for (let i = messages.length - 1; i >= 0; i--) {\n const msg = messages[i];\n\n // Skip if already added (system message)\n if (msg.role === 'system') continue;\n\n const content = typeof msg.content === 'string'\n ? msg.content\n : msg.content.map(c => c.type === 'text' ? c.text : '').join('');\n\n const tokens = this.estimateTokens(content);\n\n if (totalTokens + tokens > maxTokens) {\n console.log(`📊 Context pruned: Saved ${messages.length - optimized.length} messages, ~${totalTokens} tokens kept`);\n break;\n }\n\n optimized.unshift(msg);\n totalTokens += tokens;\n }\n\n // Ensure at least user message exists\n if (optimized.length === 0 || !optimized.some(m => m.role === 'user')) {\n const lastUserMsg = messages.slice().reverse().find(m => m.role === 'user');\n if (lastUserMsg) optimized.push(lastUserMsg);\n }\n\n return optimized;\n }\n\n /**\n * Optimize prompt for better quality output\n */\n private optimizePrompt(messages: Message[]): Message[] {\n if (!this.optimizedConfig.promptOptimization) {\n return messages;\n }\n\n const optimized = messages.map(msg => {\n if (msg.role === 'user') {\n const content = typeof msg.content === 'string'\n ? msg.content\n : msg.content.map(c => c.type === 'text' ? c.text : '').join('');\n\n // Add quality indicators for code tasks\n const isCodeTask = /write|create|implement|generate|code|function|class|api/i.test(content);\n\n if (isCodeTask && !content.includes('include') && !content.includes('with')) {\n const enhancedContent = `${content}. Include: proper error handling, type hints/types, and edge case handling. Return clean, production-ready code.`;\n\n return {\n ...msg,\n content: enhancedContent\n };\n }\n }\n\n return msg;\n });\n\n return optimized;\n }\n\n /**\n * Enhanced chat with optimization\n */\n async chat(params: ChatParams): Promise<ChatResponse> {\n // Step 1: Optimize context (sliding window)\n let messages = this.optimizeContext(params.messages);\n\n // Step 2: Optimize prompts for quality\n messages = this.optimizePrompt(messages);\n\n // Step 3: Call base implementation with optimized messages\n const enhancedParams = {\n ...params,\n messages,\n temperature: params.temperature || this.optimizedConfig.temperature,\n maxTokens: params.maxTokens || this.optimizedConfig.maxTokens\n };\n\n const response = await super.chat(enhancedParams);\n\n // Add optimization metadata\n if (response.metadata) {\n response.metadata.optimizations = {\n contextPruning: this.optimizedConfig.slidingWindow,\n promptOptimization: this.optimizedConfig.promptOptimization,\n systemPromptCaching: this.optimizedConfig.cacheSystemPrompts,\n originalMessageCount: params.messages.length,\n optimizedMessageCount: messages.length\n };\n }\n\n return response;\n }\n\n /**\n * Get optimization info\n */\n getOptimizationInfo() {\n return {\n ...super.getModelInfo(),\n optimizations: {\n maxContextTokens: this.optimizedConfig.maxContextTokens,\n slidingWindow: this.optimizedConfig.slidingWindow,\n cacheSystemPrompts: this.optimizedConfig.cacheSystemPrompts,\n promptOptimization: this.optimizedConfig.promptOptimization,\n temperature: this.optimizedConfig.temperature,\n topK: this.optimizedConfig.topK,\n topP: this.optimizedConfig.topP,\n repetitionPenalty: this.optimizedConfig.repetitionPenalty\n },\n cacheStats: {\n kvCachePoolSize: this.kvCachePool.size,\n systemPromptCacheSize: this.systemPromptCache.size\n }\n };\n }\n\n /**\n * Clear caches\n */\n clearCaches() {\n this.kvCachePool.clear();\n this.systemPromptCache.clear();\n console.log('🧹 Caches cleared');\n }\n}\n"]}
|
|
@@ -3,15 +3,6 @@
|
|
|
3
3
|
*
|
|
4
4
|
* Uses onnxruntime-node for true local CPU/GPU inference
|
|
5
5
|
* Falls back gracefully when native module isn't available (Windows)
|
|
6
|
-
*
|
|
7
|
-
* NOTE (ruvnet/ruflo#2048): `onnxruntime-node` is loaded LAZILY on first
|
|
8
|
-
* `initializeSession()` call, not at module import. The previous top-level
|
|
9
|
-
* `await import('onnxruntime-node')` fired the native-binding load
|
|
10
|
-
* (`onnxruntime_binding.node`) at module load time, which crashed Windows
|
|
11
|
-
* environments where the NAPI binary cannot be loaded — even when the
|
|
12
|
-
* consumer (e.g. `agentic-flow/reasoningbank`) never actually invokes
|
|
13
|
-
* the router. Moving the import inside `loadOrt()` keeps importing
|
|
14
|
-
* `reasoningbank` side-effect-free with respect to native bindings.
|
|
15
6
|
*/
|
|
16
7
|
import type { LLMProvider, ChatParams, ChatResponse, StreamChunk } from '../types.js';
|
|
17
8
|
export interface ONNXLocalConfig {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"onnx-local.d.ts","sourceRoot":"","sources":["../../../src/router/providers/onnx-local.ts"],"names":[],"mappings":"AAAA
|
|
1
|
+
{"version":3,"file":"onnx-local.d.ts","sourceRoot":"","sources":["../../../src/router/providers/onnx-local.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAiBH,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,CAAa;IAC5B,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;IAiD/B;;OAEG;IACH,OAAO,CAAC,cAAc;IAqBtB;;;OAGG;IACH,OAAO,CAAC,iBAAiB;IA8BzB;;OAEG;IACG,IAAI,CAAC,MAAM,EAAE,UAAU,GAAG,OAAO,CAAC,YAAY,CAAC;IA8IrD;;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"}
|
|
@@ -3,30 +3,16 @@
|
|
|
3
3
|
*
|
|
4
4
|
* Uses onnxruntime-node for true local CPU/GPU inference
|
|
5
5
|
* Falls back gracefully when native module isn't available (Windows)
|
|
6
|
-
*
|
|
7
|
-
* NOTE (ruvnet/ruflo#2048): `onnxruntime-node` is loaded LAZILY on first
|
|
8
|
-
* `initializeSession()` call, not at module import. The previous top-level
|
|
9
|
-
* `await import('onnxruntime-node')` fired the native-binding load
|
|
10
|
-
* (`onnxruntime_binding.node`) at module load time, which crashed Windows
|
|
11
|
-
* environments where the NAPI binary cannot be loaded — even when the
|
|
12
|
-
* consumer (e.g. `agentic-flow/reasoningbank`) never actually invokes
|
|
13
|
-
* the router. Moving the import inside `loadOrt()` keeps importing
|
|
14
|
-
* `reasoningbank` side-effect-free with respect to native bindings.
|
|
15
6
|
*/
|
|
16
7
|
let ort = null;
|
|
17
8
|
let ortAvailable = false;
|
|
18
|
-
|
|
19
|
-
|
|
20
|
-
|
|
21
|
-
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
ortAvailable = true;
|
|
26
|
-
}
|
|
27
|
-
catch {
|
|
28
|
-
console.warn('[ONNX] onnxruntime-node not available - local inference disabled');
|
|
29
|
-
}
|
|
9
|
+
// Dynamic import for optional onnxruntime-node
|
|
10
|
+
try {
|
|
11
|
+
ort = await import('onnxruntime-node');
|
|
12
|
+
ortAvailable = true;
|
|
13
|
+
}
|
|
14
|
+
catch {
|
|
15
|
+
console.warn('[ONNX] onnxruntime-node not available - local inference disabled');
|
|
30
16
|
}
|
|
31
17
|
import { get_encoding } from 'tiktoken';
|
|
32
18
|
import { ensurePhi4Model, ModelDownloader } from '../../utils/model-downloader.js';
|
|
@@ -96,7 +82,6 @@ export class ONNXLocalProvider {
|
|
|
96
82
|
async initializeSession() {
|
|
97
83
|
if (this.session)
|
|
98
84
|
return;
|
|
99
|
-
await loadOrt();
|
|
100
85
|
if (!ortAvailable || !ort) {
|
|
101
86
|
throw new Error('onnxruntime-node not available - install with: npm install onnxruntime-node');
|
|
102
87
|
}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"onnx-local.js","sourceRoot":"","sources":["../../../src/router/providers/onnx-local.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,IAAI,GAAG,GAAQ,IAAI,CAAC;AACpB,IAAI,YAAY,GAAG,KAAK,CAAC;AACzB,IAAI,SAAS,GAAG,KAAK,CAAC;AAEtB,KAAK,UAAU,OAAO;IACpB,IAAI,SAAS;QAAE,OAAO;IACtB,SAAS,GAAG,IAAI,CAAC;IACjB,IAAI,CAAC;QACH,GAAG,GAAG,MAAM,MAAM,CAAC,kBAAkB,CAAC,CAAC;QACvC,YAAY,GAAG,IAAI,CAAC;IACtB,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,CAAC,IAAI,CAAC,kEAAkE,CAAC,CAAC;IACnF,CAAC;AACH,CAAC;AAID,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,GAAQ,IAAI,CAAC;IACpB,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,MAAM,OAAO,EAAE,CAAC;QAChB,IAAI,CAAC,YAAY,IAAI,CAAC,GAAG,EAAE,CAAC;YAC1B,MAAM,IAAI,KAAK,CAAC,6EAA6E,CAAC,CAAC;QACjG,CAAC;QAED,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,GAAwB,EAAE,CAAC;QAExC,uDAAuD;QACvD,MAAM,WAAW,GAAI,GAAW,CAAC,MAAM,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,OAAO,CAAC,mBAAmB,CAAC,MAAM,CAAC,GAAG,IAAI,WAAW,CACnD,SAAS,EACT,UAAU,EACV,CAAC,SAAS,EAAE,UAAU,EAAE,CAAC,EAAE,OAAO,CAAC,CACpC,CAAC;YAEF,OAAO,CAAC,mBAAmB,CAAC,QAAQ,CAAC,GAAG,IAAI,WAAW,CACrD,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,uDAAuD;gBACvD,MAAM,WAAW,GAAI,GAAW,CAAC,MAAM,CAAC;gBAExC,uCAAuC;gBACvC,MAAM,WAAW,GAAG,IAAI,WAAW,CACjC,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,WAAW,CACnC,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 * Falls back gracefully when native module isn't available (Windows)\n *\n * NOTE (ruvnet/ruflo#2048): `onnxruntime-node` is loaded LAZILY on first\n * `initializeSession()` call, not at module import. The previous top-level\n * `await import('onnxruntime-node')` fired the native-binding load\n * (`onnxruntime_binding.node`) at module load time, which crashed Windows\n * environments where the NAPI binary cannot be loaded — even when the\n * consumer (e.g. `agentic-flow/reasoningbank`) never actually invokes\n * the router. Moving the import inside `loadOrt()` keeps importing\n * `reasoningbank` side-effect-free with respect to native bindings.\n */\n\nlet ort: any = null;\nlet ortAvailable = false;\nlet ortLoaded = false;\n\nasync function loadOrt(): Promise<void> {\n if (ortLoaded) return;\n ortLoaded = true;\n try {\n ort = await import('onnxruntime-node');\n ortAvailable = true;\n } catch {\n console.warn('[ONNX] onnxruntime-node not available - local inference disabled');\n }\n}\n\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: any = 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 await loadOrt();\n if (!ortAvailable || !ort) {\n throw new Error('onnxruntime-node not available - install with: npm install onnxruntime-node');\n }\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, any> = {};\n\n // Get Tensor constructor - use any for flexible access\n const TensorClass = (ort as any).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.${i}.key`] = new TensorClass(\n 'float32',\n emptyCache,\n [batchSize, numKVHeads, 0, headDim]\n );\n\n kvCache[`past_key_values.${i}.value`] = new TensorClass(\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 // Get Tensor constructor - use any for flexible access\n const TensorClass = (ort as any).Tensor;\n\n // Create input tensor for current step\n const inputTensor = new TensorClass(\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 TensorClass(\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"]}
|
|
1
|
+
{"version":3,"file":"onnx-local.js","sourceRoot":"","sources":["../../../src/router/providers/onnx-local.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,IAAI,GAAG,GAAQ,IAAI,CAAC;AACpB,IAAI,YAAY,GAAG,KAAK,CAAC;AAEzB,+CAA+C;AAC/C,IAAI,CAAC;IACH,GAAG,GAAG,MAAM,MAAM,CAAC,kBAAkB,CAAC,CAAC;IACvC,YAAY,GAAG,IAAI,CAAC;AACtB,CAAC;AAAC,MAAM,CAAC;IACP,OAAO,CAAC,IAAI,CAAC,kEAAkE,CAAC,CAAC;AACnF,CAAC;AAID,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,GAAQ,IAAI,CAAC;IACpB,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,YAAY,IAAI,CAAC,GAAG,EAAE,CAAC;YAC1B,MAAM,IAAI,KAAK,CAAC,6EAA6E,CAAC,CAAC;QACjG,CAAC;QAED,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,GAAwB,EAAE,CAAC;QAExC,uDAAuD;QACvD,MAAM,WAAW,GAAI,GAAW,CAAC,MAAM,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,OAAO,CAAC,mBAAmB,CAAC,MAAM,CAAC,GAAG,IAAI,WAAW,CACnD,SAAS,EACT,UAAU,EACV,CAAC,SAAS,EAAE,UAAU,EAAE,CAAC,EAAE,OAAO,CAAC,CACpC,CAAC;YAEF,OAAO,CAAC,mBAAmB,CAAC,QAAQ,CAAC,GAAG,IAAI,WAAW,CACrD,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,uDAAuD;gBACvD,MAAM,WAAW,GAAI,GAAW,CAAC,MAAM,CAAC;gBAExC,uCAAuC;gBACvC,MAAM,WAAW,GAAG,IAAI,WAAW,CACjC,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,WAAW,CACnC,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 * Falls back gracefully when native module isn't available (Windows)\n */\n\nlet ort: any = null;\nlet ortAvailable = false;\n\n// Dynamic import for optional onnxruntime-node\ntry {\n ort = await import('onnxruntime-node');\n ortAvailable = true;\n} catch {\n console.warn('[ONNX] onnxruntime-node not available - local inference disabled');\n}\n\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: any = 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 if (!ortAvailable || !ort) {\n throw new Error('onnxruntime-node not available - install with: npm install onnxruntime-node');\n }\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, any> = {};\n\n // Get Tensor constructor - use any for flexible access\n const TensorClass = (ort as any).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.${i}.key`] = new TensorClass(\n 'float32',\n emptyCache,\n [batchSize, numKVHeads, 0, headDim]\n );\n\n kvCache[`past_key_values.${i}.value`] = new TensorClass(\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 // Get Tensor constructor - use any for flexible access\n const TensorClass = (ort as any).Tensor;\n\n // Create input tensor for current step\n const inputTensor = new TensorClass(\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 TensorClass(\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/router/router.d.ts
CHANGED
|
@@ -1,8 +1,13 @@
|
|
|
1
1
|
import { LLMProvider, RouterConfig, ChatParams, ChatResponse, StreamChunk, ProviderType, RouterMetrics } from './types.js';
|
|
2
|
+
import { CostOptimalRouter } from './cost-optimal-router.js';
|
|
2
3
|
export declare class ModelRouter {
|
|
3
4
|
private config;
|
|
4
5
|
private providers;
|
|
5
6
|
private metrics;
|
|
7
|
+
private costOptimalRouter?;
|
|
8
|
+
private embedQuery?;
|
|
9
|
+
private embedCache;
|
|
10
|
+
private static readonly EMBED_CACHE_MAX;
|
|
6
11
|
constructor(configPath?: string);
|
|
7
12
|
private loadConfig;
|
|
8
13
|
private createConfigFromEnv;
|
|
@@ -16,6 +21,30 @@ export declare class ModelRouter {
|
|
|
16
21
|
private selectByRules;
|
|
17
22
|
private matchesRule;
|
|
18
23
|
private selectByCost;
|
|
24
|
+
/**
|
|
25
|
+
* Enable learned cost-optimal routing (ADR-073). Attaches a
|
|
26
|
+
* {@link CostOptimalRouter} plus an embedder and switches the routing mode to
|
|
27
|
+
* `'cost-optimal'`. Opt-in: until called, routing behaves exactly as before.
|
|
28
|
+
*/
|
|
29
|
+
enableCostOptimalRouting(opts: {
|
|
30
|
+
router: CostOptimalRouter;
|
|
31
|
+
embed: (text: string) => Promise<number[]> | number[];
|
|
32
|
+
}): void;
|
|
33
|
+
/**
|
|
34
|
+
* Embed `text`, caching the result. Bounded LRU: a cache hit refreshes
|
|
35
|
+
* recency; the oldest entry is evicted past {@link EMBED_CACHE_MAX}.
|
|
36
|
+
*/
|
|
37
|
+
private embedCached;
|
|
38
|
+
/** Extract the most recent user-message text to embed for routing. */
|
|
39
|
+
private lastUserText;
|
|
40
|
+
/**
|
|
41
|
+
* Learned cost-optimal selection (ADR-073): embed the query, route to the
|
|
42
|
+
* cheapest model predicted to clear the quality bar, steer params.model to it,
|
|
43
|
+
* and return the matching provider. Falls back to the {@link selectByCost}
|
|
44
|
+
* heuristic when the learned router/embedder isn't configured or errors —
|
|
45
|
+
* routing never hard-fails on the cost-optimal path.
|
|
46
|
+
*/
|
|
47
|
+
private selectByCostOptimal;
|
|
19
48
|
private selectByPerformance;
|
|
20
49
|
private handleProviderError;
|
|
21
50
|
private updateMetrics;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"router.d.ts","sourceRoot":"","sources":["../../src/router/router.ts"],"names":[],"mappings":"AAIA,OAAO,EACL,WAAW,EACX,YAAY,EACZ,UAAU,EACV,YAAY,EACZ,WAAW,EACX,YAAY,EACZ,aAAa,
|
|
1
|
+
{"version":3,"file":"router.d.ts","sourceRoot":"","sources":["../../src/router/router.ts"],"names":[],"mappings":"AAIA,OAAO,EACL,WAAW,EACX,YAAY,EACZ,UAAU,EACV,YAAY,EACZ,WAAW,EACX,YAAY,EACZ,aAAa,EAGd,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,iBAAiB,EAAE,MAAM,0BAA0B,CAAC;AAO7D,qBAAa,WAAW;IACtB,OAAO,CAAC,MAAM,CAAe;IAC7B,OAAO,CAAC,SAAS,CAA6C;IAC9D,OAAO,CAAC,OAAO,CAAgB;IAE/B,OAAO,CAAC,iBAAiB,CAAC,CAAoB;IAC9C,OAAO,CAAC,UAAU,CAAC,CAAiD;IAGpE,OAAO,CAAC,UAAU,CAA+B;IACjD,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,eAAe,CAAO;gBAElC,UAAU,CAAC,EAAE,MAAM;IAM/B,OAAO,CAAC,UAAU;IAwBlB,OAAO,CAAC,mBAAmB;IAiD3B,OAAO,CAAC,iBAAiB;IAwBzB,OAAO,CAAC,mBAAmB;IAkE3B,OAAO,CAAC,iBAAiB;IAUnB,IAAI,CAAC,MAAM,EAAE,UAAU,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,YAAY,CAAC;IAuBlE,MAAM,CAAC,MAAM,EAAE,UAAU,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,cAAc,CAAC,WAAW,CAAC;YAapE,cAAc;IAmC5B,OAAO,CAAC,kBAAkB;IAQ1B,OAAO,CAAC,aAAa;IAgBrB,OAAO,CAAC,WAAW;IAkBnB,OAAO,CAAC,YAAY;IAgBpB;;;;OAIG;IACH,wBAAwB,CAAC,IAAI,EAAE;QAC7B,MAAM,EAAE,iBAAiB,CAAC;QAC1B,KAAK,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC;KACvD,GAAG,IAAI;IAMR;;;OAGG;YACW,WAAW;IAgBzB,sEAAsE;IACtE,OAAO,CAAC,YAAY;IAepB;;;;;;OAMG;YACW,mBAAmB;IAyBjC,OAAO,CAAC,mBAAmB;YAqBb,mBAAmB;IA4BjC,OAAO,CAAC,aAAa;IA8CrB,UAAU,IAAI,aAAa;IAI3B,SAAS,IAAI,YAAY;IAIzB,YAAY,IAAI,GAAG,CAAC,YAAY,EAAE,WAAW,CAAC;CAG/C"}
|
package/dist/router/router.js
CHANGED
|
@@ -11,6 +11,13 @@ export class ModelRouter {
|
|
|
11
11
|
config;
|
|
12
12
|
providers = new Map();
|
|
13
13
|
metrics;
|
|
14
|
+
// ADR-073: learned cost-optimal routing (opt-in via enableCostOptimalRouting).
|
|
15
|
+
costOptimalRouter;
|
|
16
|
+
embedQuery;
|
|
17
|
+
// Bounded LRU embedding cache — embedding dominates the cost-optimal hot path
|
|
18
|
+
// (≈ms) vs the µs-scale k-NN, and recurring prompts are common.
|
|
19
|
+
embedCache = new Map();
|
|
20
|
+
static EMBED_CACHE_MAX = 512;
|
|
14
21
|
constructor(configPath) {
|
|
15
22
|
this.config = this.loadConfig(configPath);
|
|
16
23
|
this.initializeProviders();
|
|
@@ -227,6 +234,8 @@ export class ModelRouter {
|
|
|
227
234
|
return this.selectByRules(params, agentType);
|
|
228
235
|
case 'cost-optimized':
|
|
229
236
|
return this.selectByCost(params);
|
|
237
|
+
case 'cost-optimal':
|
|
238
|
+
return this.selectByCostOptimal(params);
|
|
230
239
|
case 'performance-optimized':
|
|
231
240
|
return this.selectByPerformance(params);
|
|
232
241
|
default:
|
|
@@ -280,6 +289,80 @@ export class ModelRouter {
|
|
|
280
289
|
}
|
|
281
290
|
return this.getDefaultProvider();
|
|
282
291
|
}
|
|
292
|
+
/**
|
|
293
|
+
* Enable learned cost-optimal routing (ADR-073). Attaches a
|
|
294
|
+
* {@link CostOptimalRouter} plus an embedder and switches the routing mode to
|
|
295
|
+
* `'cost-optimal'`. Opt-in: until called, routing behaves exactly as before.
|
|
296
|
+
*/
|
|
297
|
+
enableCostOptimalRouting(opts) {
|
|
298
|
+
this.costOptimalRouter = opts.router;
|
|
299
|
+
this.embedQuery = opts.embed;
|
|
300
|
+
this.config.routing = { ...(this.config.routing ?? { mode: 'cost-optimal' }), mode: 'cost-optimal' };
|
|
301
|
+
}
|
|
302
|
+
/**
|
|
303
|
+
* Embed `text`, caching the result. Bounded LRU: a cache hit refreshes
|
|
304
|
+
* recency; the oldest entry is evicted past {@link EMBED_CACHE_MAX}.
|
|
305
|
+
*/
|
|
306
|
+
async embedCached(text) {
|
|
307
|
+
const hit = this.embedCache.get(text);
|
|
308
|
+
if (hit) {
|
|
309
|
+
this.embedCache.delete(text); // move to most-recently-used
|
|
310
|
+
this.embedCache.set(text, hit);
|
|
311
|
+
return hit;
|
|
312
|
+
}
|
|
313
|
+
const embedding = await this.embedQuery(text);
|
|
314
|
+
this.embedCache.set(text, embedding);
|
|
315
|
+
if (this.embedCache.size > ModelRouter.EMBED_CACHE_MAX) {
|
|
316
|
+
const oldest = this.embedCache.keys().next().value;
|
|
317
|
+
if (oldest !== undefined)
|
|
318
|
+
this.embedCache.delete(oldest);
|
|
319
|
+
}
|
|
320
|
+
return embedding;
|
|
321
|
+
}
|
|
322
|
+
/** Extract the most recent user-message text to embed for routing. */
|
|
323
|
+
lastUserText(messages) {
|
|
324
|
+
for (let i = messages.length - 1; i >= 0; i--) {
|
|
325
|
+
const m = messages[i];
|
|
326
|
+
if (m.role !== 'user')
|
|
327
|
+
continue;
|
|
328
|
+
return typeof m.content === 'string'
|
|
329
|
+
? m.content
|
|
330
|
+
: m.content.map((b) => b.text ?? '').join(' ');
|
|
331
|
+
}
|
|
332
|
+
// No user turn — fall back to all text content.
|
|
333
|
+
return messages
|
|
334
|
+
.map((m) => (typeof m.content === 'string' ? m.content : ''))
|
|
335
|
+
.join(' ')
|
|
336
|
+
.trim();
|
|
337
|
+
}
|
|
338
|
+
/**
|
|
339
|
+
* Learned cost-optimal selection (ADR-073): embed the query, route to the
|
|
340
|
+
* cheapest model predicted to clear the quality bar, steer params.model to it,
|
|
341
|
+
* and return the matching provider. Falls back to the {@link selectByCost}
|
|
342
|
+
* heuristic when the learned router/embedder isn't configured or errors —
|
|
343
|
+
* routing never hard-fails on the cost-optimal path.
|
|
344
|
+
*/
|
|
345
|
+
async selectByCostOptimal(params) {
|
|
346
|
+
if (this.costOptimalRouter && this.embedQuery) {
|
|
347
|
+
try {
|
|
348
|
+
const text = this.lastUserText(params.messages);
|
|
349
|
+
const embedding = await this.embedCached(text);
|
|
350
|
+
const decision = this.costOptimalRouter.route(embedding);
|
|
351
|
+
const provider = this.providers.get(decision.provider);
|
|
352
|
+
if (provider) {
|
|
353
|
+
params.model = decision.model; // steer model, not just provider
|
|
354
|
+
console.log(`💡 Cost-optimal routing → ${decision.id} ` +
|
|
355
|
+
`(q̂=${decision.predictedQuality.toFixed(2)}, $${decision.costPerMTok}/Mtok, metBar=${decision.metBar})`);
|
|
356
|
+
return provider;
|
|
357
|
+
}
|
|
358
|
+
console.warn(`⚠️ Cost-optimal pick '${decision.provider}' not initialized; falling back to heuristic`);
|
|
359
|
+
}
|
|
360
|
+
catch (err) {
|
|
361
|
+
console.warn('⚠️ Cost-optimal routing failed; falling back to heuristic', err);
|
|
362
|
+
}
|
|
363
|
+
}
|
|
364
|
+
return this.selectByCost(params);
|
|
365
|
+
}
|
|
283
366
|
selectByPerformance(params) {
|
|
284
367
|
// For now, use metrics to select fastest provider
|
|
285
368
|
let fastestProvider = null;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"router.js","sourceRoot":"","sources":["../../src/router/router.ts"],"names":[],"mappings":"AAAA,yCAAyC;AACzC,OAAO,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,IAAI,CAAC;AAC9C,OAAO,EAAE,OAAO,EAAE,MAAM,IAAI,CAAC;AAC7B,OAAO,EAAE,IAAI,EAAE,MAAM,MAAM,CAAC;AAW5B,OAAO,EAAE,kBAAkB,EAAE,MAAM,2BAA2B,CAAC;AAC/D,OAAO,EAAE,iBAAiB,EAAE,MAAM,0BAA0B,CAAC;AAC7D,OAAO,EAAE,iBAAiB,EAAE,MAAM,2BAA2B,CAAC;AAC9D,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AACvD,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AAEvD,MAAM,OAAO,WAAW;IACd,MAAM,CAAe;IACrB,SAAS,GAAmC,IAAI,GAAG,EAAE,CAAC;IACtD,OAAO,CAAgB;IAE/B,YAAY,UAAmB;QAC7B,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,CAAC;QAC1C,IAAI,CAAC,mBAAmB,EAAE,CAAC;QAC3B,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,iBAAiB,EAAE,CAAC;IAC1C,CAAC;IAEO,UAAU,CAAC,UAAmB;QACpC,MAAM,KAAK,GAAG;YACZ,UAAU;YACV,OAAO,CAAC,GAAG,CAAC,0BAA0B;YACtC,IAAI,CAAC,OAAO,EAAE,EAAE,eAAe,EAAE,oBAAoB,CAAC;YACtD,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,oBAAoB,CAAC;YACzC,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,QAAQ,EAAE,oBAAoB,CAAC;YACnD,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,4BAA4B,CAAC;SAClD,CAAC,MAAM,CAAC,OAAO,CAAa,CAAC;QAE9B,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,IAAI,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;gBACrB,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;gBAC5C,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;gBAEnC,mCAAmC;gBACnC,OAAO,IAAI,CAAC,iBAAiB,CAAC,MAAM,CAAC,CAAC;YACxC,CAAC;QACH,CAAC;QAED,oEAAoE;QACpE,OAAO,IAAI,CAAC,mBAAmB,EAAE,CAAC;IACpC,CAAC;IAEO,mBAAmB;QACzB,mDAAmD;QACnD,MAAM,MAAM,GAAQ;YAClB,OAAO,EAAE,KAAK;YACd,eAAe,EAAG,OAAO,CAAC,GAAG,CAAC,QAAgB,IAAI,WAAW;YAC7D,OAAO,EAAE,EAAE,IAAI,EAAE,QAAiB,EAAE;YACpC,SAAS,EAAE,EAAS;SACrB,CAAC;QAEF,kCAAkC;QAClC,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,EAAE,CAAC;YAClC,MAAM,CAAC,SAAS,CAAC,SAAS,GAAG;gBAC3B,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,iBAAiB;gBACrC,OAAO,EAAE,OAAO,CAAC,GAAG,CAAC,kBAAkB;aACxC,CAAC;QACJ,CAAC;QAED,mCAAmC;QACnC,IAAI,OAAO,CAAC,GAAG,CAAC,kBAAkB,EAAE,CAAC;YACnC,MAAM,CAAC,SAAS,CAAC,UAAU,GAAG;gBAC5B,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,kBAAkB;gBACtC,OAAO,EAAE,OAAO,CAAC,GAAG,CAAC,mBAAmB;aACzC,CAAC;QACJ,CAAC;QAED,+BAA+B;QAC/B,IAAI,OAAO,CAAC,GAAG,CAAC,qBAAqB,EAAE,CAAC;YACtC,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG;gBACxB,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,qBAAqB;aAC1C,CAAC;QACJ,CAAC;QAED,+EAA+E;QAC/E,IAAI,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,OAAO,CAAC,GAAG,CAAC,eAAe,EAAE,CAAC;YAC9D,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG;gBACxB,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,cAAc;gBAClC,OAAO,EAAE,OAAO,CAAC,GAAG,CAAC,eAAe;aACrC,CAAC;QACJ,CAAC;QAED,+CAA+C;QAC/C,MAAM,CAAC,SAAS,CAAC,IAAI,GAAG;YACtB,SAAS,EAAE,OAAO,CAAC,GAAG,CAAC,eAAe;YACtC,kBAAkB,EAAE,CAAC,KAAK,CAAC;SAC5B,CAAC;QAEF,OAAO,MAAsB,CAAC;IAChC,CAAC;IAEO,iBAAiB,CAAC,GAAQ;QAChC,IAAI,OAAO,GAAG,KAAK,QAAQ,EAAE,CAAC;YAC5B,sDAAsD;YACtD,OAAO,GAAG,CAAC,OAAO,CAAC,gBAAgB,EAAE,CAAC,CAAC,EAAE,GAAG,EAAE,EAAE;gBAC9C,MAAM,CAAC,OAAO,EAAE,YAAY,CAAC,GAAG,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBAChD,OAAO,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,YAAY,IAAI,EAAE,CAAC;YACpD,CAAC,CAAC,CAAC;QACL,CAAC;QAED,IAAI,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;YACvB,OAAO,GAAG,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC;QACzD,CAAC;QAED,IAAI,GAAG,IAAI,OAAO,GAAG,KAAK,QAAQ,EAAE,CAAC;YACnC,MAAM,MAAM,GAAQ,EAAE,CAAC;YACvB,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;gBAC/C,MAAM,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC,iBAAiB,CAAC,KAAK,CAAC,CAAC;YAC9C,CAAC;YACD,OAAO,MAAM,CAAC;QAChB,CAAC;QAED,OAAO,GAAG,CAAC;IACb,CAAC;IAEO,mBAAmB;QACzB,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,cAAc,KAAK,MAAM,CAAC;QAEtD,uBAAuB;QACvB,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,SAAS,EAAE,CAAC;YACpC,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,IAAI,iBAAiB,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,SAAS,CAAC,CAAC;gBACxE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,WAAW,EAAE,QAAQ,CAAC,CAAC;gBAC1C,IAAI,OAAO;oBAAE,OAAO,CAAC,GAAG,CAAC,kCAAkC,CAAC,CAAC;YAC/D,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,IAAI,OAAO;oBAAE,OAAO,CAAC,KAAK,CAAC,mCAAmC,EAAE,KAAK,CAAC,CAAC;YACzE,CAAC;QACH,CAAC;QAED,wBAAwB;QACxB,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,UAAU,EAAE,CAAC;YACrC,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,IAAI,kBAAkB,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,UAAU,CAAC,CAAC;gBAC1E,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,YAAY,EAAE,QAAQ,CAAC,CAAC;gBAC3C,IAAI,OAAO;oBAAE,OAAO,CAAC,GAAG,CAAC,mCAAmC,CAAC,CAAC;YAChE,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,IAAI,OAAO;oBAAE,OAAO,CAAC,KAAK,CAAC,oCAAoC,EAAE,KAAK,CAAC,CAAC;YAC1E,CAAC;QACH,CAAC;QAED,wBAAwB;QACxB,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC;YAC/B,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,IAAI,iBAAiB,CAAC;oBACrC,SAAS,EACP,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,SAAS;wBACpC,4EAA4E;oBAC9E,kBAAkB,EAAE,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,kBAAkB,IAAI,CAAC,KAAK,CAAC;oBAC5E,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,SAAS,IAAI,GAAG;oBACtD,WAAW,EAAE,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,WAAW,IAAI,GAAG;iBAC3D,CAAC,CAAC;gBACH,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,EAAE,QAAQ,CAAC,CAAC;gBACrC,IAAI,OAAO;oBAAE,OAAO,CAAC,GAAG,CAAC,mCAAmC,CAAC,CAAC;YAChE,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,IAAI,OAAO;oBAAE,OAAO,CAAC,KAAK,CAAC,8BAA8B,EAAE,KAAK,CAAC,CAAC;YACpE,CAAC;QACH,CAAC;QAED,oBAAoB;QACpB,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,EAAE,CAAC;YACjC,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,IAAI,cAAc,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;gBAClE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;gBACvC,IAAI,OAAO;oBAAE,OAAO,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;YAC5D,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,IAAI,OAAO;oBAAE,OAAO,CAAC,KAAK,CAAC,gCAAgC,EAAE,KAAK,CAAC,CAAC;YACtE,CAAC;QACH,CAAC;QAED,qEAAqE;QACrE,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,EAAE,CAAC;YACjC,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,IAAI,cAAc,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;gBAClE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;gBACvC,IAAI,OAAO;oBAAE,OAAO,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;YAC5D,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,IAAI,OAAO;oBAAE,OAAO,CAAC,KAAK,CAAC,gCAAgC,EAAE,KAAK,CAAC,CAAC;YACtE,CAAC;QACH,CAAC;IACH,CAAC;IAEO,iBAAiB;QACvB,OAAO;YACL,aAAa,EAAE,CAAC;YAChB,SAAS,EAAE,CAAC;YACZ,WAAW,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE;YACpC,iBAAiB,EAAE,EAAE;YACrB,cAAc,EAAE,EAAE;SACnB,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,MAAkB,EAAE,SAAkB;QAC/C,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QAE9D,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YAE7C,iBAAiB;YACjB,IAAI,CAAC,aAAa,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,SAAS,CAAC,CAAC;YAE/E,eAAe;YACf,QAAQ,CAAC,QAAQ,GAAG;gBAClB,GAAG,QAAQ,CAAC,QAAQ;gBACpB,QAAQ,EAAE,QAAQ,CAAC,IAAI;gBACvB,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;aAChC,CAAC;YAEF,OAAO,QAAQ,CAAC;QAClB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,IAAI,CAAC,mBAAmB,CAAC,KAAsB,EAAE,MAAM,EAAE,SAAS,CAAC,CAAC;QAC7E,CAAC;IACH,CAAC;IAED,KAAK,CAAC,CAAC,MAAM,CAAC,MAAkB,EAAE,SAAkB;QAClD,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QAE9D,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC;YACrB,MAAM,IAAI,KAAK,CAAC,YAAY,QAAQ,CAAC,IAAI,6BAA6B,CAAC,CAAC;QAC1E,CAAC;QAED,MAAM,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QACzC,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,QAAQ,EAAE,CAAC;YACnC,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;IAEO,KAAK,CAAC,cAAc,CAAC,MAAkB,EAAE,SAAkB;QACjE,wDAAwD;QACxD,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACpB,MAAM,cAAc,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,QAAwB,CAAC,CAAC;YAC3E,IAAI,cAAc,EAAE,CAAC;gBACnB,OAAO,cAAc,CAAC;YACxB,CAAC;YACD,OAAO,CAAC,IAAI,CACV,2BAA2B,MAAM,CAAC,QAAQ,gDAAgD,CAC3F,CAAC;QACJ,CAAC;QAED,MAAM,WAAW,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,IAAI,IAAI,QAAQ,CAAC;QAE1D,QAAQ,WAAW,EAAE,CAAC;YACpB,KAAK,QAAQ;gBACX,OAAO,IAAI,CAAC,kBAAkB,EAAE,CAAC;YAEnC,KAAK,YAAY;gBACf,OAAO,IAAI,CAAC,aAAa,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;YAE/C,KAAK,gBAAgB;gBACnB,OAAO,IAAI,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC;YAEnC,KAAK,uBAAuB;gBAC1B,OAAO,IAAI,CAAC,mBAAmB,CAAC,MAAM,CAAC,CAAC;YAE1C;gBACE,OAAO,IAAI,CAAC,kBAAkB,EAAE,CAAC;QACrC,CAAC;IACH,CAAC;IAEO,kBAAkB;QACxB,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC;QACjE,IAAI,CAAC,QAAQ,EAAE,CAAC;YACd,MAAM,IAAI,KAAK,CAAC,oBAAoB,IAAI,CAAC,MAAM,CAAC,eAAe,kBAAkB,CAAC,CAAC;QACrF,CAAC;QACD,OAAO,QAAQ,CAAC;IAClB,CAAC;IAEO,aAAa,CAAC,MAAkB,EAAE,SAAkB;QAC1D,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,KAAK,IAAI,EAAE,CAAC;QAE/C,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,IAAI,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,CAAC;gBACxD,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;gBAC1D,IAAI,QAAQ,EAAE,CAAC;oBACb,OAAO,CAAC,GAAG,CAAC,wBAAwB,IAAI,CAAC,MAAM,IAAI,mBAAmB,EAAE,CAAC,CAAC;oBAC1E,OAAO,QAAQ,CAAC;gBAClB,CAAC;YACH,CAAC;QACH,CAAC;QAED,OAAO,IAAI,CAAC,kBAAkB,EAAE,CAAC;IACnC,CAAC;IAEO,WAAW,CAAC,SAAc,EAAE,MAAkB,EAAE,SAAkB;QACxE,IAAI,SAAS,CAAC,SAAS,IAAI,SAAS,EAAE,CAAC;YACrC,IAAI,CAAC,SAAS,CAAC,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;gBAC7C,OAAO,KAAK,CAAC;YACf,CAAC;QACH,CAAC;QAED,IAAI,SAAS,CAAC,aAAa,KAAK,SAAS,EAAE,CAAC;YAC1C,IAAI,SAAS,CAAC,aAAa,IAAI,CAAC,CAAC,MAAM,CAAC,KAAK,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,EAAE,CAAC;gBAC5E,OAAO,KAAK,CAAC;YACf,CAAC;QACH,CAAC;QAED,0CAA0C;QAE1C,OAAO,IAAI,CAAC;IACd,CAAC;IAEO,YAAY,CAAC,MAAkB;QACrC,oCAAoC;QACpC,0CAA0C;QAC1C,MAAM,aAAa,GAAmB,CAAC,YAAY,EAAE,WAAW,EAAE,QAAQ,CAAC,CAAC;QAE5E,KAAK,MAAM,YAAY,IAAI,aAAa,EAAE,CAAC;YACzC,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;YAClD,IAAI,QAAQ,EAAE,CAAC;gBACb,OAAO,CAAC,GAAG,CAAC,uCAAuC,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC;gBACpE,OAAO,QAAQ,CAAC;YAClB,CAAC;QACH,CAAC;QAED,OAAO,IAAI,CAAC,kBAAkB,EAAE,CAAC;IACnC,CAAC;IAEO,mBAAmB,CAAC,MAAkB;QAC5C,kDAAkD;QAClD,IAAI,eAAe,GAAuB,IAAI,CAAC;QAC/C,IAAI,aAAa,GAAG,QAAQ,CAAC;QAE7B,KAAK,MAAM,CAAC,YAAY,EAAE,QAAQ,CAAC,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;YACtD,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,iBAAiB,CAAC,YAAY,CAAC,CAAC;YAC/D,IAAI,SAAS,IAAI,SAAS,CAAC,UAAU,GAAG,aAAa,EAAE,CAAC;gBACtD,aAAa,GAAG,SAAS,CAAC,UAAU,CAAC;gBACrC,eAAe,GAAG,QAAQ,CAAC;YAC7B,CAAC;QACH,CAAC;QAED,IAAI,eAAe,EAAE,CAAC;YACpB,OAAO,CAAC,GAAG,CAAC,6CAA6C,eAAe,CAAC,IAAI,EAAE,CAAC,CAAC;YACjF,OAAO,eAAe,CAAC;QACzB,CAAC;QAED,OAAO,IAAI,CAAC,kBAAkB,EAAE,CAAC;IACnC,CAAC;IAEO,KAAK,CAAC,mBAAmB,CAC/B,KAAoB,EACpB,MAAkB,EAClB,SAAkB;QAElB,OAAO,CAAC,KAAK,CAAC,yBAAyB,KAAK,CAAC,QAAQ,GAAG,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;QAEzE,qBAAqB;QACrB,MAAM,aAAa,GAAG,IAAI,CAAC,MAAM,CAAC,aAAa,IAAI,EAAE,CAAC;QAEtD,KAAK,MAAM,YAAY,IAAI,aAAa,EAAE,CAAC;YACzC,IAAI,YAAY,KAAK,KAAK,CAAC,QAAQ;gBAAE,SAAS,CAAC,uBAAuB;YAEtE,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;YAClD,IAAI,QAAQ,EAAE,CAAC;gBACb,OAAO,CAAC,GAAG,CAAC,sBAAsB,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC;gBACnD,IAAI,CAAC;oBACH,OAAO,MAAM,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACrC,CAAC;gBAAC,OAAO,aAAa,EAAE,CAAC;oBACvB,OAAO,CAAC,KAAK,CAAC,uBAAuB,QAAQ,CAAC,IAAI,cAAc,CAAC,CAAC;oBAClE,SAAS;gBACX,CAAC;YACH,CAAC;QACH,CAAC;QAED,MAAM,KAAK,CAAC,CAAC,wBAAwB;IACvC,CAAC;IAEO,aAAa,CACnB,YAAoB,EACpB,QAAsB,EACtB,OAAe,EACf,SAAkB;QAElB,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,CAAC;QAE7B,IAAI,QAAQ,CAAC,KAAK,EAAE,CAAC;YACnB,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC,KAAK,IAAI,QAAQ,CAAC,KAAK,CAAC,WAAW,CAAC;YAC7D,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC,MAAM,IAAI,QAAQ,CAAC,KAAK,CAAC,YAAY,CAAC;QACjE,CAAC;QAED,IAAI,QAAQ,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC;YAC5B,IAAI,CAAC,OAAO,CAAC,SAAS,IAAI,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC;QACnD,CAAC;QAED,qBAAqB;QACrB,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,iBAAiB,CAAC,YAAY,CAAC,EAAE,CAAC;YAClD,IAAI,CAAC,OAAO,CAAC,iBAAiB,CAAC,YAAY,CAAC,GAAG;gBAC7C,QAAQ,EAAE,CAAC;gBACX,IAAI,EAAE,CAAC;gBACP,UAAU,EAAE,CAAC;gBACb,MAAM,EAAE,CAAC;aACV,CAAC;QACJ,CAAC;QAED,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,iBAAiB,CAAC,YAAY,CAAC,CAAC;QAC/D,SAAS,CAAC,QAAQ,EAAE,CAAC;QACrB,SAAS,CAAC,IAAI,IAAI,QAAQ,CAAC,QAAQ,EAAE,IAAI,IAAI,CAAC,CAAC;QAC/C,SAAS,CAAC,UAAU;YAClB,CAAC,SAAS,CAAC,UAAU,GAAG,CAAC,SAAS,CAAC,QAAQ,GAAG,CAAC,CAAC,GAAG,OAAO,CAAC,GAAG,SAAS,CAAC,QAAQ,CAAC;QAEnF,kBAAkB;QAClB,IAAI,SAAS,EAAE,CAAC;YACd,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,cAAc,EAAE,CAAC;gBACjC,IAAI,CAAC,OAAO,CAAC,cAAc,GAAG,EAAE,CAAC;YACnC,CAAC;YACD,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC,SAAS,CAAC,EAAE,CAAC;gBAC5C,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC,SAAS,CAAC,GAAG,EAAE,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;YACpE,CAAC;YACD,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC,SAAS,CAAC,CAAC,QAAQ,EAAE,CAAC;YAClD,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC,SAAS,CAAC,CAAC,IAAI,IAAI,QAAQ,CAAC,QAAQ,EAAE,IAAI,IAAI,CAAC,CAAC;QAC9E,CAAC;IACH,CAAC;IAED,UAAU;QACR,OAAO,EAAE,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC;IAC7B,CAAC;IAED,SAAS;QACP,OAAO,EAAE,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC;IAC5B,CAAC;IAED,YAAY;QACV,OAAO,IAAI,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACjC,CAAC;CACF","sourcesContent":["// Multi-model router core implementation\nimport { readFileSync, existsSync } from 'fs';\nimport { homedir } from 'os';\nimport { join } from 'path';\nimport {\n LLMProvider,\n RouterConfig,\n ChatParams,\n ChatResponse,\n StreamChunk,\n ProviderType,\n RouterMetrics,\n ProviderError,\n} from './types.js';\nimport { OpenRouterProvider } from './providers/openrouter.js';\nimport { AnthropicProvider } from './providers/anthropic.js';\nimport { ONNXLocalProvider } from './providers/onnx-local.js';\nimport { GeminiProvider } from './providers/gemini.js';\nimport { OllamaProvider } from './providers/ollama.js';\n\nexport class ModelRouter {\n private config: RouterConfig;\n private providers: Map<ProviderType, LLMProvider> = new Map();\n private metrics: RouterMetrics;\n\n constructor(configPath?: string) {\n this.config = this.loadConfig(configPath);\n this.initializeProviders();\n this.metrics = this.initializeMetrics();\n }\n\n private loadConfig(configPath?: string): RouterConfig {\n const paths = [\n configPath,\n process.env.AGENTIC_FLOW_ROUTER_CONFIG,\n join(homedir(), '.agentic-flow', 'router.config.json'),\n join(process.cwd(), 'router.config.json'),\n join(process.cwd(), 'config', 'router.config.json'),\n join(process.cwd(), 'router.config.example.json'),\n ].filter(Boolean) as string[];\n\n for (const path of paths) {\n if (existsSync(path)) {\n const content = readFileSync(path, 'utf-8');\n const config = JSON.parse(content);\n\n // Substitute environment variables\n return this.substituteEnvVars(config);\n }\n }\n\n // If no config file found, create config from environment variables\n return this.createConfigFromEnv();\n }\n\n private createConfigFromEnv(): RouterConfig {\n // Create minimal config from environment variables\n const config: any = {\n version: '1.0',\n defaultProvider: (process.env.PROVIDER as any) || 'anthropic',\n routing: { mode: 'manual' as const },\n providers: {} as any,\n };\n\n // Add Anthropic if API key exists\n if (process.env.ANTHROPIC_API_KEY) {\n config.providers.anthropic = {\n apiKey: process.env.ANTHROPIC_API_KEY,\n baseUrl: process.env.ANTHROPIC_BASE_URL,\n };\n }\n\n // Add OpenRouter if API key exists\n if (process.env.OPENROUTER_API_KEY) {\n config.providers.openrouter = {\n apiKey: process.env.OPENROUTER_API_KEY,\n baseUrl: process.env.OPENROUTER_BASE_URL,\n };\n }\n\n // Add Gemini if API key exists\n if (process.env.GOOGLE_GEMINI_API_KEY) {\n config.providers.gemini = {\n apiKey: process.env.GOOGLE_GEMINI_API_KEY,\n };\n }\n\n // Add Ollama if API key OR explicit base URL is set (self-hosted needs no key)\n if (process.env.OLLAMA_API_KEY || process.env.OLLAMA_BASE_URL) {\n config.providers.ollama = {\n apiKey: process.env.OLLAMA_API_KEY,\n baseUrl: process.env.OLLAMA_BASE_URL,\n };\n }\n\n // ONNX is always available (no API key needed)\n config.providers.onnx = {\n modelPath: process.env.ONNX_MODEL_PATH,\n executionProviders: ['cpu'],\n };\n\n return config as RouterConfig;\n }\n\n private substituteEnvVars(obj: any): any {\n if (typeof obj === 'string') {\n // Replace ${VAR_NAME} with environment variable value\n return obj.replace(/\\$\\{([^}]+)\\}/g, (_, key) => {\n const [varName, defaultValue] = key.split(':-');\n return process.env[varName] || defaultValue || '';\n });\n }\n\n if (Array.isArray(obj)) {\n return obj.map((item) => this.substituteEnvVars(item));\n }\n\n if (obj && typeof obj === 'object') {\n const result: any = {};\n for (const [key, value] of Object.entries(obj)) {\n result[key] = this.substituteEnvVars(value);\n }\n return result;\n }\n\n return obj;\n }\n\n private initializeProviders(): void {\n const verbose = process.env.ROUTER_VERBOSE === 'true';\n\n // Initialize Anthropic\n if (this.config.providers.anthropic) {\n try {\n const provider = new AnthropicProvider(this.config.providers.anthropic);\n this.providers.set('anthropic', provider);\n if (verbose) console.log('✅ Anthropic provider initialized');\n } catch (error) {\n if (verbose) console.error('❌ Failed to initialize Anthropic:', error);\n }\n }\n\n // Initialize OpenRouter\n if (this.config.providers.openrouter) {\n try {\n const provider = new OpenRouterProvider(this.config.providers.openrouter);\n this.providers.set('openrouter', provider);\n if (verbose) console.log('✅ OpenRouter provider initialized');\n } catch (error) {\n if (verbose) console.error('❌ Failed to initialize OpenRouter:', error);\n }\n }\n\n // Initialize ONNX Local\n if (this.config.providers.onnx) {\n try {\n const provider = new ONNXLocalProvider({\n modelPath:\n this.config.providers.onnx.modelPath ||\n './models/phi-4/cpu_and_mobile/cpu-int4-rtn-block-32-acc-level-4/model.onnx',\n executionProviders: this.config.providers.onnx.executionProviders || ['cpu'],\n maxTokens: this.config.providers.onnx.maxTokens || 100,\n temperature: this.config.providers.onnx.temperature || 0.7,\n });\n this.providers.set('onnx', provider);\n if (verbose) console.log('✅ ONNX Local provider initialized');\n } catch (error) {\n if (verbose) console.error('❌ Failed to initialize ONNX:', error);\n }\n }\n\n // Initialize Gemini\n if (this.config.providers.gemini) {\n try {\n const provider = new GeminiProvider(this.config.providers.gemini);\n this.providers.set('gemini', provider);\n if (verbose) console.log('✅ Gemini provider initialized');\n } catch (error) {\n if (verbose) console.error('❌ Failed to initialize Gemini:', error);\n }\n }\n\n // Initialize Ollama (OpenAI-compat, self-hosted or ollama.com Cloud)\n if (this.config.providers.ollama) {\n try {\n const provider = new OllamaProvider(this.config.providers.ollama);\n this.providers.set('ollama', provider);\n if (verbose) console.log('✅ Ollama provider initialized');\n } catch (error) {\n if (verbose) console.error('❌ Failed to initialize Ollama:', error);\n }\n }\n }\n\n private initializeMetrics(): RouterMetrics {\n return {\n totalRequests: 0,\n totalCost: 0,\n totalTokens: { input: 0, output: 0 },\n providerBreakdown: {},\n agentBreakdown: {},\n };\n }\n\n async chat(params: ChatParams, agentType?: string): Promise<ChatResponse> {\n const startTime = Date.now();\n const provider = await this.selectProvider(params, agentType);\n\n try {\n const response = await provider.chat(params);\n\n // Update metrics\n this.updateMetrics(provider.name, response, Date.now() - startTime, agentType);\n\n // Add metadata\n response.metadata = {\n ...response.metadata,\n provider: provider.name,\n latency: Date.now() - startTime,\n };\n\n return response;\n } catch (error) {\n return this.handleProviderError(error as ProviderError, params, agentType);\n }\n }\n\n async *stream(params: ChatParams, agentType?: string): AsyncGenerator<StreamChunk> {\n const provider = await this.selectProvider(params, agentType);\n\n if (!provider.stream) {\n throw new Error(`Provider ${provider.name} does not support streaming`);\n }\n\n const iterator = provider.stream(params);\n for await (const chunk of iterator) {\n yield chunk;\n }\n }\n\n private async selectProvider(params: ChatParams, agentType?: string): Promise<LLMProvider> {\n // If provider is explicitly specified in params, use it\n if (params.provider) {\n const forcedProvider = this.providers.get(params.provider as ProviderType);\n if (forcedProvider) {\n return forcedProvider;\n }\n console.warn(\n `⚠️ Requested provider '${params.provider}' not available, falling back to routing logic`\n );\n }\n\n const routingMode = this.config.routing?.mode || 'manual';\n\n switch (routingMode) {\n case 'manual':\n return this.getDefaultProvider();\n\n case 'rule-based':\n return this.selectByRules(params, agentType);\n\n case 'cost-optimized':\n return this.selectByCost(params);\n\n case 'performance-optimized':\n return this.selectByPerformance(params);\n\n default:\n return this.getDefaultProvider();\n }\n }\n\n private getDefaultProvider(): LLMProvider {\n const provider = this.providers.get(this.config.defaultProvider);\n if (!provider) {\n throw new Error(`Default provider ${this.config.defaultProvider} not initialized`);\n }\n return provider;\n }\n\n private selectByRules(params: ChatParams, agentType?: string): LLMProvider {\n const rules = this.config.routing?.rules || [];\n\n for (const rule of rules) {\n if (this.matchesRule(rule.condition, params, agentType)) {\n const provider = this.providers.get(rule.action.provider);\n if (provider) {\n console.log(`🎯 Routing via rule: ${rule.reason || 'matched condition'}`);\n return provider;\n }\n }\n }\n\n return this.getDefaultProvider();\n }\n\n private matchesRule(condition: any, params: ChatParams, agentType?: string): boolean {\n if (condition.agentType && agentType) {\n if (!condition.agentType.includes(agentType)) {\n return false;\n }\n }\n\n if (condition.requiresTools !== undefined) {\n if (condition.requiresTools && (!params.tools || params.tools.length === 0)) {\n return false;\n }\n }\n\n // TODO: Add more condition matching logic\n\n return true;\n }\n\n private selectByCost(params: ChatParams): LLMProvider {\n // For now, prefer cheaper providers\n // TODO: Implement actual cost calculation\n const providerOrder: ProviderType[] = ['openrouter', 'anthropic', 'openai'];\n\n for (const providerType of providerOrder) {\n const provider = this.providers.get(providerType);\n if (provider) {\n console.log(`💰 Cost-optimized routing: selected ${provider.name}`);\n return provider;\n }\n }\n\n return this.getDefaultProvider();\n }\n\n private selectByPerformance(params: ChatParams): LLMProvider {\n // For now, use metrics to select fastest provider\n let fastestProvider: LLMProvider | null = null;\n let lowestLatency = Infinity;\n\n for (const [providerType, provider] of this.providers) {\n const breakdown = this.metrics.providerBreakdown[providerType];\n if (breakdown && breakdown.avgLatency < lowestLatency) {\n lowestLatency = breakdown.avgLatency;\n fastestProvider = provider;\n }\n }\n\n if (fastestProvider) {\n console.log(`⚡ Performance-optimized routing: selected ${fastestProvider.name}`);\n return fastestProvider;\n }\n\n return this.getDefaultProvider();\n }\n\n private async handleProviderError(\n error: ProviderError,\n params: ChatParams,\n agentType?: string\n ): Promise<ChatResponse> {\n console.error(`❌ Provider error from ${error.provider}:`, error.message);\n\n // Try fallback chain\n const fallbackChain = this.config.fallbackChain || [];\n\n for (const providerType of fallbackChain) {\n if (providerType === error.provider) continue; // Skip failed provider\n\n const provider = this.providers.get(providerType);\n if (provider) {\n console.log(`🔄 Falling back to ${provider.name}`);\n try {\n return await provider.chat(params);\n } catch (fallbackError) {\n console.error(`❌ Fallback provider ${provider.name} also failed`);\n continue;\n }\n }\n }\n\n throw error; // No fallback succeeded\n }\n\n private updateMetrics(\n providerName: string,\n response: ChatResponse,\n latency: number,\n agentType?: string\n ): void {\n this.metrics.totalRequests++;\n\n if (response.usage) {\n this.metrics.totalTokens.input += response.usage.inputTokens;\n this.metrics.totalTokens.output += response.usage.outputTokens;\n }\n\n if (response.metadata?.cost) {\n this.metrics.totalCost += response.metadata.cost;\n }\n\n // Provider breakdown\n if (!this.metrics.providerBreakdown[providerName]) {\n this.metrics.providerBreakdown[providerName] = {\n requests: 0,\n cost: 0,\n avgLatency: 0,\n errors: 0,\n };\n }\n\n const breakdown = this.metrics.providerBreakdown[providerName];\n breakdown.requests++;\n breakdown.cost += response.metadata?.cost || 0;\n breakdown.avgLatency =\n (breakdown.avgLatency * (breakdown.requests - 1) + latency) / breakdown.requests;\n\n // Agent breakdown\n if (agentType) {\n if (!this.metrics.agentBreakdown) {\n this.metrics.agentBreakdown = {};\n }\n if (!this.metrics.agentBreakdown[agentType]) {\n this.metrics.agentBreakdown[agentType] = { requests: 0, cost: 0 };\n }\n this.metrics.agentBreakdown[agentType].requests++;\n this.metrics.agentBreakdown[agentType].cost += response.metadata?.cost || 0;\n }\n }\n\n getMetrics(): RouterMetrics {\n return { ...this.metrics };\n }\n\n getConfig(): RouterConfig {\n return { ...this.config };\n }\n\n getProviders(): Map<ProviderType, LLMProvider> {\n return new Map(this.providers);\n }\n}\n"]}
|
|
1
|
+
{"version":3,"file":"router.js","sourceRoot":"","sources":["../../src/router/router.ts"],"names":[],"mappings":"AAAA,yCAAyC;AACzC,OAAO,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,IAAI,CAAC;AAC9C,OAAO,EAAE,OAAO,EAAE,MAAM,IAAI,CAAC;AAC7B,OAAO,EAAE,IAAI,EAAE,MAAM,MAAM,CAAC;AAa5B,OAAO,EAAE,kBAAkB,EAAE,MAAM,2BAA2B,CAAC;AAC/D,OAAO,EAAE,iBAAiB,EAAE,MAAM,0BAA0B,CAAC;AAC7D,OAAO,EAAE,iBAAiB,EAAE,MAAM,2BAA2B,CAAC;AAC9D,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AACvD,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AAEvD,MAAM,OAAO,WAAW;IACd,MAAM,CAAe;IACrB,SAAS,GAAmC,IAAI,GAAG,EAAE,CAAC;IACtD,OAAO,CAAgB;IAC/B,+EAA+E;IACvE,iBAAiB,CAAqB;IACtC,UAAU,CAAkD;IACpE,8EAA8E;IAC9E,gEAAgE;IACxD,UAAU,GAAG,IAAI,GAAG,EAAoB,CAAC;IACzC,MAAM,CAAU,eAAe,GAAG,GAAG,CAAC;IAE9C,YAAY,UAAmB;QAC7B,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,CAAC;QAC1C,IAAI,CAAC,mBAAmB,EAAE,CAAC;QAC3B,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,iBAAiB,EAAE,CAAC;IAC1C,CAAC;IAEO,UAAU,CAAC,UAAmB;QACpC,MAAM,KAAK,GAAG;YACZ,UAAU;YACV,OAAO,CAAC,GAAG,CAAC,0BAA0B;YACtC,IAAI,CAAC,OAAO,EAAE,EAAE,eAAe,EAAE,oBAAoB,CAAC;YACtD,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,oBAAoB,CAAC;YACzC,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,QAAQ,EAAE,oBAAoB,CAAC;YACnD,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,4BAA4B,CAAC;SAClD,CAAC,MAAM,CAAC,OAAO,CAAa,CAAC;QAE9B,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,IAAI,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;gBACrB,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;gBAC5C,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;gBAEnC,mCAAmC;gBACnC,OAAO,IAAI,CAAC,iBAAiB,CAAC,MAAM,CAAC,CAAC;YACxC,CAAC;QACH,CAAC;QAED,oEAAoE;QACpE,OAAO,IAAI,CAAC,mBAAmB,EAAE,CAAC;IACpC,CAAC;IAEO,mBAAmB;QACzB,mDAAmD;QACnD,MAAM,MAAM,GAAQ;YAClB,OAAO,EAAE,KAAK;YACd,eAAe,EAAG,OAAO,CAAC,GAAG,CAAC,QAAgB,IAAI,WAAW;YAC7D,OAAO,EAAE,EAAE,IAAI,EAAE,QAAiB,EAAE;YACpC,SAAS,EAAE,EAAS;SACrB,CAAC;QAEF,kCAAkC;QAClC,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,EAAE,CAAC;YAClC,MAAM,CAAC,SAAS,CAAC,SAAS,GAAG;gBAC3B,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,iBAAiB;gBACrC,OAAO,EAAE,OAAO,CAAC,GAAG,CAAC,kBAAkB;aACxC,CAAC;QACJ,CAAC;QAED,mCAAmC;QACnC,IAAI,OAAO,CAAC,GAAG,CAAC,kBAAkB,EAAE,CAAC;YACnC,MAAM,CAAC,SAAS,CAAC,UAAU,GAAG;gBAC5B,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,kBAAkB;gBACtC,OAAO,EAAE,OAAO,CAAC,GAAG,CAAC,mBAAmB;aACzC,CAAC;QACJ,CAAC;QAED,+BAA+B;QAC/B,IAAI,OAAO,CAAC,GAAG,CAAC,qBAAqB,EAAE,CAAC;YACtC,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG;gBACxB,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,qBAAqB;aAC1C,CAAC;QACJ,CAAC;QAED,+EAA+E;QAC/E,IAAI,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,OAAO,CAAC,GAAG,CAAC,eAAe,EAAE,CAAC;YAC9D,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG;gBACxB,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,cAAc;gBAClC,OAAO,EAAE,OAAO,CAAC,GAAG,CAAC,eAAe;aACrC,CAAC;QACJ,CAAC;QAED,+CAA+C;QAC/C,MAAM,CAAC,SAAS,CAAC,IAAI,GAAG;YACtB,SAAS,EAAE,OAAO,CAAC,GAAG,CAAC,eAAe;YACtC,kBAAkB,EAAE,CAAC,KAAK,CAAC;SAC5B,CAAC;QAEF,OAAO,MAAsB,CAAC;IAChC,CAAC;IAEO,iBAAiB,CAAC,GAAQ;QAChC,IAAI,OAAO,GAAG,KAAK,QAAQ,EAAE,CAAC;YAC5B,sDAAsD;YACtD,OAAO,GAAG,CAAC,OAAO,CAAC,gBAAgB,EAAE,CAAC,CAAC,EAAE,GAAG,EAAE,EAAE;gBAC9C,MAAM,CAAC,OAAO,EAAE,YAAY,CAAC,GAAG,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBAChD,OAAO,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,YAAY,IAAI,EAAE,CAAC;YACpD,CAAC,CAAC,CAAC;QACL,CAAC;QAED,IAAI,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;YACvB,OAAO,GAAG,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC;QACzD,CAAC;QAED,IAAI,GAAG,IAAI,OAAO,GAAG,KAAK,QAAQ,EAAE,CAAC;YACnC,MAAM,MAAM,GAAQ,EAAE,CAAC;YACvB,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;gBAC/C,MAAM,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC,iBAAiB,CAAC,KAAK,CAAC,CAAC;YAC9C,CAAC;YACD,OAAO,MAAM,CAAC;QAChB,CAAC;QAED,OAAO,GAAG,CAAC;IACb,CAAC;IAEO,mBAAmB;QACzB,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,cAAc,KAAK,MAAM,CAAC;QAEtD,uBAAuB;QACvB,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,SAAS,EAAE,CAAC;YACpC,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,IAAI,iBAAiB,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,SAAS,CAAC,CAAC;gBACxE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,WAAW,EAAE,QAAQ,CAAC,CAAC;gBAC1C,IAAI,OAAO;oBAAE,OAAO,CAAC,GAAG,CAAC,kCAAkC,CAAC,CAAC;YAC/D,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,IAAI,OAAO;oBAAE,OAAO,CAAC,KAAK,CAAC,mCAAmC,EAAE,KAAK,CAAC,CAAC;YACzE,CAAC;QACH,CAAC;QAED,wBAAwB;QACxB,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,UAAU,EAAE,CAAC;YACrC,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,IAAI,kBAAkB,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,UAAU,CAAC,CAAC;gBAC1E,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,YAAY,EAAE,QAAQ,CAAC,CAAC;gBAC3C,IAAI,OAAO;oBAAE,OAAO,CAAC,GAAG,CAAC,mCAAmC,CAAC,CAAC;YAChE,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,IAAI,OAAO;oBAAE,OAAO,CAAC,KAAK,CAAC,oCAAoC,EAAE,KAAK,CAAC,CAAC;YAC1E,CAAC;QACH,CAAC;QAED,wBAAwB;QACxB,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC;YAC/B,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,IAAI,iBAAiB,CAAC;oBACrC,SAAS,EACP,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,SAAS;wBACpC,4EAA4E;oBAC9E,kBAAkB,EAAE,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,kBAAkB,IAAI,CAAC,KAAK,CAAC;oBAC5E,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,SAAS,IAAI,GAAG;oBACtD,WAAW,EAAE,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,WAAW,IAAI,GAAG;iBAC3D,CAAC,CAAC;gBACH,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,EAAE,QAAQ,CAAC,CAAC;gBACrC,IAAI,OAAO;oBAAE,OAAO,CAAC,GAAG,CAAC,mCAAmC,CAAC,CAAC;YAChE,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,IAAI,OAAO;oBAAE,OAAO,CAAC,KAAK,CAAC,8BAA8B,EAAE,KAAK,CAAC,CAAC;YACpE,CAAC;QACH,CAAC;QAED,oBAAoB;QACpB,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,EAAE,CAAC;YACjC,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,IAAI,cAAc,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;gBAClE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;gBACvC,IAAI,OAAO;oBAAE,OAAO,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;YAC5D,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,IAAI,OAAO;oBAAE,OAAO,CAAC,KAAK,CAAC,gCAAgC,EAAE,KAAK,CAAC,CAAC;YACtE,CAAC;QACH,CAAC;QAED,qEAAqE;QACrE,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,EAAE,CAAC;YACjC,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,IAAI,cAAc,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;gBAClE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;gBACvC,IAAI,OAAO;oBAAE,OAAO,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;YAC5D,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,IAAI,OAAO;oBAAE,OAAO,CAAC,KAAK,CAAC,gCAAgC,EAAE,KAAK,CAAC,CAAC;YACtE,CAAC;QACH,CAAC;IACH,CAAC;IAEO,iBAAiB;QACvB,OAAO;YACL,aAAa,EAAE,CAAC;YAChB,SAAS,EAAE,CAAC;YACZ,WAAW,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE;YACpC,iBAAiB,EAAE,EAAE;YACrB,cAAc,EAAE,EAAE;SACnB,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,MAAkB,EAAE,SAAkB;QAC/C,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QAE9D,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YAE7C,iBAAiB;YACjB,IAAI,CAAC,aAAa,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,SAAS,CAAC,CAAC;YAE/E,eAAe;YACf,QAAQ,CAAC,QAAQ,GAAG;gBAClB,GAAG,QAAQ,CAAC,QAAQ;gBACpB,QAAQ,EAAE,QAAQ,CAAC,IAAI;gBACvB,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;aAChC,CAAC;YAEF,OAAO,QAAQ,CAAC;QAClB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,IAAI,CAAC,mBAAmB,CAAC,KAAsB,EAAE,MAAM,EAAE,SAAS,CAAC,CAAC;QAC7E,CAAC;IACH,CAAC;IAED,KAAK,CAAC,CAAC,MAAM,CAAC,MAAkB,EAAE,SAAkB;QAClD,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QAE9D,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC;YACrB,MAAM,IAAI,KAAK,CAAC,YAAY,QAAQ,CAAC,IAAI,6BAA6B,CAAC,CAAC;QAC1E,CAAC;QAED,MAAM,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QACzC,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,QAAQ,EAAE,CAAC;YACnC,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;IAEO,KAAK,CAAC,cAAc,CAAC,MAAkB,EAAE,SAAkB;QACjE,wDAAwD;QACxD,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACpB,MAAM,cAAc,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,QAAwB,CAAC,CAAC;YAC3E,IAAI,cAAc,EAAE,CAAC;gBACnB,OAAO,cAAc,CAAC;YACxB,CAAC;YACD,OAAO,CAAC,IAAI,CACV,2BAA2B,MAAM,CAAC,QAAQ,gDAAgD,CAC3F,CAAC;QACJ,CAAC;QAED,MAAM,WAAW,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,IAAI,IAAI,QAAQ,CAAC;QAE1D,QAAQ,WAAW,EAAE,CAAC;YACpB,KAAK,QAAQ;gBACX,OAAO,IAAI,CAAC,kBAAkB,EAAE,CAAC;YAEnC,KAAK,YAAY;gBACf,OAAO,IAAI,CAAC,aAAa,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;YAE/C,KAAK,gBAAgB;gBACnB,OAAO,IAAI,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC;YAEnC,KAAK,cAAc;gBACjB,OAAO,IAAI,CAAC,mBAAmB,CAAC,MAAM,CAAC,CAAC;YAE1C,KAAK,uBAAuB;gBAC1B,OAAO,IAAI,CAAC,mBAAmB,CAAC,MAAM,CAAC,CAAC;YAE1C;gBACE,OAAO,IAAI,CAAC,kBAAkB,EAAE,CAAC;QACrC,CAAC;IACH,CAAC;IAEO,kBAAkB;QACxB,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC;QACjE,IAAI,CAAC,QAAQ,EAAE,CAAC;YACd,MAAM,IAAI,KAAK,CAAC,oBAAoB,IAAI,CAAC,MAAM,CAAC,eAAe,kBAAkB,CAAC,CAAC;QACrF,CAAC;QACD,OAAO,QAAQ,CAAC;IAClB,CAAC;IAEO,aAAa,CAAC,MAAkB,EAAE,SAAkB;QAC1D,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,KAAK,IAAI,EAAE,CAAC;QAE/C,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,IAAI,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,CAAC;gBACxD,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;gBAC1D,IAAI,QAAQ,EAAE,CAAC;oBACb,OAAO,CAAC,GAAG,CAAC,wBAAwB,IAAI,CAAC,MAAM,IAAI,mBAAmB,EAAE,CAAC,CAAC;oBAC1E,OAAO,QAAQ,CAAC;gBAClB,CAAC;YACH,CAAC;QACH,CAAC;QAED,OAAO,IAAI,CAAC,kBAAkB,EAAE,CAAC;IACnC,CAAC;IAEO,WAAW,CAAC,SAAc,EAAE,MAAkB,EAAE,SAAkB;QACxE,IAAI,SAAS,CAAC,SAAS,IAAI,SAAS,EAAE,CAAC;YACrC,IAAI,CAAC,SAAS,CAAC,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;gBAC7C,OAAO,KAAK,CAAC;YACf,CAAC;QACH,CAAC;QAED,IAAI,SAAS,CAAC,aAAa,KAAK,SAAS,EAAE,CAAC;YAC1C,IAAI,SAAS,CAAC,aAAa,IAAI,CAAC,CAAC,MAAM,CAAC,KAAK,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,EAAE,CAAC;gBAC5E,OAAO,KAAK,CAAC;YACf,CAAC;QACH,CAAC;QAED,0CAA0C;QAE1C,OAAO,IAAI,CAAC;IACd,CAAC;IAEO,YAAY,CAAC,MAAkB;QACrC,oCAAoC;QACpC,0CAA0C;QAC1C,MAAM,aAAa,GAAmB,CAAC,YAAY,EAAE,WAAW,EAAE,QAAQ,CAAC,CAAC;QAE5E,KAAK,MAAM,YAAY,IAAI,aAAa,EAAE,CAAC;YACzC,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;YAClD,IAAI,QAAQ,EAAE,CAAC;gBACb,OAAO,CAAC,GAAG,CAAC,uCAAuC,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC;gBACpE,OAAO,QAAQ,CAAC;YAClB,CAAC;QACH,CAAC;QAED,OAAO,IAAI,CAAC,kBAAkB,EAAE,CAAC;IACnC,CAAC;IAED;;;;OAIG;IACH,wBAAwB,CAAC,IAGxB;QACC,IAAI,CAAC,iBAAiB,GAAG,IAAI,CAAC,MAAM,CAAC;QACrC,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC;QAC7B,IAAI,CAAC,MAAM,CAAC,OAAO,GAAG,EAAE,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,EAAE,IAAI,EAAE,cAAc,EAAE,CAAC,EAAE,IAAI,EAAE,cAAc,EAAE,CAAC;IACvG,CAAC;IAED;;;OAGG;IACK,KAAK,CAAC,WAAW,CAAC,IAAY;QACpC,MAAM,GAAG,GAAG,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACtC,IAAI,GAAG,EAAE,CAAC;YACR,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,6BAA6B;YAC3D,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;YAC/B,OAAO,GAAG,CAAC;QACb,CAAC;QACD,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,UAAW,CAAC,IAAI,CAAC,CAAC;QAC/C,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;QACrC,IAAI,IAAI,CAAC,UAAU,CAAC,IAAI,GAAG,WAAW,CAAC,eAAe,EAAE,CAAC;YACvD,MAAM,MAAM,GAAG,IAAI,CAAC,UAAU,CAAC,IAAI,EAAE,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC;YACnD,IAAI,MAAM,KAAK,SAAS;gBAAE,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QAC3D,CAAC;QACD,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,sEAAsE;IAC9D,YAAY,CAAC,QAAmB;QACtC,KAAK,IAAI,CAAC,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;YAC9C,MAAM,CAAC,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;YACtB,IAAI,CAAC,CAAC,IAAI,KAAK,MAAM;gBAAE,SAAS;YAChC,OAAO,OAAO,CAAC,CAAC,OAAO,KAAK,QAAQ;gBAClC,CAAC,CAAC,CAAC,CAAC,OAAO;gBACX,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACnD,CAAC;QACD,gDAAgD;QAChD,OAAO,QAAQ;aACZ,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;aAC5D,IAAI,CAAC,GAAG,CAAC;aACT,IAAI,EAAE,CAAC;IACZ,CAAC;IAED;;;;;;OAMG;IACK,KAAK,CAAC,mBAAmB,CAAC,MAAkB;QAClD,IAAI,IAAI,CAAC,iBAAiB,IAAI,IAAI,CAAC,UAAU,EAAE,CAAC;YAC9C,IAAI,CAAC;gBACH,MAAM,IAAI,GAAG,IAAI,CAAC,YAAY,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;gBAChD,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC;gBAC/C,MAAM,QAAQ,GAAG,IAAI,CAAC,iBAAiB,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;gBACzD,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;gBACvD,IAAI,QAAQ,EAAE,CAAC;oBACb,MAAM,CAAC,KAAK,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,iCAAiC;oBAChE,OAAO,CAAC,GAAG,CACT,6BAA6B,QAAQ,CAAC,EAAE,GAAG;wBACzC,OAAO,QAAQ,CAAC,gBAAgB,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,QAAQ,CAAC,WAAW,iBAAiB,QAAQ,CAAC,MAAM,GAAG,CAC3G,CAAC;oBACF,OAAO,QAAQ,CAAC;gBAClB,CAAC;gBACD,OAAO,CAAC,IAAI,CACV,0BAA0B,QAAQ,CAAC,QAAQ,8CAA8C,CAC1F,CAAC;YACJ,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,OAAO,CAAC,IAAI,CAAC,4DAA4D,EAAE,GAAG,CAAC,CAAC;YAClF,CAAC;QACH,CAAC;QACD,OAAO,IAAI,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC;IACnC,CAAC;IAEO,mBAAmB,CAAC,MAAkB;QAC5C,kDAAkD;QAClD,IAAI,eAAe,GAAuB,IAAI,CAAC;QAC/C,IAAI,aAAa,GAAG,QAAQ,CAAC;QAE7B,KAAK,MAAM,CAAC,YAAY,EAAE,QAAQ,CAAC,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;YACtD,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,iBAAiB,CAAC,YAAY,CAAC,CAAC;YAC/D,IAAI,SAAS,IAAI,SAAS,CAAC,UAAU,GAAG,aAAa,EAAE,CAAC;gBACtD,aAAa,GAAG,SAAS,CAAC,UAAU,CAAC;gBACrC,eAAe,GAAG,QAAQ,CAAC;YAC7B,CAAC;QACH,CAAC;QAED,IAAI,eAAe,EAAE,CAAC;YACpB,OAAO,CAAC,GAAG,CAAC,6CAA6C,eAAe,CAAC,IAAI,EAAE,CAAC,CAAC;YACjF,OAAO,eAAe,CAAC;QACzB,CAAC;QAED,OAAO,IAAI,CAAC,kBAAkB,EAAE,CAAC;IACnC,CAAC;IAEO,KAAK,CAAC,mBAAmB,CAC/B,KAAoB,EACpB,MAAkB,EAClB,SAAkB;QAElB,OAAO,CAAC,KAAK,CAAC,yBAAyB,KAAK,CAAC,QAAQ,GAAG,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;QAEzE,qBAAqB;QACrB,MAAM,aAAa,GAAG,IAAI,CAAC,MAAM,CAAC,aAAa,IAAI,EAAE,CAAC;QAEtD,KAAK,MAAM,YAAY,IAAI,aAAa,EAAE,CAAC;YACzC,IAAI,YAAY,KAAK,KAAK,CAAC,QAAQ;gBAAE,SAAS,CAAC,uBAAuB;YAEtE,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;YAClD,IAAI,QAAQ,EAAE,CAAC;gBACb,OAAO,CAAC,GAAG,CAAC,sBAAsB,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC;gBACnD,IAAI,CAAC;oBACH,OAAO,MAAM,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACrC,CAAC;gBAAC,OAAO,aAAa,EAAE,CAAC;oBACvB,OAAO,CAAC,KAAK,CAAC,uBAAuB,QAAQ,CAAC,IAAI,cAAc,CAAC,CAAC;oBAClE,SAAS;gBACX,CAAC;YACH,CAAC;QACH,CAAC;QAED,MAAM,KAAK,CAAC,CAAC,wBAAwB;IACvC,CAAC;IAEO,aAAa,CACnB,YAAoB,EACpB,QAAsB,EACtB,OAAe,EACf,SAAkB;QAElB,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,CAAC;QAE7B,IAAI,QAAQ,CAAC,KAAK,EAAE,CAAC;YACnB,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC,KAAK,IAAI,QAAQ,CAAC,KAAK,CAAC,WAAW,CAAC;YAC7D,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC,MAAM,IAAI,QAAQ,CAAC,KAAK,CAAC,YAAY,CAAC;QACjE,CAAC;QAED,IAAI,QAAQ,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC;YAC5B,IAAI,CAAC,OAAO,CAAC,SAAS,IAAI,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC;QACnD,CAAC;QAED,qBAAqB;QACrB,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,iBAAiB,CAAC,YAAY,CAAC,EAAE,CAAC;YAClD,IAAI,CAAC,OAAO,CAAC,iBAAiB,CAAC,YAAY,CAAC,GAAG;gBAC7C,QAAQ,EAAE,CAAC;gBACX,IAAI,EAAE,CAAC;gBACP,UAAU,EAAE,CAAC;gBACb,MAAM,EAAE,CAAC;aACV,CAAC;QACJ,CAAC;QAED,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,iBAAiB,CAAC,YAAY,CAAC,CAAC;QAC/D,SAAS,CAAC,QAAQ,EAAE,CAAC;QACrB,SAAS,CAAC,IAAI,IAAI,QAAQ,CAAC,QAAQ,EAAE,IAAI,IAAI,CAAC,CAAC;QAC/C,SAAS,CAAC,UAAU;YAClB,CAAC,SAAS,CAAC,UAAU,GAAG,CAAC,SAAS,CAAC,QAAQ,GAAG,CAAC,CAAC,GAAG,OAAO,CAAC,GAAG,SAAS,CAAC,QAAQ,CAAC;QAEnF,kBAAkB;QAClB,IAAI,SAAS,EAAE,CAAC;YACd,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,cAAc,EAAE,CAAC;gBACjC,IAAI,CAAC,OAAO,CAAC,cAAc,GAAG,EAAE,CAAC;YACnC,CAAC;YACD,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC,SAAS,CAAC,EAAE,CAAC;gBAC5C,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC,SAAS,CAAC,GAAG,EAAE,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;YACpE,CAAC;YACD,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC,SAAS,CAAC,CAAC,QAAQ,EAAE,CAAC;YAClD,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC,SAAS,CAAC,CAAC,IAAI,IAAI,QAAQ,CAAC,QAAQ,EAAE,IAAI,IAAI,CAAC,CAAC;QAC9E,CAAC;IACH,CAAC;IAED,UAAU;QACR,OAAO,EAAE,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC;IAC7B,CAAC;IAED,SAAS;QACP,OAAO,EAAE,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC;IAC5B,CAAC;IAED,YAAY;QACV,OAAO,IAAI,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACjC,CAAC","sourcesContent":["// Multi-model router core implementation\nimport { readFileSync, existsSync } from 'fs';\nimport { homedir } from 'os';\nimport { join } from 'path';\nimport {\n LLMProvider,\n RouterConfig,\n ChatParams,\n ChatResponse,\n StreamChunk,\n ProviderType,\n RouterMetrics,\n ProviderError,\n Message,\n} from './types.js';\nimport { CostOptimalRouter } from './cost-optimal-router.js';\nimport { OpenRouterProvider } from './providers/openrouter.js';\nimport { AnthropicProvider } from './providers/anthropic.js';\nimport { ONNXLocalProvider } from './providers/onnx-local.js';\nimport { GeminiProvider } from './providers/gemini.js';\nimport { OllamaProvider } from './providers/ollama.js';\n\nexport class ModelRouter {\n private config: RouterConfig;\n private providers: Map<ProviderType, LLMProvider> = new Map();\n private metrics: RouterMetrics;\n // ADR-073: learned cost-optimal routing (opt-in via enableCostOptimalRouting).\n private costOptimalRouter?: CostOptimalRouter;\n private embedQuery?: (text: string) => Promise<number[]> | number[];\n // Bounded LRU embedding cache — embedding dominates the cost-optimal hot path\n // (≈ms) vs the µs-scale k-NN, and recurring prompts are common.\n private embedCache = new Map<string, number[]>();\n private static readonly EMBED_CACHE_MAX = 512;\n\n constructor(configPath?: string) {\n this.config = this.loadConfig(configPath);\n this.initializeProviders();\n this.metrics = this.initializeMetrics();\n }\n\n private loadConfig(configPath?: string): RouterConfig {\n const paths = [\n configPath,\n process.env.AGENTIC_FLOW_ROUTER_CONFIG,\n join(homedir(), '.agentic-flow', 'router.config.json'),\n join(process.cwd(), 'router.config.json'),\n join(process.cwd(), 'config', 'router.config.json'),\n join(process.cwd(), 'router.config.example.json'),\n ].filter(Boolean) as string[];\n\n for (const path of paths) {\n if (existsSync(path)) {\n const content = readFileSync(path, 'utf-8');\n const config = JSON.parse(content);\n\n // Substitute environment variables\n return this.substituteEnvVars(config);\n }\n }\n\n // If no config file found, create config from environment variables\n return this.createConfigFromEnv();\n }\n\n private createConfigFromEnv(): RouterConfig {\n // Create minimal config from environment variables\n const config: any = {\n version: '1.0',\n defaultProvider: (process.env.PROVIDER as any) || 'anthropic',\n routing: { mode: 'manual' as const },\n providers: {} as any,\n };\n\n // Add Anthropic if API key exists\n if (process.env.ANTHROPIC_API_KEY) {\n config.providers.anthropic = {\n apiKey: process.env.ANTHROPIC_API_KEY,\n baseUrl: process.env.ANTHROPIC_BASE_URL,\n };\n }\n\n // Add OpenRouter if API key exists\n if (process.env.OPENROUTER_API_KEY) {\n config.providers.openrouter = {\n apiKey: process.env.OPENROUTER_API_KEY,\n baseUrl: process.env.OPENROUTER_BASE_URL,\n };\n }\n\n // Add Gemini if API key exists\n if (process.env.GOOGLE_GEMINI_API_KEY) {\n config.providers.gemini = {\n apiKey: process.env.GOOGLE_GEMINI_API_KEY,\n };\n }\n\n // Add Ollama if API key OR explicit base URL is set (self-hosted needs no key)\n if (process.env.OLLAMA_API_KEY || process.env.OLLAMA_BASE_URL) {\n config.providers.ollama = {\n apiKey: process.env.OLLAMA_API_KEY,\n baseUrl: process.env.OLLAMA_BASE_URL,\n };\n }\n\n // ONNX is always available (no API key needed)\n config.providers.onnx = {\n modelPath: process.env.ONNX_MODEL_PATH,\n executionProviders: ['cpu'],\n };\n\n return config as RouterConfig;\n }\n\n private substituteEnvVars(obj: any): any {\n if (typeof obj === 'string') {\n // Replace ${VAR_NAME} with environment variable value\n return obj.replace(/\\$\\{([^}]+)\\}/g, (_, key) => {\n const [varName, defaultValue] = key.split(':-');\n return process.env[varName] || defaultValue || '';\n });\n }\n\n if (Array.isArray(obj)) {\n return obj.map((item) => this.substituteEnvVars(item));\n }\n\n if (obj && typeof obj === 'object') {\n const result: any = {};\n for (const [key, value] of Object.entries(obj)) {\n result[key] = this.substituteEnvVars(value);\n }\n return result;\n }\n\n return obj;\n }\n\n private initializeProviders(): void {\n const verbose = process.env.ROUTER_VERBOSE === 'true';\n\n // Initialize Anthropic\n if (this.config.providers.anthropic) {\n try {\n const provider = new AnthropicProvider(this.config.providers.anthropic);\n this.providers.set('anthropic', provider);\n if (verbose) console.log('✅ Anthropic provider initialized');\n } catch (error) {\n if (verbose) console.error('❌ Failed to initialize Anthropic:', error);\n }\n }\n\n // Initialize OpenRouter\n if (this.config.providers.openrouter) {\n try {\n const provider = new OpenRouterProvider(this.config.providers.openrouter);\n this.providers.set('openrouter', provider);\n if (verbose) console.log('✅ OpenRouter provider initialized');\n } catch (error) {\n if (verbose) console.error('❌ Failed to initialize OpenRouter:', error);\n }\n }\n\n // Initialize ONNX Local\n if (this.config.providers.onnx) {\n try {\n const provider = new ONNXLocalProvider({\n modelPath:\n this.config.providers.onnx.modelPath ||\n './models/phi-4/cpu_and_mobile/cpu-int4-rtn-block-32-acc-level-4/model.onnx',\n executionProviders: this.config.providers.onnx.executionProviders || ['cpu'],\n maxTokens: this.config.providers.onnx.maxTokens || 100,\n temperature: this.config.providers.onnx.temperature || 0.7,\n });\n this.providers.set('onnx', provider);\n if (verbose) console.log('✅ ONNX Local provider initialized');\n } catch (error) {\n if (verbose) console.error('❌ Failed to initialize ONNX:', error);\n }\n }\n\n // Initialize Gemini\n if (this.config.providers.gemini) {\n try {\n const provider = new GeminiProvider(this.config.providers.gemini);\n this.providers.set('gemini', provider);\n if (verbose) console.log('✅ Gemini provider initialized');\n } catch (error) {\n if (verbose) console.error('❌ Failed to initialize Gemini:', error);\n }\n }\n\n // Initialize Ollama (OpenAI-compat, self-hosted or ollama.com Cloud)\n if (this.config.providers.ollama) {\n try {\n const provider = new OllamaProvider(this.config.providers.ollama);\n this.providers.set('ollama', provider);\n if (verbose) console.log('✅ Ollama provider initialized');\n } catch (error) {\n if (verbose) console.error('❌ Failed to initialize Ollama:', error);\n }\n }\n }\n\n private initializeMetrics(): RouterMetrics {\n return {\n totalRequests: 0,\n totalCost: 0,\n totalTokens: { input: 0, output: 0 },\n providerBreakdown: {},\n agentBreakdown: {},\n };\n }\n\n async chat(params: ChatParams, agentType?: string): Promise<ChatResponse> {\n const startTime = Date.now();\n const provider = await this.selectProvider(params, agentType);\n\n try {\n const response = await provider.chat(params);\n\n // Update metrics\n this.updateMetrics(provider.name, response, Date.now() - startTime, agentType);\n\n // Add metadata\n response.metadata = {\n ...response.metadata,\n provider: provider.name,\n latency: Date.now() - startTime,\n };\n\n return response;\n } catch (error) {\n return this.handleProviderError(error as ProviderError, params, agentType);\n }\n }\n\n async *stream(params: ChatParams, agentType?: string): AsyncGenerator<StreamChunk> {\n const provider = await this.selectProvider(params, agentType);\n\n if (!provider.stream) {\n throw new Error(`Provider ${provider.name} does not support streaming`);\n }\n\n const iterator = provider.stream(params);\n for await (const chunk of iterator) {\n yield chunk;\n }\n }\n\n private async selectProvider(params: ChatParams, agentType?: string): Promise<LLMProvider> {\n // If provider is explicitly specified in params, use it\n if (params.provider) {\n const forcedProvider = this.providers.get(params.provider as ProviderType);\n if (forcedProvider) {\n return forcedProvider;\n }\n console.warn(\n `⚠️ Requested provider '${params.provider}' not available, falling back to routing logic`\n );\n }\n\n const routingMode = this.config.routing?.mode || 'manual';\n\n switch (routingMode) {\n case 'manual':\n return this.getDefaultProvider();\n\n case 'rule-based':\n return this.selectByRules(params, agentType);\n\n case 'cost-optimized':\n return this.selectByCost(params);\n\n case 'cost-optimal':\n return this.selectByCostOptimal(params);\n\n case 'performance-optimized':\n return this.selectByPerformance(params);\n\n default:\n return this.getDefaultProvider();\n }\n }\n\n private getDefaultProvider(): LLMProvider {\n const provider = this.providers.get(this.config.defaultProvider);\n if (!provider) {\n throw new Error(`Default provider ${this.config.defaultProvider} not initialized`);\n }\n return provider;\n }\n\n private selectByRules(params: ChatParams, agentType?: string): LLMProvider {\n const rules = this.config.routing?.rules || [];\n\n for (const rule of rules) {\n if (this.matchesRule(rule.condition, params, agentType)) {\n const provider = this.providers.get(rule.action.provider);\n if (provider) {\n console.log(`🎯 Routing via rule: ${rule.reason || 'matched condition'}`);\n return provider;\n }\n }\n }\n\n return this.getDefaultProvider();\n }\n\n private matchesRule(condition: any, params: ChatParams, agentType?: string): boolean {\n if (condition.agentType && agentType) {\n if (!condition.agentType.includes(agentType)) {\n return false;\n }\n }\n\n if (condition.requiresTools !== undefined) {\n if (condition.requiresTools && (!params.tools || params.tools.length === 0)) {\n return false;\n }\n }\n\n // TODO: Add more condition matching logic\n\n return true;\n }\n\n private selectByCost(params: ChatParams): LLMProvider {\n // For now, prefer cheaper providers\n // TODO: Implement actual cost calculation\n const providerOrder: ProviderType[] = ['openrouter', 'anthropic', 'openai'];\n\n for (const providerType of providerOrder) {\n const provider = this.providers.get(providerType);\n if (provider) {\n console.log(`💰 Cost-optimized routing: selected ${provider.name}`);\n return provider;\n }\n }\n\n return this.getDefaultProvider();\n }\n\n /**\n * Enable learned cost-optimal routing (ADR-073). Attaches a\n * {@link CostOptimalRouter} plus an embedder and switches the routing mode to\n * `'cost-optimal'`. Opt-in: until called, routing behaves exactly as before.\n */\n enableCostOptimalRouting(opts: {\n router: CostOptimalRouter;\n embed: (text: string) => Promise<number[]> | number[];\n }): void {\n this.costOptimalRouter = opts.router;\n this.embedQuery = opts.embed;\n this.config.routing = { ...(this.config.routing ?? { mode: 'cost-optimal' }), mode: 'cost-optimal' };\n }\n\n /**\n * Embed `text`, caching the result. Bounded LRU: a cache hit refreshes\n * recency; the oldest entry is evicted past {@link EMBED_CACHE_MAX}.\n */\n private async embedCached(text: string): Promise<number[]> {\n const hit = this.embedCache.get(text);\n if (hit) {\n this.embedCache.delete(text); // move to most-recently-used\n this.embedCache.set(text, hit);\n return hit;\n }\n const embedding = await this.embedQuery!(text);\n this.embedCache.set(text, embedding);\n if (this.embedCache.size > ModelRouter.EMBED_CACHE_MAX) {\n const oldest = this.embedCache.keys().next().value;\n if (oldest !== undefined) this.embedCache.delete(oldest);\n }\n return embedding;\n }\n\n /** Extract the most recent user-message text to embed for routing. */\n private lastUserText(messages: Message[]): string {\n for (let i = messages.length - 1; i >= 0; i--) {\n const m = messages[i];\n if (m.role !== 'user') continue;\n return typeof m.content === 'string'\n ? m.content\n : m.content.map((b) => b.text ?? '').join(' ');\n }\n // No user turn — fall back to all text content.\n return messages\n .map((m) => (typeof m.content === 'string' ? m.content : ''))\n .join(' ')\n .trim();\n }\n\n /**\n * Learned cost-optimal selection (ADR-073): embed the query, route to the\n * cheapest model predicted to clear the quality bar, steer params.model to it,\n * and return the matching provider. Falls back to the {@link selectByCost}\n * heuristic when the learned router/embedder isn't configured or errors —\n * routing never hard-fails on the cost-optimal path.\n */\n private async selectByCostOptimal(params: ChatParams): Promise<LLMProvider> {\n if (this.costOptimalRouter && this.embedQuery) {\n try {\n const text = this.lastUserText(params.messages);\n const embedding = await this.embedCached(text);\n const decision = this.costOptimalRouter.route(embedding);\n const provider = this.providers.get(decision.provider);\n if (provider) {\n params.model = decision.model; // steer model, not just provider\n console.log(\n `💡 Cost-optimal routing → ${decision.id} ` +\n `(q̂=${decision.predictedQuality.toFixed(2)}, $${decision.costPerMTok}/Mtok, metBar=${decision.metBar})`,\n );\n return provider;\n }\n console.warn(\n `⚠️ Cost-optimal pick '${decision.provider}' not initialized; falling back to heuristic`,\n );\n } catch (err) {\n console.warn('⚠️ Cost-optimal routing failed; falling back to heuristic', err);\n }\n }\n return this.selectByCost(params);\n }\n\n private selectByPerformance(params: ChatParams): LLMProvider {\n // For now, use metrics to select fastest provider\n let fastestProvider: LLMProvider | null = null;\n let lowestLatency = Infinity;\n\n for (const [providerType, provider] of this.providers) {\n const breakdown = this.metrics.providerBreakdown[providerType];\n if (breakdown && breakdown.avgLatency < lowestLatency) {\n lowestLatency = breakdown.avgLatency;\n fastestProvider = provider;\n }\n }\n\n if (fastestProvider) {\n console.log(`⚡ Performance-optimized routing: selected ${fastestProvider.name}`);\n return fastestProvider;\n }\n\n return this.getDefaultProvider();\n }\n\n private async handleProviderError(\n error: ProviderError,\n params: ChatParams,\n agentType?: string\n ): Promise<ChatResponse> {\n console.error(`❌ Provider error from ${error.provider}:`, error.message);\n\n // Try fallback chain\n const fallbackChain = this.config.fallbackChain || [];\n\n for (const providerType of fallbackChain) {\n if (providerType === error.provider) continue; // Skip failed provider\n\n const provider = this.providers.get(providerType);\n if (provider) {\n console.log(`🔄 Falling back to ${provider.name}`);\n try {\n return await provider.chat(params);\n } catch (fallbackError) {\n console.error(`❌ Fallback provider ${provider.name} also failed`);\n continue;\n }\n }\n }\n\n throw error; // No fallback succeeded\n }\n\n private updateMetrics(\n providerName: string,\n response: ChatResponse,\n latency: number,\n agentType?: string\n ): void {\n this.metrics.totalRequests++;\n\n if (response.usage) {\n this.metrics.totalTokens.input += response.usage.inputTokens;\n this.metrics.totalTokens.output += response.usage.outputTokens;\n }\n\n if (response.metadata?.cost) {\n this.metrics.totalCost += response.metadata.cost;\n }\n\n // Provider breakdown\n if (!this.metrics.providerBreakdown[providerName]) {\n this.metrics.providerBreakdown[providerName] = {\n requests: 0,\n cost: 0,\n avgLatency: 0,\n errors: 0,\n };\n }\n\n const breakdown = this.metrics.providerBreakdown[providerName];\n breakdown.requests++;\n breakdown.cost += response.metadata?.cost || 0;\n breakdown.avgLatency =\n (breakdown.avgLatency * (breakdown.requests - 1) + latency) / breakdown.requests;\n\n // Agent breakdown\n if (agentType) {\n if (!this.metrics.agentBreakdown) {\n this.metrics.agentBreakdown = {};\n }\n if (!this.metrics.agentBreakdown[agentType]) {\n this.metrics.agentBreakdown[agentType] = { requests: 0, cost: 0 };\n }\n this.metrics.agentBreakdown[agentType].requests++;\n this.metrics.agentBreakdown[agentType].cost += response.metadata?.cost || 0;\n }\n }\n\n getMetrics(): RouterMetrics {\n return { ...this.metrics };\n }\n\n getConfig(): RouterConfig {\n return { ...this.config };\n }\n\n getProviders(): Map<ProviderType, LLMProvider> {\n return new Map(this.providers);\n }\n}\n"]}
|
package/dist/router/types.d.ts
CHANGED
|
@@ -113,7 +113,7 @@ export interface RouterConfig {
|
|
|
113
113
|
cache?: CacheConfig;
|
|
114
114
|
}
|
|
115
115
|
export interface RoutingConfig {
|
|
116
|
-
mode: 'manual' | 'cost-optimized' | 'performance-optimized' | 'quality-optimized' | 'rule-based';
|
|
116
|
+
mode: 'manual' | 'cost-optimized' | 'cost-optimal' | 'performance-optimized' | 'quality-optimized' | 'rule-based';
|
|
117
117
|
rules?: RoutingRule[];
|
|
118
118
|
costOptimization?: {
|
|
119
119
|
enabled: boolean;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/router/types.ts"],"names":[],"mappings":"AAEA,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,YAAY,CAAC;IACnB,iBAAiB,EAAE,OAAO,CAAC;IAC3B,aAAa,EAAE,OAAO,CAAC;IACvB,WAAW,EAAE,OAAO,CAAC;IACrB,IAAI,CAAC,MAAM,EAAE,UAAU,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC;IAChD,MAAM,CAAC,CAAC,MAAM,EAAE,UAAU,GAAG,cAAc,CAAC,WAAW,CAAC,CAAC;IACzD,oBAAoB,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC;CACnD;AAED,MAAM,MAAM,YAAY,GAAG,WAAW,GAAG,QAAQ,GAAG,YAAY,GAAG,QAAQ,GAAG,SAAS,GAAG,MAAM,GAAG,QAAQ,GAAG,QAAQ,CAAC;AAEvH,MAAM,WAAW,UAAU;IACzB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,OAAO,EAAE,CAAC;IACpB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,IAAI,EAAE,CAAC;IACf,UAAU,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IACtE,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC/B,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,QAAQ,CAAC;IACtC,OAAO,EAAE,MAAM,GAAG,YAAY,EAAE,CAAC;CAClC;AAED,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,GAAG,UAAU,GAAG,aAAa,CAAC;IAC1C,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,KAAK,CAAC,EAAE,GAAG,CAAC;IACZ,OAAO,CAAC,EAAE,GAAG,CAAC;IACd,QAAQ,CAAC,EAAE,OAAO,CAAC;CACpB;AAED,MAAM,WAAW,IAAI;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE;QACZ,IAAI,EAAE,QAAQ,CAAC;QACf,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;QAChC,QAAQ,CAAC,EAAE,MAAM,EAAE,CAAC;KACrB,CAAC;CACH;AAED,MAAM,WAAW,YAAY;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,YAAY,EAAE,CAAC;IACxB,UAAU,CAAC,EAAE,UAAU,GAAG,YAAY,GAAG,UAAU,GAAG,eAAe,CAAC;IACtE,KAAK,CAAC,EAAE;QACN,WAAW,EAAE,MAAM,CAAC;QACpB,YAAY,EAAE,MAAM,CAAC;KACtB,CAAC;IACF,QAAQ,CAAC,EAAE;QACT,QAAQ,EAAE,MAAM,CAAC;QACjB,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;QAC9B,CAAC,GAAG,EAAE,MAAM,GAAG,GAAG,CAAC;KACpB,CAAC;CACH;AAED,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,qBAAqB,GAAG,qBAAqB,GAAG,oBAAoB,GAAG,eAAe,GAAG,eAAe,GAAG,cAAc,CAAC;IAChI,KAAK,CAAC,EAAE;QACN,IAAI,EAAE,YAAY,GAAG,kBAAkB,CAAC;QACxC,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,YAAY,CAAC,EAAE,MAAM,CAAC;KACvB,CAAC;IACF,aAAa,CAAC,EAAE,YAAY,CAAC;IAC7B,OAAO,CAAC,EAAE,OAAO,CAAC,YAAY,CAAC,CAAC;IAChC,KAAK,CAAC,EAAE;QACN,WAAW,EAAE,MAAM,CAAC;QACpB,YAAY,EAAE,MAAM,CAAC;KACtB,CAAC;CACH;AAED,MAAM,WAAW,cAAc;IAC7B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,MAAM,CAAC,EAAE;QACP,OAAO,EAAE,MAAM,CAAC;QAChB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,QAAQ,CAAC,EAAE,MAAM,CAAC;QAClB,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAAC;KACnC,CAAC;IACF,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE;QACV,iBAAiB,CAAC,EAAE,MAAM,CAAC;QAC3B,eAAe,CAAC,EAAE,MAAM,CAAC;KAC1B,CAAC;IACF,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAElC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,eAAe,CAAC,EAAE,OAAO,CAAC;CAC3B;AAED,MAAM,WAAW,YAAY;IAC3B,OAAO,EAAE,MAAM,CAAC;IAChB,eAAe,EAAE,YAAY,CAAC;IAC9B,aAAa,CAAC,EAAE,YAAY,EAAE,CAAC;IAC/B,SAAS,EAAE,MAAM,CAAC,YAAY,EAAE,cAAc,CAAC,CAAC;IAChD,OAAO,CAAC,EAAE,aAAa,CAAC;IACxB,WAAW,CAAC,EAAE,iBAAiB,CAAC;IAChC,UAAU,CAAC,EAAE,gBAAgB,CAAC;IAC9B,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB;AAED,MAAM,WAAW,aAAa;
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/router/types.ts"],"names":[],"mappings":"AAEA,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,YAAY,CAAC;IACnB,iBAAiB,EAAE,OAAO,CAAC;IAC3B,aAAa,EAAE,OAAO,CAAC;IACvB,WAAW,EAAE,OAAO,CAAC;IACrB,IAAI,CAAC,MAAM,EAAE,UAAU,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC;IAChD,MAAM,CAAC,CAAC,MAAM,EAAE,UAAU,GAAG,cAAc,CAAC,WAAW,CAAC,CAAC;IACzD,oBAAoB,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC;CACnD;AAED,MAAM,MAAM,YAAY,GAAG,WAAW,GAAG,QAAQ,GAAG,YAAY,GAAG,QAAQ,GAAG,SAAS,GAAG,MAAM,GAAG,QAAQ,GAAG,QAAQ,CAAC;AAEvH,MAAM,WAAW,UAAU;IACzB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,OAAO,EAAE,CAAC;IACpB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,IAAI,EAAE,CAAC;IACf,UAAU,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IACtE,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC/B,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,QAAQ,CAAC;IACtC,OAAO,EAAE,MAAM,GAAG,YAAY,EAAE,CAAC;CAClC;AAED,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,GAAG,UAAU,GAAG,aAAa,CAAC;IAC1C,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,KAAK,CAAC,EAAE,GAAG,CAAC;IACZ,OAAO,CAAC,EAAE,GAAG,CAAC;IACd,QAAQ,CAAC,EAAE,OAAO,CAAC;CACpB;AAED,MAAM,WAAW,IAAI;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE;QACZ,IAAI,EAAE,QAAQ,CAAC;QACf,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;QAChC,QAAQ,CAAC,EAAE,MAAM,EAAE,CAAC;KACrB,CAAC;CACH;AAED,MAAM,WAAW,YAAY;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,YAAY,EAAE,CAAC;IACxB,UAAU,CAAC,EAAE,UAAU,GAAG,YAAY,GAAG,UAAU,GAAG,eAAe,CAAC;IACtE,KAAK,CAAC,EAAE;QACN,WAAW,EAAE,MAAM,CAAC;QACpB,YAAY,EAAE,MAAM,CAAC;KACtB,CAAC;IACF,QAAQ,CAAC,EAAE;QACT,QAAQ,EAAE,MAAM,CAAC;QACjB,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;QAC9B,CAAC,GAAG,EAAE,MAAM,GAAG,GAAG,CAAC;KACpB,CAAC;CACH;AAED,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,qBAAqB,GAAG,qBAAqB,GAAG,oBAAoB,GAAG,eAAe,GAAG,eAAe,GAAG,cAAc,CAAC;IAChI,KAAK,CAAC,EAAE;QACN,IAAI,EAAE,YAAY,GAAG,kBAAkB,CAAC;QACxC,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,YAAY,CAAC,EAAE,MAAM,CAAC;KACvB,CAAC;IACF,aAAa,CAAC,EAAE,YAAY,CAAC;IAC7B,OAAO,CAAC,EAAE,OAAO,CAAC,YAAY,CAAC,CAAC;IAChC,KAAK,CAAC,EAAE;QACN,WAAW,EAAE,MAAM,CAAC;QACpB,YAAY,EAAE,MAAM,CAAC;KACtB,CAAC;CACH;AAED,MAAM,WAAW,cAAc;IAC7B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,MAAM,CAAC,EAAE;QACP,OAAO,EAAE,MAAM,CAAC;QAChB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,QAAQ,CAAC,EAAE,MAAM,CAAC;QAClB,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAAC;KACnC,CAAC;IACF,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE;QACV,iBAAiB,CAAC,EAAE,MAAM,CAAC;QAC3B,eAAe,CAAC,EAAE,MAAM,CAAC;KAC1B,CAAC;IACF,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAElC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,kBAAkB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,eAAe,CAAC,EAAE,OAAO,CAAC;CAC3B;AAED,MAAM,WAAW,YAAY;IAC3B,OAAO,EAAE,MAAM,CAAC;IAChB,eAAe,EAAE,YAAY,CAAC;IAC9B,aAAa,CAAC,EAAE,YAAY,EAAE,CAAC;IAC/B,SAAS,EAAE,MAAM,CAAC,YAAY,EAAE,cAAc,CAAC,CAAC;IAChD,OAAO,CAAC,EAAE,aAAa,CAAC;IACxB,WAAW,CAAC,EAAE,iBAAiB,CAAC;IAChC,UAAU,CAAC,EAAE,gBAAgB,CAAC;IAC9B,KAAK,CAAC,EAAE,WAAW,CAAC;CACrB;AAED,MAAM,WAAW,aAAa;IAG5B,IAAI,EAAE,QAAQ,GAAG,gBAAgB,GAAG,cAAc,GAAG,uBAAuB,GAAG,mBAAmB,GAAG,YAAY,CAAC;IAClH,KAAK,CAAC,EAAE,WAAW,EAAE,CAAC;IACtB,gBAAgB,CAAC,EAAE;QACjB,OAAO,EAAE,OAAO,CAAC;QACjB,iBAAiB,CAAC,EAAE,MAAM,CAAC;QAC3B,YAAY,CAAC,EAAE;YACb,KAAK,CAAC,EAAE,MAAM,CAAC;YACf,OAAO,CAAC,EAAE,MAAM,CAAC;SAClB,CAAC;QACF,aAAa,CAAC,EAAE,OAAO,CAAC;QACxB,aAAa,CAAC,EAAE,MAAM,CAAC;KACxB,CAAC;IACF,WAAW,CAAC,EAAE;QACZ,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,kBAAkB,CAAC,EAAE,MAAM,CAAC;QAC5B,cAAc,CAAC,EAAE;YACf,OAAO,EAAE,OAAO,CAAC;YACjB,SAAS,EAAE,MAAM,CAAC;YAClB,OAAO,EAAE,MAAM,CAAC;YAChB,YAAY,CAAC,EAAE,MAAM,CAAC;SACvB,CAAC;KACH,CAAC;CACH;AAED,MAAM,WAAW,WAAW;IAC1B,SAAS,EAAE;QACT,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;QACrB,aAAa,CAAC,EAAE,OAAO,CAAC;QACxB,UAAU,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;QACvC,OAAO,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;QACpC,SAAS,CAAC,EAAE,OAAO,CAAC;QACpB,iBAAiB,CAAC,EAAE,OAAO,CAAC;KAC7B,CAAC;IACF,MAAM,EAAE;QACN,QAAQ,EAAE,YAAY,CAAC;QACvB,KAAK,EAAE,MAAM,CAAC;QACd,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,SAAS,CAAC,EAAE,MAAM,CAAC;KACpB,CAAC;IACF,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,iBAAiB;IAChC,kBAAkB,EAAE,OAAO,CAAC;IAC5B,aAAa,EAAE,WAAW,GAAG,QAAQ,GAAG,aAAa,CAAC;IACtD,aAAa,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACvC,gBAAgB,CAAC,EAAE,eAAe,GAAG,UAAU,GAAG,MAAM,CAAC;CAC1D;AAED,MAAM,WAAW,gBAAgB;IAC/B,OAAO,EAAE,OAAO,CAAC;IACjB,QAAQ,EAAE,OAAO,GAAG,MAAM,GAAG,MAAM,GAAG,OAAO,CAAC;IAC9C,OAAO,CAAC,EAAE;QACR,SAAS,CAAC,EAAE,OAAO,CAAC;QACpB,YAAY,CAAC,EAAE,OAAO,CAAC;QACvB,WAAW,CAAC,EAAE,OAAO,CAAC;QACtB,WAAW,CAAC,EAAE,OAAO,CAAC;KACvB,CAAC;IACF,MAAM,CAAC,EAAE;QACP,aAAa,CAAC,EAAE,MAAM,CAAC;QACvB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,gBAAgB,CAAC,EAAE,MAAM,CAAC;KAC3B,CAAC;CACH;AAED,MAAM,WAAW,WAAW;IAC1B,OAAO,EAAE,OAAO,CAAC;IACjB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,KAAK,GAAG,MAAM,GAAG,KAAK,CAAC;IAClC,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE;QAAE,GAAG,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CAC9C;AAED,MAAM,WAAW,aAAa;IAC5B,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE;QACX,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;KAChB,CAAC;IACF,iBAAiB,EAAE,MAAM,CAAC,MAAM,EAAE;QAChC,QAAQ,EAAE,MAAM,CAAC;QACjB,IAAI,EAAE,MAAM,CAAC;QACb,UAAU,EAAE,MAAM,CAAC;QACnB,MAAM,EAAE,MAAM,CAAC;KAChB,CAAC,CAAC;IACH,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE;QAC9B,QAAQ,EAAE,MAAM,CAAC;QACjB,IAAI,EAAE,MAAM,CAAC;KACd,CAAC,CAAC;CACJ;AAED,MAAM,WAAW,aAAc,SAAQ,KAAK;IAC1C,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,OAAO,CAAC;CACpB"}
|