profclaw 2.0.0 → 2.3.0

package/dist/providers/core/models.js

@@ -32,8 +32,11 @@ export const MODEL_ALIASES = {
     local: { provider: 'ollama', model: 'llama3.2' },
     llama: { provider: 'ollama', model: 'llama3.2' },
     'deepseek-local': { provider: 'ollama', model: 'deepseek-r1:7b' },
-    qwen: { provider: 'ollama', model: 'qwen2.5:14b' },
+    qwen: { provider: 'ollama', model: 'qwen3:4b' },
+    'qwen-14b': { provider: 'ollama', model: 'qwen2.5:14b' },
     'mistral-local': { provider: 'ollama', model: 'mistral:7b' },
+    gemma4: { provider: 'ollama', model: 'gemma4:e4b' },
+    'gemma4-26b': { provider: 'ollama', model: 'gemma4:26b' },
     // xAI (Grok)
     grok: { provider: 'xai', model: 'grok-2' },
     'grok-3': { provider: 'xai', model: 'grok-3' },
@@ -378,6 +381,44 @@ export const MODEL_CATALOG = [
         costPer1MInput: 30.0,
         costPer1MOutput: 60.0,
     },
+    // Qwen 3 (local) — best small model for tool calling
+    {
+        id: 'qwen3:4b',
+        name: 'Qwen 3 4B (Local)',
+        provider: 'ollama',
+        contextWindow: 32768,
+        maxOutput: 8192,
+        supportsVision: false,
+        supportsStreaming: true,
+        supportsTools: true,
+        costPer1MInput: 0,
+        costPer1MOutput: 0,
+    },
+    // Gemma 4 (local) — supports native tool calling + multimodal
+    {
+        id: 'gemma4:e4b',
+        name: 'Gemma 4 E4B (Local)',
+        provider: 'ollama',
+        contextWindow: 128000,
+        maxOutput: 8192,
+        supportsVision: true,
+        supportsStreaming: true,
+        supportsTools: true,
+        costPer1MInput: 0,
+        costPer1MOutput: 0,
+    },
+    {
+        id: 'gemma4:26b',
+        name: 'Gemma 4 26B (Local)',
+        provider: 'ollama',
+        contextWindow: 128000,
+        maxOutput: 8192,
+        supportsVision: true,
+        supportsStreaming: true,
+        supportsTools: true,
+        costPer1MInput: 0,
+        costPer1MOutput: 0,
+    },
     // Ollama (local) - Most models don't reliably support native tool calling
     {
         id: 'llama3.2',
@@ -884,6 +925,31 @@ export const MODEL_CATALOG = [
         costPer1MInput: 0.40,
         costPer1MOutput: 1.20,
     },
+    // Cerebras (fast inference)
+    {
+        id: 'llama3.1-8b',
+        name: 'Llama 3.1 8B (Cerebras)',
+        provider: 'cerebras',
+        contextWindow: 8192,
+        maxOutput: 8192,
+        supportsVision: false,
+        supportsStreaming: true,
+        supportsTools: false,
+        costPer1MInput: 0.10,
+        costPer1MOutput: 0.10,
+    },
+    {
+        id: 'llama-3.3-70b',
+        name: 'Llama 3.3 70B (Cerebras)',
+        provider: 'cerebras',
+        contextWindow: 8192,
+        maxOutput: 8192,
+        supportsVision: false,
+        supportsStreaming: true,
+        supportsTools: true,
+        costPer1MInput: 0.60,
+        costPer1MOutput: 0.60,
+    },
     // IBM Watsonx
     {
         id: 'watsonx/ibm/granite-13b-chat-v2',

package/dist/providers/core/types.d.ts

@@ -58,10 +58,13 @@ export interface ChatResponse {
     };
     duration: number;
 }
+export type ToolTier = 'essential' | 'standard' | 'full';
+export type ModelCapabilityLevel = 'basic' | 'instruction' | 'reasoning';
 export interface ToolDefinition {
     name: string;
     description: string;
     parameters?: Record<string, unknown>;
+    tier?: ToolTier;
 }
 export interface ToolCall {
     id: string;

package/dist/providers/index.d.ts

@@ -4,4 +4,5 @@
  * Multi-provider AI chat using Vercel AI SDK.
  */
 export { aiProvider, chat, chatStream, ProviderType, MODEL_ALIASES, MODEL_CATALOG, type ProviderConfig, type ModelInfo, type ChatMessage, type ChatRequest, type ChatResponse, type NativeToolCall, type NativeToolResult, type NativeToolDefinition, type ToolExecutionResult, } from './ai-sdk.js';
+export { routeQuery, classifyComplexity, selectModel, getRoutingStats, recordRoutingDecision, configureSmartRouter, isSmartRouterEnabled, getCostComparison, getCostOptimizationAdvice, resolveDefaultModel, type ComplexityTier, type ComplexityResult, type RoutingDecision, type SmartRouterConfig, type ProviderRecommendation, } from './smart-router.js';
 //# sourceMappingURL=index.d.ts.map

package/dist/providers/index.js

@@ -4,4 +4,5 @@
  * Multi-provider AI chat using Vercel AI SDK.
  */
 export { aiProvider, chat, chatStream, ProviderType, MODEL_ALIASES, MODEL_CATALOG, } from './ai-sdk.js';
+export { routeQuery, classifyComplexity, selectModel, getRoutingStats, recordRoutingDecision, configureSmartRouter, isSmartRouterEnabled, getCostComparison, getCostOptimizationAdvice, resolveDefaultModel, } from './smart-router.js';
 //# sourceMappingURL=index.js.map

package/dist/providers/smart-router.d.ts

@@ -0,0 +1,118 @@
+/**
+ * Smart Model Router
+ *
+ * Classifies query complexity and routes to the cheapest model
+ * that can handle it. This is the engine behind profClaw's "10x cheaper" claim.
+ *
+ * Complexity tiers:
+ *   trivial  -> Haiku/Groq/Gemini Flash ($0.001-0.005/conv)
+ *   standard -> Sonnet/GPT-4o/Gemini Pro ($0.02-0.05/conv)
+ *   complex  -> Opus/GPT-4.5/Gemini Ultra ($0.05-0.15/conv)
+ *
+ * The router saves users 60-90% vs always using a top-tier model.
+ */
+import type { ModelInfo, ProviderType } from './core/types.js';
+export type ComplexityTier = 'trivial' | 'standard' | 'complex';
+export interface ComplexitySignal {
+    name: string;
+    weight: number;
+    matched: boolean;
+    detail?: string;
+}
+export interface ComplexityResult {
+    tier: ComplexityTier;
+    score: number;
+    confidence: number;
+    signals: ComplexitySignal[];
+    reasoning: string;
+}
+export interface ProviderRecommendation {
+    provider: ProviderType;
+    model: string;
+    reason: string;
+    estimatedSavings: string;
+    signupUrl: string;
+    freeTrialAvailable: boolean;
+}
+export interface RoutingDecision {
+    complexity: ComplexityResult;
+    selectedModel: ModelInfo;
+    alternativeModels: ModelInfo[];
+    estimatedCost: number;
+    savedVsDefault: number;
+    savingsPercent: number;
+    missingProviderHint?: ProviderRecommendation;
+}
+export interface SmartRouterConfig {
+    enabled: boolean;
+    defaultModel: string;
+    preferredProviders: ProviderType[];
+    maxCostPerRequest: number;
+    enableLocalFallback: boolean;
+    complexityThresholds: {
+        trivialMax: number;
+        standardMax: number;
+    };
+}
+/**
+ * Resolve the default model from whatever providers are available.
+ * Picks the best mid-range model from configured providers.
+ * Truly provider-agnostic - works with any combination.
+ */
+export declare function resolveDefaultModel(availableProviders: Set<ProviderType>): string;
+interface RoutingContext {
+    conversationLength: number;
+    hasToolUse: boolean;
+    hasCodeContext: boolean;
+    hasImages: boolean;
+    previousModel?: string;
+    sessionCostSoFar: number;
+}
+/**
+ * Classify the complexity of a user prompt.
+ */
+export declare function classifyComplexity(prompt: string, context?: Partial<RoutingContext>): ComplexityResult;
+/**
+ * Select the best model for a given complexity tier.
+ * Fully dynamic - queries the MODEL_CATALOG based on available providers.
+ * No hardcoded model IDs or provider assumptions.
+ */
+export declare function selectModel(complexity: ComplexityResult, availableProviders: Set<ProviderType>, overrideModel?: string): RoutingDecision;
+/**
+ * One-shot: classify + select in a single call.
+ */
+export declare function routeQuery(prompt: string, availableProviders: Set<ProviderType>, context?: Partial<RoutingContext>, overrideModel?: string): RoutingDecision;
+/**
+ * Analyze usage patterns and generate cost optimization recommendations.
+ * Call this from the dashboard or CLI doctor command.
+ */
+export declare function getCostOptimizationAdvice(availableProviders: Set<ProviderType>): Array<{
+    priority: 'high' | 'medium' | 'low';
+    action: string;
+    reason: string;
+    estimatedMonthlySavings: string;
+    envVar: string;
+    signupUrl: string;
+}>;
+/**
+ * Get a human-readable cost comparison for display.
+ */
+export declare function getCostComparison(decision: RoutingDecision): string;
+/**
+ * Get cumulative savings stats for display in dashboard.
+ */
+export declare function getRoutingStats(): {
+    totalRouted: number;
+    totalSaved: number;
+    byTier: Record<ComplexityTier, number>;
+    avgSavingsPercent: number;
+};
+/**
+ * Record a routing decision for stats.
+ */
+export declare function recordRoutingDecision(decision: RoutingDecision): void;
+export declare function configureSmartRouter(newConfig: Partial<SmartRouterConfig>): void;
+export declare function getSmartRouterConfig(): SmartRouterConfig;
+export declare function isSmartRouterEnabled(): boolean;
+export {};
+//# sourceMappingURL=smart-router.d.ts.map