rax-flow-providers 0.2.1 → 2.0.1

package/dist/index.d.ts

@@ -8,5 +8,6 @@ export * from "./cohere-adapter.js";
 export * from "./rest-adapter.js";
 export * from "./host-bridge-adapter.js";
 export * from "./strategy.js";
+export * from "./model-strategy.js";
 export * from "./utils.js";
 //# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,mBAAmB,CAAC;AAClC,cAAc,qBAAqB,CAAC;AACpC,cAAc,qBAAqB,CAAC;AACpC,cAAc,qBAAqB,CAAC;AACpC,cAAc,mBAAmB,CAAC;AAClC,cAAc,sBAAsB,CAAC;AACrC,cAAc,qBAAqB,CAAC;AACpC,cAAc,mBAAmB,CAAC;AAClC,cAAc,0BAA0B,CAAC;AACzC,cAAc,eAAe,CAAC;AAC9B,cAAc,YAAY,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,mBAAmB,CAAC;AAClC,cAAc,qBAAqB,CAAC;AACpC,cAAc,qBAAqB,CAAC;AACpC,cAAc,qBAAqB,CAAC;AACpC,cAAc,mBAAmB,CAAC;AAClC,cAAc,sBAAsB,CAAC;AACrC,cAAc,qBAAqB,CAAC;AACpC,cAAc,mBAAmB,CAAC;AAClC,cAAc,0BAA0B,CAAC;AACzC,cAAc,eAAe,CAAC;AAC9B,cAAc,qBAAqB,CAAC;AACpC,cAAc,YAAY,CAAC"}

package/dist/index.js

@@ -8,5 +8,6 @@ export * from "./cohere-adapter.js";
 export * from "./rest-adapter.js";
 export * from "./host-bridge-adapter.js";
 export * from "./strategy.js";
+export * from "./model-strategy.js";
 export * from "./utils.js";
 //# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,mBAAmB,CAAC;AAClC,cAAc,qBAAqB,CAAC;AACpC,cAAc,qBAAqB,CAAC;AACpC,cAAc,qBAAqB,CAAC;AACpC,cAAc,mBAAmB,CAAC;AAClC,cAAc,sBAAsB,CAAC;AACrC,cAAc,qBAAqB,CAAC;AACpC,cAAc,mBAAmB,CAAC;AAClC,cAAc,0BAA0B,CAAC;AACzC,cAAc,eAAe,CAAC;AAC9B,cAAc,YAAY,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,mBAAmB,CAAC;AAClC,cAAc,qBAAqB,CAAC;AACpC,cAAc,qBAAqB,CAAC;AACpC,cAAc,qBAAqB,CAAC;AACpC,cAAc,mBAAmB,CAAC;AAClC,cAAc,sBAAsB,CAAC;AACrC,cAAc,qBAAqB,CAAC;AACpC,cAAc,mBAAmB,CAAC;AAClC,cAAc,0BAA0B,CAAC;AACzC,cAAc,eAAe,CAAC;AAC9B,cAAc,qBAAqB,CAAC;AACpC,cAAc,YAAY,CAAC"}

package/dist/model-strategy.d.ts

@@ -0,0 +1,83 @@
+import { IModelProvider, ModelResponse, ProviderCallOptions, Intent } from "rax-flow-core";
+export type ModelTier = "economy" | "standard" | "premium" | "reasoning";
+export type SelectionCriteria = "cost" | "latency" | "quality" | "balanced";
+export interface ModelProfile {
+    provider: string;
+    model: string;
+    tier: ModelTier;
+    costPer1kTokens: {
+        input: number;
+        output: number;
+    };
+    avgLatencyMs: number;
+    maxTokens: number;
+    supportsStructured: boolean;
+    supportsVision: boolean;
+    supportsStreaming: boolean;
+    qualityScore: number;
+    capabilities: string[];
+}
+export interface SelectionContext {
+    prompt: string;
+    intent?: Intent;
+    estimatedTokens?: number;
+    maxCostUsd?: number;
+    maxLatencyMs?: number;
+    requiresStructured?: boolean;
+    requiresVision?: boolean;
+    requiresStreaming?: boolean;
+    preferredTier?: ModelTier;
+    criteria?: SelectionCriteria;
+    fallbackChain?: string[];
+}
+export interface SelectionResult {
+    provider: IModelProvider;
+    profile: ModelProfile;
+    fallbackChain: ModelProfile[];
+    estimatedCost: number;
+    estimatedLatency: number;
+    reasoning: string;
+}
+export interface ProviderMetrics {
+    provider: string;
+    totalCalls: number;
+    successRate: number;
+    avgLatencyMs: number;
+    totalCostUsd: number;
+    lastError?: string;
+    lastErrorTime?: number;
+}
+export declare class ModelStrategyEngine {
+    private profiles;
+    private providers;
+    private metrics;
+    private selectionHistory;
+    constructor(providerInstances: Record<string, IModelProvider>, customProfiles?: ModelProfile[]);
+    select(context: SelectionContext): SelectionResult;
+    callWithStrategy<T>(context: SelectionContext, prompt: string, schema?: object, options?: ProviderCallOptions): Promise<ModelResponse<T>>;
+    private getCandidates;
+    private meetsRequirements;
+    private scoreCandidate;
+    private scoreByCost;
+    private scoreByLatency;
+    private calculateCost;
+    private estimateTokens;
+    private generateReasoning;
+    private getFallbackProvider;
+    private getProviderName;
+    private recordSuccess;
+    private recordFailure;
+    getMetrics(): Record<string, ProviderMetrics>;
+    getSelectionHistory(limit?: number): typeof this.selectionHistory;
+    getProfiles(): ModelProfile[];
+    addProfile(profile: ModelProfile): void;
+    getIntentTier(intent: Intent): ModelTier;
+    recommendForIntent(intent: Intent): ModelProfile[];
+}
+export declare function smartFallbackCall(providers: IModelProvider[], prompt: string, options?: {
+    maxRetriesPerProvider?: number;
+    retryDelayMs?: number;
+    callOptions?: ProviderCallOptions;
+    onFallback?: (from: string, to: string, reason: string) => void;
+}): Promise<ModelResponse<string>>;
+//# sourceMappingURL=model-strategy.d.ts.map

package/dist/model-strategy.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"model-strategy.d.ts","sourceRoot":"","sources":["../src/model-strategy.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,aAAa,EAAE,mBAAmB,EAAE,MAAM,EAAE,MAAM,eAAe,CAAC;AAG3F,MAAM,MAAM,SAAS,GAAG,SAAS,GAAG,UAAU,GAAG,SAAS,GAAG,WAAW,CAAC;AACzE,MAAM,MAAM,iBAAiB,GAAG,MAAM,GAAG,SAAS,GAAG,SAAS,GAAG,UAAU,CAAC;AAE5E,MAAM,WAAW,YAAY;IAC3B,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,SAAS,CAAC;IAChB,eAAe,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC;IACnD,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;IAClB,kBAAkB,EAAE,OAAO,CAAC;IAC5B,cAAc,EAAE,OAAO,CAAC;IACxB,iBAAiB,EAAE,OAAO,CAAC;IAC3B,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,gBAAgB;IAC/B,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAC7B,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,iBAAiB,CAAC,EAAE,OAAO,CAAC;IAC5B,aAAa,CAAC,EAAE,SAAS,CAAC;IAC1B,QAAQ,CAAC,EAAE,iBAAiB,CAAC;IAC7B,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,eAAe;IAC9B,QAAQ,EAAE,cAAc,CAAC;IACzB,OAAO,EAAE,YAAY,CAAC;IACtB,aAAa,EAAE,YAAY,EAAE,CAAC;IAC9B,aAAa,EAAE,MAAM,CAAC;IACtB,gBAAgB,EAAE,MAAM,CAAC;IACzB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,eAAe;IAC9B,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAyCD,qBAAa,mBAAmB;IAC9B,OAAO,CAAC,QAAQ,CAAwC;IACxD,OAAO,CAAC,SAAS,CAA0C;IAC3D,OAAO,CAAC,OAAO,CAA2C;IAC1D,OAAO,CAAC,gBAAgB,CAAsH;gBAG5I,iBAAiB,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,EACjD,cAAc,CAAC,EAAE,YAAY,EAAE;IAoBjC,MAAM,CAAC,OAAO,EAAE,gBAAgB,GAAG,eAAe;IAsC5C,gBAAgB,CAAC,CAAC,EACtB,OAAO,EAAE,gBAAgB,EACzB,MAAM,EAAE,MAAM,EACd,MAAM,CAAC,EAAE,MAAM,EACf,OAAO,CAAC,EAAE,mBAAmB,GAC5B,OAAO,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC;IA6C5B,OAAO,CAAC,aAAa;IAerB,OAAO,CAAC,iBAAiB;IAuBzB,OAAO,CAAC,cAAc;IAqCtB,OAAO,CAAC,WAAW;IAMnB,OAAO,CAAC,cAAc;IAItB,OAAO,CAAC,aAAa;IAMrB,OAAO,CAAC,cAAc;IAItB,OAAO,CAAC,iBAAiB;IAOzB,OAAO,CAAC,mBAAmB;IAa3B,OAAO,CAAC,eAAe;IAOvB,OAAO,CAAC,aAAa;IAmBrB,OAAO,CAAC,aAAa;IAmBrB,UAAU,IAAI,MAAM,CAAC,MAAM,EAAE,eAAe,CAAC;IAI7C,mBAAmB,CAAC,KAAK,GAAE,MAAY,GAAG,OAAO,IAAI,CAAC,gBAAgB;IAItE,WAAW,IAAI,YAAY,EAAE;IAI7B,UAAU,CAAC,OAAO,EAAE,YAAY,GAAG,IAAI;IAKvC,aAAa,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS;IAIxC,kBAAkB,CAAC,MAAM,EAAE,MAAM,GAAG,YAAY,EAAE;CAMnD;AAED,wBAAsB,iBAAiB,CACrC,SAAS,EAAE,cAAc,EAAE,EAC3B,MAAM,EAAE,MAAM,EACd,OAAO,GAAE;IACP,qBAAqB,CAAC,EAAE,MAAM,CAAC;IAC/B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,mBAAmB,CAAC;IAClC,UAAU,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,KAAK,IAAI,CAAC;CAC5D,GACL,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC,CA2ChC"}

package/dist/model-strategy.js

@@ -0,0 +1,330 @@
+import { RaxProviderError } from "./error-mapper.js";
+const MODEL_CATALOG = [
+    // OpenAI models
+    { provider: "openai", model: "gpt-4o-mini", tier: "economy", costPer1kTokens: { input: 0.00015, output: 0.0006 }, avgLatencyMs: 400, maxTokens: 128000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.75, capabilities: ["chat", "code", "analysis"] },
+    { provider: "openai", model: "gpt-4.1-mini", tier: "economy", costPer1kTokens: { input: 0.0004, output: 0.0016 }, avgLatencyMs: 450, maxTokens: 128000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.78, capabilities: ["chat", "code", "analysis"] },
+    { provider: "openai", model: "gpt-4o", tier: "standard", costPer1kTokens: { input: 0.0025, output: 0.01 }, avgLatencyMs: 600, maxTokens: 128000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.88, capabilities: ["chat", "code", "analysis", "reasoning"] },
+    { provider: "openai", model: "gpt-4.1", tier: "standard", costPer1kTokens: { input: 0.002, output: 0.008 }, avgLatencyMs: 550, maxTokens: 128000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.90, capabilities: ["chat", "code", "analysis", "reasoning"] },
+    { provider: "openai", model: "o1", tier: "reasoning", costPer1kTokens: { input: 0.015, output: 0.06 }, avgLatencyMs: 5000, maxTokens: 200000, supportsStructured: false, supportsVision: false, supportsStreaming: false, qualityScore: 0.95, capabilities: ["reasoning", "math", "science"] },
+    { provider: "openai", model: "o3-mini", tier: "premium", costPer1kTokens: { input: 0.0011, output: 0.0044 }, avgLatencyMs: 2000, maxTokens: 200000, supportsStructured: true, supportsVision: false, supportsStreaming: true, qualityScore: 0.92, capabilities: ["reasoning", "code", "analysis"] },
+    // Claude models
+    { provider: "claude", model: "claude-3-5-haiku-latest", tier: "economy", costPer1kTokens: { input: 0.0008, output: 0.004 }, avgLatencyMs: 300, maxTokens: 200000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.80, capabilities: ["chat", "code", "analysis"] },
+    { provider: "claude", model: "claude-3-5-sonnet-latest", tier: "standard", costPer1kTokens: { input: 0.003, output: 0.015 }, avgLatencyMs: 500, maxTokens: 200000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.91, capabilities: ["chat", "code", "analysis", "reasoning"] },
+    { provider: "claude", model: "claude-3-opus-latest", tier: "premium", costPer1kTokens: { input: 0.015, output: 0.075 }, avgLatencyMs: 1200, maxTokens: 200000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.94, capabilities: ["chat", "code", "analysis", "reasoning", "creative"] },
+    // Gemini models
+    { provider: "gemini", model: "gemini-2.0-flash", tier: "economy", costPer1kTokens: { input: 0.0001, output: 0.0004 }, avgLatencyMs: 250, maxTokens: 1000000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.72, capabilities: ["chat", "code", "multimodal"] },
+    { provider: "gemini", model: "gemini-1.5-pro", tier: "standard", costPer1kTokens: { input: 0.00125, output: 0.005 }, avgLatencyMs: 600, maxTokens: 2000000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.85, capabilities: ["chat", "code", "analysis", "multimodal"] },
+    // Groq models (fast inference)
+    { provider: "groq", model: "llama-3.3-70b-versatile", tier: "economy", costPer1kTokens: { input: 0.00059, output: 0.00079 }, avgLatencyMs: 80, maxTokens: 128000, supportsStructured: true, supportsVision: false, supportsStreaming: true, qualityScore: 0.78, capabilities: ["chat", "code"] },
+    { provider: "groq", model: "llama-3.1-8b-instant", tier: "economy", costPer1kTokens: { input: 0.00002, output: 0.00002 }, avgLatencyMs: 40, maxTokens: 128000, supportsStructured: true, supportsVision: false, supportsStreaming: true, qualityScore: 0.65, capabilities: ["chat"] },
+    // Host bridge
+    { provider: "host", model: "host-managed", tier: "standard", costPer1kTokens: { input: 0, output: 0 }, avgLatencyMs: 1000, maxTokens: 1000000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.85, capabilities: ["chat", "code", "analysis"] }
+];
+const INTENT_TIER_MAP = {
+    brainstorm: "economy",
+    spec: "standard",
+    architecture: "premium",
+    plan: "standard",
+    generate_code: "standard",
+    test: "economy",
+    fix: "standard",
+    optimize: "premium",
+    document: "economy",
+    benchmark: "economy"
+};
+export class ModelStrategyEngine {
+    profiles = new Map();
+    providers = new Map();
+    metrics = new Map();
+    selectionHistory = [];
+    constructor(providerInstances, customProfiles) {
+        const catalog = customProfiles ?? MODEL_CATALOG;
+        for (const profile of catalog) {
+            const key = `${profile.provider}:${profile.model}`;
+            this.profiles.set(key, profile);
+        }
+        for (const [name, provider] of Object.entries(providerInstances)) {
+            this.providers.set(name, provider);
+            this.metrics.set(name, {
+                provider: name,
+                totalCalls: 0,
+                successRate: 1,
+                avgLatencyMs: 0,
+                totalCostUsd: 0
+            });
+        }
+    }
+    select(context) {
+        const candidates = this.getCandidates(context);
+        if (candidates.length === 0) {
+            const fallback = this.getFallbackProvider();
+            return {
+                provider: fallback.provider,
+                profile: fallback.profile,
+                fallbackChain: [],
+                estimatedCost: 0,
+                estimatedLatency: 1000,
+                reasoning: "No suitable model found, using fallback"
+            };
+        }
+        const scored = candidates.map(c => ({
+            ...c,
+            score: this.scoreCandidate(c, context)
+        }));
+        scored.sort((a, b) => b.score - a.score);
+        const selected = scored[0];
+        const fallbackChain = scored.slice(1, 4);
+        const estimatedTokens = context.estimatedTokens ?? this.estimateTokens(context.prompt);
+        const estimatedCost = this.calculateCost(selected.profile, estimatedTokens);
+        return {
+            provider: selected.provider,
+            profile: selected.profile,
+            fallbackChain: fallbackChain.map(f => f.profile),
+            estimatedCost,
+            estimatedLatency: selected.profile.avgLatencyMs,
+            reasoning: this.generateReasoning(selected, context)
+        };
+    }
+    async callWithStrategy(context, prompt, schema, options) {
+        const selection = this.select(context);
+        const fallbackProviders = selection.fallbackChain
+            .map(p => this.providers.get(p.provider))
+            .filter((p) => p !== undefined);
+        const startTime = Date.now();
+        let lastError;
+        // Try primary
+        try {
+            const result = schema
+                ? await selection.provider.callStructured(prompt, schema, options)
+                : await selection.provider.callModel(prompt, options);
+            this.recordSuccess(selection.profile, Date.now() - startTime, result.costUsd ?? 0);
+            return result;
+        }
+        catch (err) {
+            lastError = err;
+            this.recordFailure(selection.profile, err);
+        }
+        // Try fallbacks
+        for (const fallbackProvider of fallbackProviders) {
+            const fallbackProfile = selection.fallbackChain.find(p => p.provider === this.getProviderName(fallbackProvider));
+            if (!fallbackProfile)
+                continue;
+            try {
+                const result = schema
+                    ? await fallbackProvider.callStructured(prompt, schema, options)
+                    : await fallbackProvider.callModel(prompt, options);
+                this.recordSuccess(fallbackProfile, Date.now() - startTime, result.costUsd ?? 0);
+                return result;
+            }
+            catch (err) {
+                lastError = err;
+                if (fallbackProfile) {
+                    this.recordFailure(fallbackProfile, err);
+                }
+            }
+        }
+        throw new Error(`All providers failed. Last error: ${lastError instanceof Error ? lastError.message : String(lastError)}`);
+    }
+    getCandidates(context) {
+        const candidates = [];
+        for (const [key, profile] of this.profiles) {
+            if (!this.meetsRequirements(profile, context))
+                continue;
+            const provider = this.providers.get(profile.provider);
+            if (!provider)
+                continue;
+            candidates.push({ provider, profile });
+        }
+        return candidates;
+    }
+    meetsRequirements(profile, context) {
+        if (context.requiresStructured && !profile.supportsStructured)
+            return false;
+        if (context.requiresVision && !profile.supportsVision)
+            return false;
+        if (context.requiresStreaming && !profile.supportsStreaming)
+            return false;
+        if (context.maxCostUsd !== undefined) {
+            const estimatedTokens = context.estimatedTokens ?? this.estimateTokens(context.prompt);
+            const estimatedCost = this.calculateCost(profile, estimatedTokens);
+            if (estimatedCost > context.maxCostUsd)
+                return false;
+        }
+        if (context.maxLatencyMs !== undefined && profile.avgLatencyMs > context.maxLatencyMs)
+            return false;
+        if (context.preferredTier && profile.tier !== context.preferredTier) {
+            const tierPriority = ["economy", "standard", "premium", "reasoning"];
+            const preferredIdx = tierPriority.indexOf(context.preferredTier);
+            const profileIdx = tierPriority.indexOf(profile.tier);
+            if (profileIdx > preferredIdx + 1)
+                return false;
+        }
+        return true;
+    }
+    scoreCandidate(candidate, context) {
+        const criteria = context.criteria ?? "balanced";
+        let score = 0;
+        const metrics = this.metrics.get(candidate.profile.provider);
+        const successRate = metrics?.successRate ?? 1;
+        switch (criteria) {
+            case "cost":
+                score = this.scoreByCost(candidate.profile, context);
+                break;
+            case "latency":
+                score = this.scoreByLatency(candidate.profile);
+                break;
+            case "quality":
+                score = candidate.profile.qualityScore * 100;
+                break;
+            case "balanced":
+            default:
+                score = (this.scoreByCost(candidate.profile, context) * 0.3 +
+                    this.scoreByLatency(candidate.profile) * 0.3 +
+                    candidate.profile.qualityScore * 30 +
+                    successRate * 10);
+        }
+        if (context.intent) {
+            const preferredTier = INTENT_TIER_MAP[context.intent];
+            if (preferredTier && candidate.profile.tier === preferredTier) {
+                score += 10;
+            }
+        }
+        return score;
+    }
+    scoreByCost(profile, context) {
+        const tokens = context.estimatedTokens ?? this.estimateTokens(context.prompt);
+        const cost = this.calculateCost(profile, tokens);
+        return Math.max(0, 100 - cost * 1000);
+    }
+    scoreByLatency(profile) {
+        return Math.max(0, 100 - profile.avgLatencyMs / 50);
+    }
+    calculateCost(profile, tokens) {
+        const inputTokens = Math.ceil(tokens * 0.7);
+        const outputTokens = Math.ceil(tokens * 0.3);
+        return (inputTokens * profile.costPer1kTokens.input + outputTokens * profile.costPer1kTokens.output) / 1000;
+    }
+    estimateTokens(prompt) {
+        return Math.ceil(prompt.split(/\s+/).length * 1.3);
+    }
+    generateReasoning(selected, context) {
+        const criteria = context.criteria ?? "balanced";
+        return `Selected ${selected.profile.provider}/${selected.profile.model} (tier: ${selected.profile.tier}) ` +
+            `based on ${criteria} criteria with score ${selected.score.toFixed(1)}. ` +
+            `Est. latency: ${selected.profile.avgLatencyMs}ms, quality: ${(selected.profile.qualityScore * 100).toFixed(0)}%`;
+    }
+    getFallbackProvider() {
+        const hostProvider = this.providers.get("host");
+        const hostProfile = Array.from(this.profiles.values()).find(p => p.provider === "host");
+        if (hostProvider && hostProfile) {
+            return { provider: hostProvider, profile: hostProfile };
+        }
+        const firstProvider = this.providers.values().next().value;
+        const firstProfile = this.profiles.values().next().value;
+        return { provider: firstProvider, profile: firstProfile };
+    }
+    getProviderName(provider) {
+        for (const [name, p] of this.providers) {
+            if (p === provider)
+                return name;
+        }
+        return "unknown";
+    }
+    recordSuccess(profile, latency, cost) {
+        const metrics = this.metrics.get(profile.provider);
+        if (!metrics)
+            return;
+        metrics.totalCalls++;
+        metrics.successRate = (metrics.successRate * (metrics.totalCalls - 1) + 1) / metrics.totalCalls;
+        metrics.avgLatencyMs = (metrics.avgLatencyMs * (metrics.totalCalls - 1) + latency) / metrics.totalCalls;
+        metrics.totalCostUsd += cost;
+        this.selectionHistory.push({
+            timestamp: Date.now(),
+            provider: profile.provider,
+            model: profile.model,
+            success: true,
+            latency,
+            cost
+        });
+    }
+    recordFailure(profile, error) {
+        const metrics = this.metrics.get(profile.provider);
+        if (!metrics)
+            return;
+        metrics.totalCalls++;
+        metrics.successRate = (metrics.successRate * (metrics.totalCalls - 1)) / metrics.totalCalls;
+        metrics.lastError = error instanceof Error ? error.message : String(error);
+        metrics.lastErrorTime = Date.now();
+        this.selectionHistory.push({
+            timestamp: Date.now(),
+            provider: profile.provider,
+            model: profile.model,
+            success: false,
+            latency: 0,
+            cost: 0
+        });
+    }
+    getMetrics() {
+        return Object.fromEntries(this.metrics);
+    }
+    getSelectionHistory(limit = 100) {
+        return this.selectionHistory.slice(-limit);
+    }
+    getProfiles() {
+        return Array.from(this.profiles.values());
+    }
+    addProfile(profile) {
+        const key = `${profile.provider}:${profile.model}`;
+        this.profiles.set(key, profile);
+    }
+    getIntentTier(intent) {
+        return INTENT_TIER_MAP[intent] ?? "standard";
+    }
+    recommendForIntent(intent) {
+        const tier = this.getIntentTier(intent);
+        return Array.from(this.profiles.values())
+            .filter(p => p.tier === tier)
+            .sort((a, b) => b.qualityScore - a.qualityScore);
+    }
+}
+export async function smartFallbackCall(providers, prompt, options = {}) {
+    const { maxRetriesPerProvider = 1, retryDelayMs = 500, callOptions, onFallback } = options;
+    const errors = [];
+    let lastProvider = "";
+    for (const provider of providers) {
+        const providerName = provider.constructor?.name ?? "unknown";
+        const healthy = await provider.healthCheck().catch(() => false);
+        if (!healthy) {
+            if (onFallback && lastProvider) {
+                onFallback(lastProvider, providerName, "health_check_failed");
+            }
+            continue;
+        }
+        let attempt = 0;
+        while (attempt <= maxRetriesPerProvider) {
+            try {
+                return await provider.callModel(prompt, callOptions);
+            }
+            catch (err) {
+                const errorMsg = err instanceof Error ? err.message : String(err);
+                errors.push(`${providerName}: ${errorMsg}`);
+                if (err instanceof RaxProviderError && err.isRetryable && attempt < maxRetriesPerProvider) {
+                    await new Promise(r => setTimeout(r, retryDelayMs * (attempt + 1)));
+                    attempt++;
+                    continue;
+                }
+                if (onFallback) {
+                    const nextProvider = providers[providers.indexOf(provider) + 1];
+                    if (nextProvider) {
+                        const nextName = nextProvider.constructor?.name ?? "unknown";
+                        onFallback(providerName, nextName, errorMsg);
+                    }
+                }
+                break;
+            }
+        }
+        lastProvider = providerName;
+    }
+    throw new Error(`smart_fallback_exhausted: ${errors.join("; ")}`);
+}
+//# sourceMappingURL=model-strategy.js.map

package/dist/model-strategy.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"model-strategy.js","sourceRoot":"","sources":["../src/model-strategy.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAoDrD,MAAM,aAAa,GAAmB;IACpC,gBAAgB;IAChB,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,aAAa,EAAE,IAAI,EAAE,SAAS,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,SAAS,EAAE,MAAM,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,IAAI,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE;IACjS,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,cAAc,EAAE,IAAI,EAAE,SAAS,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,SAAS,EAAE,MAAM,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,IAAI,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE;IACjS,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,QAAQ,EAAE,IAAI,EAAE,UAAU,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,SAAS,EAAE,MAAM,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,IAAI,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,WAAW,CAAC,EAAE;IACvS,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,UAAU,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,SAAS,EAAE,MAAM,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,IAAI,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,WAAW,CAAC,EAAE;IACxS,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,EAAE,YAAY,EAAE,IAAI,EAAE,SAAS,EAAE,MAAM,EAAE,kBAAkB,EAAE,KAAK,EAAE,cAAc,EAAE,KAAK,EAAE,iBAAiB,EAAE,KAAK,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,WAAW,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE;IAC9R,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,SAAS,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,EAAE,YAAY,EAAE,IAAI,EAAE,SAAS,EAAE,MAAM,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,KAAK,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,WAAW,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE;IAEnS,gBAAgB;IAChB,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,yBAAyB,EAAE,IAAI,EAAE,SAAS,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,SAAS,EAAE,MAAM,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,IAAI,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE;IAC3S,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,0BAA0B,EAAE,IAAI,EAAE,UAAU,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,SAAS,EAAE,MAAM,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,IAAI,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,WAAW,CAAC,EAAE;IACzT,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,sBAAsB,EAAE,IAAI,EAAE,SAAS,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,YAAY,EAAE,IAAI,EAAE,SAAS,EAAE,MAAM,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,IAAI,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,WAAW,EAAE,UAAU,CAAC,EAAE;IAEjU,gBAAgB;IAChB,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,kBAAkB,EAAE,IAAI,EAAE,SAAS,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,SAAS,EAAE,OAAO,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,IAAI,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,YAAY,CAAC,EAAE;IACxS,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,gBAAgB,EAAE,IAAI,EAAE,UAAU,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,SAAS,EAAE,OAAO,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,IAAI,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,YAAY,CAAC,EAAE;IAEnT,+BAA+B;IAC/B,EAAE,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,yBAAyB,EAAE,IAAI,EAAE,SAAS,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,YAAY,EAAE,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,KAAK,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,EAAE;IAChS,EAAE,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,sBAAsB,EAAE,IAAI,EAAE,SAAS,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,YAAY,EAAE,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,KAAK,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,MAAM,CAAC,EAAE;IAErR,cAAc;IACd,EAAE,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,cAAc,EAAE,IAAI,EAAE,UAAU,EAAE,eAAe,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,EAAE,YAAY,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,kBAAkB,EAAE,IAAI,EAAE,cAAc,EAAE,IAAI,EAAE,iBAAiB,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE;CACzR,CAAC;AAEF,MAAM,eAAe,GAAuC;IAC1D,UAAU,EAAE,SAAS;IACrB,IAAI,EAAE,UAAU;IAChB,YAAY,EAAE,SAAS;IACvB,IAAI,EAAE,UAAU;IAChB,aAAa,EAAE,UAAU;IACzB,IAAI,EAAE,SAAS;IACf,GAAG,EAAE,UAAU;IACf,QAAQ,EAAE,SAAS;IACnB,QAAQ,EAAE,SAAS;IACnB,SAAS,EAAE,SAAS;CACrB,CAAC;AAEF,MAAM,OAAO,mBAAmB;IACtB,QAAQ,GAA8B,IAAI,GAAG,EAAE,CAAC;IAChD,SAAS,GAAgC,IAAI,GAAG,EAAE,CAAC;IACnD,OAAO,GAAiC,IAAI,GAAG,EAAE,CAAC;IAClD,gBAAgB,GAAmH,EAAE,CAAC;IAE9I,YACE,iBAAiD,EACjD,cAA+B;QAE/B,MAAM,OAAO,GAAG,cAAc,IAAI,aAAa,CAAC;QAChD,KAAK,MAAM,OAAO,IAAI,OAAO,EAAE,CAAC;YAC9B,MAAM,GAAG,GAAG,GAAG,OAAO,CAAC,QAAQ,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;YACnD,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;QAClC,CAAC;QAED,KAAK,MAAM,CAAC,IAAI,EAAE,QAAQ,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,iBAAiB,CAAC,EAAE,CAAC;YACjE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;YACnC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,EAAE;gBACrB,QAAQ,EAAE,IAAI;gBACd,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,YAAY,EAAE,CAAC;gBACf,YAAY,EAAE,CAAC;aAChB,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,MAAM,CAAC,OAAyB;QAC9B,MAAM,UAAU,GAAG,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC;QAE/C,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC5B,MAAM,QAAQ,GAAG,IAAI,CAAC,mBAAmB,EAAE,CAAC;YAC5C,OAAO;gBACL,QAAQ,EAAE,QAAQ,CAAC,QAAQ;gBAC3B,OAAO,EAAE,QAAQ,CAAC,OAAO;gBACzB,aAAa,EAAE,EAAE;gBACjB,aAAa,EAAE,CAAC;gBAChB,gBAAgB,EAAE,IAAI;gBACtB,SAAS,EAAE,yCAAyC;aACrD,CAAC;QACJ,CAAC;QAED,MAAM,MAAM,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YAClC,GAAG,CAAC;YACJ,KAAK,EAAE,IAAI,CAAC,cAAc,CAAC,CAAC,EAAE,OAAO,CAAC;SACvC,CAAC,CAAC,CAAC;QAEJ,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC;QAEzC,MAAM,QAAQ,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,aAAa,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QAEzC,MAAM,eAAe,GAAG,OAAO,CAAC,eAAe,IAAI,IAAI,CAAC,cAAc,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QACvF,MAAM,aAAa,GAAG,IAAI,CAAC,aAAa,CAAC,QAAQ,CAAC,OAAO,EAAE,eAAe,CAAC,CAAC;QAE5E,OAAO;YACL,QAAQ,EAAE,QAAQ,CAAC,QAAQ;YAC3B,OAAO,EAAE,QAAQ,CAAC,OAAO;YACzB,aAAa,EAAE,aAAa,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC;YAChD,aAAa;YACb,gBAAgB,EAAE,QAAQ,CAAC,OAAO,CAAC,YAAY;YAC/C,SAAS,EAAE,IAAI,CAAC,iBAAiB,CAAC,QAAQ,EAAE,OAAO,CAAC;SACrD,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,gBAAgB,CACpB,OAAyB,EACzB,MAAc,EACd,MAAe,EACf,OAA6B;QAE7B,MAAM,SAAS,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACvC,MAAM,iBAAiB,GAAG,SAAS,CAAC,aAAa;aAC9C,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC;aACxC,MAAM,CAAC,CAAC,CAAC,EAAuB,EAAE,CAAC,CAAC,KAAK,SAAS,CAAC,CAAC;QAEvD,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,IAAI,SAAkB,CAAC;QAEvB,cAAc;QACd,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM;gBACnB,CAAC,CAAC,MAAM,SAAS,CAAC,QAAQ,CAAC,cAAc,CAAI,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC;gBACrE,CAAC,CAAC,MAAM,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,MAAM,EAAE,OAAO,CAAqB,CAAC;YAE5E,IAAI,CAAC,aAAa,CAAC,SAAS,CAAC,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,MAAM,CAAC,OAAO,IAAI,CAAC,CAAC,CAAC;YACnF,OAAO,MAAM,CAAC;QAChB,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,SAAS,GAAG,GAAG,CAAC;YAChB,IAAI,CAAC,aAAa,CAAC,SAAS,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC;QAC7C,CAAC;QAED,gBAAgB;QAChB,KAAK,MAAM,gBAAgB,IAAI,iBAAiB,EAAE,CAAC;YACjD,MAAM,eAAe,GAAG,SAAS,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,KAAK,IAAI,CAAC,eAAe,CAAC,gBAAgB,CAAC,CAAC,CAAC;YACjH,IAAI,CAAC,eAAe;gBAAE,SAAS;YAE/B,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,MAAM;oBACnB,CAAC,CAAC,MAAM,gBAAgB,CAAC,cAAc,CAAI,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC;oBACnE,CAAC,CAAC,MAAM,gBAAgB,CAAC,SAAS,CAAC,MAAM,EAAE,OAAO,CAAqB,CAAC;gBAE1E,IAAI,CAAC,aAAa,CAAC,eAAe,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,MAAM,CAAC,OAAO,IAAI,CAAC,CAAC,CAAC;gBACjF,OAAO,MAAM,CAAC;YAChB,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,SAAS,GAAG,GAAG,CAAC;gBAChB,IAAI,eAAe,EAAE,CAAC;oBACpB,IAAI,CAAC,aAAa,CAAC,eAAe,EAAE,GAAG,CAAC,CAAC;gBAC3C,CAAC;YACH,CAAC;QACH,CAAC;QAED,MAAM,IAAI,KAAK,CAAC,qCAAqC,SAAS,YAAY,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC;IAC7H,CAAC;IAEO,aAAa,CAAC,OAAyB;QAC7C,MAAM,UAAU,GAA+D,EAAE,CAAC;QAElF,KAAK,MAAM,CAAC,GAAG,EAAE,OAAO,CAAC,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YAC3C,IAAI,CAAC,IAAI,CAAC,iBAAiB,CAAC,OAAO,EAAE,OAAO,CAAC;gBAAE,SAAS;YAExD,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;YACtD,IAAI,CAAC,QAAQ;gBAAE,SAAS;YAExB,UAAU,CAAC,IAAI,CAAC,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC;QACzC,CAAC;QAED,OAAO,UAAU,CAAC;IACpB,CAAC;IAEO,iBAAiB,CAAC,OAAqB,EAAE,OAAyB;QACxE,IAAI,OAAO,CAAC,kBAAkB,IAAI,CAAC,OAAO,CAAC,kBAAkB;YAAE,OAAO,KAAK,CAAC;QAC5E,IAAI,OAAO,CAAC,cAAc,IAAI,CAAC,OAAO,CAAC,cAAc;YAAE,OAAO,KAAK,CAAC;QACpE,IAAI,OAAO,CAAC,iBAAiB,IAAI,CAAC,OAAO,CAAC,iBAAiB;YAAE,OAAO,KAAK,CAAC;QAE1E,IAAI,OAAO,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;YACrC,MAAM,eAAe,GAAG,OAAO,CAAC,eAAe,IAAI,IAAI,CAAC,cAAc,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;YACvF,MAAM,aAAa,GAAG,IAAI,CAAC,aAAa,CAAC,OAAO,EAAE,eAAe,CAAC,CAAC;YACnE,IAAI,aAAa,GAAG,OAAO,CAAC,UAAU;gBAAE,OAAO,KAAK,CAAC;QACvD,CAAC;QAED,IAAI,OAAO,CAAC,YAAY,KAAK,SAAS,IAAI,OAAO,CAAC,YAAY,GAAG,OAAO,CAAC,YAAY;YAAE,OAAO,KAAK,CAAC;QAEpG,IAAI,OAAO,CAAC,aAAa,IAAI,OAAO,CAAC,IAAI,KAAK,OAAO,CAAC,aAAa,EAAE,CAAC;YACpE,MAAM,YAAY,GAAgB,CAAC,SAAS,EAAE,UAAU,EAAE,SAAS,EAAE,WAAW,CAAC,CAAC;YAClF,MAAM,YAAY,GAAG,YAAY,CAAC,OAAO,CAAC,OAAO,CAAC,aAAa,CAAC,CAAC;YACjE,MAAM,UAAU,GAAG,YAAY,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;YACtD,IAAI,UAAU,GAAG,YAAY,GAAG,CAAC;gBAAE,OAAO,KAAK,CAAC;QAClD,CAAC;QAED,OAAO,IAAI,CAAC;IACd,CAAC;IAEO,cAAc,CAAC,SAAoC,EAAE,OAAyB;QACpF,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,IAAI,UAAU,CAAC;QAChD,IAAI,KAAK,GAAG,CAAC,CAAC;QAEd,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;QAC7D,MAAM,WAAW,GAAG,OAAO,EAAE,WAAW,IAAI,CAAC,CAAC;QAE9C,QAAQ,QAAQ,EAAE,CAAC;YACjB,KAAK,MAAM;gBACT,KAAK,GAAG,IAAI,CAAC,WAAW,CAAC,SAAS,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;gBACrD,MAAM;YACR,KAAK,SAAS;gBACZ,KAAK,GAAG,IAAI,CAAC,cAAc,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC;gBAC/C,MAAM;YACR,KAAK,SAAS;gBACZ,KAAK,GAAG,SAAS,CAAC,OAAO,CAAC,YAAY,GAAG,GAAG,CAAC;gBAC7C,MAAM;YACR,KAAK,UAAU,CAAC;YAChB;gBACE,KAAK,GAAG,CACN,IAAI,CAAC,WAAW,CAAC,SAAS,CAAC,OAAO,EAAE,OAAO,CAAC,GAAG,GAAG;oBAClD,IAAI,CAAC,cAAc,CAAC,SAAS,CAAC,OAAO,CAAC,GAAG,GAAG;oBAC5C,SAAS,CAAC,OAAO,CAAC,YAAY,GAAG,EAAE;oBACnC,WAAW,GAAG,EAAE,CACjB,CAAC;QACN,CAAC;QAED,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;YACnB,MAAM,aAAa,GAAG,eAAe,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;YACtD,IAAI,aAAa,IAAI,SAAS,CAAC,OAAO,CAAC,IAAI,KAAK,aAAa,EAAE,CAAC;gBAC9D,KAAK,IAAI,EAAE,CAAC;YACd,CAAC;QACH,CAAC;QAED,OAAO,KAAK,CAAC;IACf,CAAC;IAEO,WAAW,CAAC,OAAqB,EAAE,OAAyB;QAClE,MAAM,MAAM,GAAG,OAAO,CAAC,eAAe,IAAI,IAAI,CAAC,cAAc,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC9E,MAAM,IAAI,GAAG,IAAI,CAAC,aAAa,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QACjD,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,GAAG,IAAI,GAAG,IAAI,CAAC,CAAC;IACxC,CAAC;IAEO,cAAc,CAAC,OAAqB;QAC1C,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,GAAG,OAAO,CAAC,YAAY,GAAG,EAAE,CAAC,CAAC;IACtD,CAAC;IAEO,aAAa,CAAC,OAAqB,EAAE,MAAc;QACzD,MAAM,WAAW,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC;QAC5C,MAAM,YAAY,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC;QAC7C,OAAO,CAAC,WAAW,GAAG,OAAO,CAAC,eAAe,CAAC,KAAK,GAAG,YAAY,GAAG,OAAO,CAAC,eAAe,CAAC,MAAM,CAAC,GAAG,IAAI,CAAC;IAC9G,CAAC;IAEO,cAAc,CAAC,MAAc;QACnC,OAAO,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC;IACrD,CAAC;IAEO,iBAAiB,CAAC,QAAkD,EAAE,OAAyB;QACrG,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,IAAI,UAAU,CAAC;QAChD,OAAO,YAAY,QAAQ,CAAC,OAAO,CAAC,QAAQ,IAAI,QAAQ,CAAC,OAAO,CAAC,KAAK,WAAW,QAAQ,CAAC,OAAO,CAAC,IAAI,IAAI;YACnG,YAAY,QAAQ,wBAAwB,QAAQ,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI;YACzE,iBAAiB,QAAQ,CAAC,OAAO,CAAC,YAAY,gBAAgB,CAAC,QAAQ,CAAC,OAAO,CAAC,YAAY,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;IAC3H,CAAC;IAEO,mBAAmB;QACzB,MAAM,YAAY,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QAChD,MAAM,WAAW,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,KAAK,MAAM,CAAC,CAAC;QAExF,IAAI,YAAY,IAAI,WAAW,EAAE,CAAC;YAChC,OAAO,EAAE,QAAQ,EAAE,YAAY,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC;QAC1D,CAAC;QAED,MAAM,aAAa,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,KAAM,CAAC;QAC5D,MAAM,YAAY,GAAG,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,KAAM,CAAC;QAC1D,OAAO,EAAE,QAAQ,EAAE,aAAa,EAAE,OAAO,EAAE,YAAY,EAAE,CAAC;IAC5D,CAAC;IAEO,eAAe,CAAC,QAAwB;QAC9C,KAAK,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;YACvC,IAAI,CAAC,KAAK,QAAQ;gBAAE,OAAO,IAAI,CAAC;QAClC,CAAC;QACD,OAAO,SAAS,CAAC;IACnB,CAAC;IAEO,aAAa,CAAC,OAAqB,EAAE,OAAe,EAAE,IAAY;QACxE,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;QACnD,IAAI,CAAC,OAAO;YAAE,OAAO;QAErB,OAAO,CAAC,UAAU,EAAE,CAAC;QACrB,OAAO,CAAC,WAAW,GAAG,CAAC,OAAO,CAAC,WAAW,GAAG,CAAC,OAAO,CAAC,UAAU,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,OAAO,CAAC,UAAU,CAAC;QAChG,OAAO,CAAC,YAAY,GAAG,CAAC,OAAO,CAAC,YAAY,GAAG,CAAC,OAAO,CAAC,UAAU,GAAG,CAAC,CAAC,GAAG,OAAO,CAAC,GAAG,OAAO,CAAC,UAAU,CAAC;QACxG,OAAO,CAAC,YAAY,IAAI,IAAI,CAAC;QAE7B,IAAI,CAAC,gBAAgB,CAAC,IAAI,CAAC;YACzB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;YACrB,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,OAAO,EAAE,IAAI;YACb,OAAO;YACP,IAAI;SACL,CAAC,CAAC;IACL,CAAC;IAEO,aAAa,CAAC,OAAqB,EAAE,KAAc;QACzD,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;QACnD,IAAI,CAAC,OAAO;YAAE,OAAO;QAErB,OAAO,CAAC,UAAU,EAAE,CAAC;QACrB,OAAO,CAAC,WAAW,GAAG,CAAC,OAAO,CAAC,WAAW,GAAG,CAAC,OAAO,CAAC,UAAU,GAAG,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,UAAU,CAAC;QAC5F,OAAO,CAAC,SAAS,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAC3E,OAAO,CAAC,aAAa,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAEnC,IAAI,CAAC,gBAAgB,CAAC,IAAI,CAAC;YACzB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;YACrB,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,OAAO,EAAE,KAAK;YACd,OAAO,EAAE,CAAC;YACV,IAAI,EAAE,CAAC;SACR,CAAC,CAAC;IACL,CAAC;IAED,UAAU;QACR,OAAO,MAAM,CAAC,WAAW,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAC1C,CAAC;IAED,mBAAmB,CAAC,QAAgB,GAAG;QACrC,OAAO,IAAI,CAAC,gBAAgB,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC;IAC7C,CAAC;IAED,WAAW;QACT,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;IAC5C,CAAC;IAED,UAAU,CAAC,OAAqB;QAC9B,MAAM,GAAG,GAAG,GAAG,OAAO,CAAC,QAAQ,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;QACnD,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;IAClC,CAAC;IAED,aAAa,CAAC,MAAc;QAC1B,OAAO,eAAe,CAAC,MAAM,CAAC,IAAI,UAAU,CAAC;IAC/C,CAAC;IAED,kBAAkB,CAAC,MAAc;QAC/B,MAAM,IAAI,GAAG,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC;QACxC,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC;aACtC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,CAAC;aAC5B,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,GAAG,CAAC,CAAC,YAAY,CAAC,CAAC;IACrD,CAAC;CACF;AAED,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAA2B,EAC3B,MAAc,EACd,UAKI,EAAE;IAEN,MAAM,EAAE,qBAAqB,GAAG,CAAC,EAAE,YAAY,GAAG,GAAG,EAAE,WAAW,EAAE,UAAU,EAAE,GAAG,OAAO,CAAC;IAC3F,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,IAAI,YAAY,GAAG,EAAE,CAAC;IAEtB,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;QACjC,MAAM,YAAY,GAAI,QAAgB,CAAC,WAAW,EAAE,IAAI,IAAI,SAAS,CAAC;QACtE,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,WAAW,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,KAAK,CAAC,CAAC;QAChE,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,IAAI,UAAU,IAAI,YAAY,EAAE,CAAC;gBAC/B,UAAU,CAAC,YAAY,EAAE,YAAY,EAAE,qBAAqB,CAAC,CAAC;YAChE,CAAC;YACD,SAAS;QACX,CAAC;QAED,IAAI,OAAO,GAAG,CAAC,CAAC;QAChB,OAAO,OAAO,IAAI,qBAAqB,EAAE,CAAC;YACxC,IAAI,CAAC;gBACH,OAAO,MAAM,QAAQ,CAAC,SAAS,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC;YACvD,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,MAAM,QAAQ,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;gBAClE,MAAM,CAAC,IAAI,CAAC,GAAG,YAAY,KAAK,QAAQ,EAAE,CAAC,CAAC;gBAE5C,IAAI,GAAG,YAAY,gBAAgB,IAAI,GAAG,CAAC,WAAW,IAAI,OAAO,GAAG,qBAAqB,EAAE,CAAC;oBAC1F,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,YAAY,GAAG,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;oBACpE,OAAO,EAAE,CAAC;oBACV,SAAS;gBACX,CAAC;gBAED,IAAI,UAAU,EAAE,CAAC;oBACf,MAAM,YAAY,GAAG,SAAS,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC;oBAChE,IAAI,YAAY,EAAE,CAAC;wBACjB,MAAM,QAAQ,GAAI,YAAoB,CAAC,WAAW,EAAE,IAAI,IAAI,SAAS,CAAC;wBACtE,UAAU,CAAC,YAAY,EAAE,QAAQ,EAAE,QAAQ,CAAC,CAAC;oBAC/C,CAAC;gBACH,CAAC;gBACD,MAAM;YACR,CAAC;QACH,CAAC;QACD,YAAY,GAAG,YAAY,CAAC;IAC9B,CAAC;IAED,MAAM,IAAI,KAAK,CAAC,6BAA6B,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AACpE,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "rax-flow-providers",
-  "version": "0.2.1",
+  "version": "2.0.1",
   "type": "module",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",

package/src/index.ts

@@ -8,4 +8,5 @@ export * from "./cohere-adapter.js";
 export * from "./rest-adapter.js";
 export * from "./host-bridge-adapter.js";
 export * from "./strategy.js";
+export * from "./model-strategy.js";
 export * from "./utils.js";

package/src/model-strategy.ts

@@ -0,0 +1,450 @@
+import { IModelProvider, ModelResponse, ProviderCallOptions, Intent } from "rax-flow-core";
+import { RaxProviderError } from "./error-mapper.js";
+
+export type ModelTier = "economy" | "standard" | "premium" | "reasoning";
+export type SelectionCriteria = "cost" | "latency" | "quality" | "balanced";
+
+export interface ModelProfile {
+  provider: string;
+  model: string;
+  tier: ModelTier;
+  costPer1kTokens: { input: number; output: number };
+  avgLatencyMs: number;
+  maxTokens: number;
+  supportsStructured: boolean;
+  supportsVision: boolean;
+  supportsStreaming: boolean;
+  qualityScore: number;
+  capabilities: string[];
+}
+
+export interface SelectionContext {
+  prompt: string;
+  intent?: Intent;
+  estimatedTokens?: number;
+  maxCostUsd?: number;
+  maxLatencyMs?: number;
+  requiresStructured?: boolean;
+  requiresVision?: boolean;
+  requiresStreaming?: boolean;
+  preferredTier?: ModelTier;
+  criteria?: SelectionCriteria;
+  fallbackChain?: string[];
+}
+
+export interface SelectionResult {
+  provider: IModelProvider;
+  profile: ModelProfile;
+  fallbackChain: ModelProfile[];
+  estimatedCost: number;
+  estimatedLatency: number;
+  reasoning: string;
+}
+
+export interface ProviderMetrics {
+  provider: string;
+  totalCalls: number;
+  successRate: number;
+  avgLatencyMs: number;
+  totalCostUsd: number;
+  lastError?: string;
+  lastErrorTime?: number;
+}
+
+const MODEL_CATALOG: ModelProfile[] = [
+  // OpenAI models
+  { provider: "openai", model: "gpt-4o-mini", tier: "economy", costPer1kTokens: { input: 0.00015, output: 0.0006 }, avgLatencyMs: 400, maxTokens: 128000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.75, capabilities: ["chat", "code", "analysis"] },
+  { provider: "openai", model: "gpt-4.1-mini", tier: "economy", costPer1kTokens: { input: 0.0004, output: 0.0016 }, avgLatencyMs: 450, maxTokens: 128000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.78, capabilities: ["chat", "code", "analysis"] },
+  { provider: "openai", model: "gpt-4o", tier: "standard", costPer1kTokens: { input: 0.0025, output: 0.01 }, avgLatencyMs: 600, maxTokens: 128000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.88, capabilities: ["chat", "code", "analysis", "reasoning"] },
+  { provider: "openai", model: "gpt-4.1", tier: "standard", costPer1kTokens: { input: 0.002, output: 0.008 }, avgLatencyMs: 550, maxTokens: 128000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.90, capabilities: ["chat", "code", "analysis", "reasoning"] },
+  { provider: "openai", model: "o1", tier: "reasoning", costPer1kTokens: { input: 0.015, output: 0.06 }, avgLatencyMs: 5000, maxTokens: 200000, supportsStructured: false, supportsVision: false, supportsStreaming: false, qualityScore: 0.95, capabilities: ["reasoning", "math", "science"] },
+  { provider: "openai", model: "o3-mini", tier: "premium", costPer1kTokens: { input: 0.0011, output: 0.0044 }, avgLatencyMs: 2000, maxTokens: 200000, supportsStructured: true, supportsVision: false, supportsStreaming: true, qualityScore: 0.92, capabilities: ["reasoning", "code", "analysis"] },
+
+  // Claude models
+  { provider: "claude", model: "claude-3-5-haiku-latest", tier: "economy", costPer1kTokens: { input: 0.0008, output: 0.004 }, avgLatencyMs: 300, maxTokens: 200000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.80, capabilities: ["chat", "code", "analysis"] },
+  { provider: "claude", model: "claude-3-5-sonnet-latest", tier: "standard", costPer1kTokens: { input: 0.003, output: 0.015 }, avgLatencyMs: 500, maxTokens: 200000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.91, capabilities: ["chat", "code", "analysis", "reasoning"] },
+  { provider: "claude", model: "claude-3-opus-latest", tier: "premium", costPer1kTokens: { input: 0.015, output: 0.075 }, avgLatencyMs: 1200, maxTokens: 200000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.94, capabilities: ["chat", "code", "analysis", "reasoning", "creative"] },
+
+  // Gemini models
+  { provider: "gemini", model: "gemini-2.0-flash", tier: "economy", costPer1kTokens: { input: 0.0001, output: 0.0004 }, avgLatencyMs: 250, maxTokens: 1000000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.72, capabilities: ["chat", "code", "multimodal"] },
+  { provider: "gemini", model: "gemini-1.5-pro", tier: "standard", costPer1kTokens: { input: 0.00125, output: 0.005 }, avgLatencyMs: 600, maxTokens: 2000000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.85, capabilities: ["chat", "code", "analysis", "multimodal"] },
+
+  // Groq models (fast inference)
+  { provider: "groq", model: "llama-3.3-70b-versatile", tier: "economy", costPer1kTokens: { input: 0.00059, output: 0.00079 }, avgLatencyMs: 80, maxTokens: 128000, supportsStructured: true, supportsVision: false, supportsStreaming: true, qualityScore: 0.78, capabilities: ["chat", "code"] },
+  { provider: "groq", model: "llama-3.1-8b-instant", tier: "economy", costPer1kTokens: { input: 0.00002, output: 0.00002 }, avgLatencyMs: 40, maxTokens: 128000, supportsStructured: true, supportsVision: false, supportsStreaming: true, qualityScore: 0.65, capabilities: ["chat"] },
+
+  // Host bridge
+  { provider: "host", model: "host-managed", tier: "standard", costPer1kTokens: { input: 0, output: 0 }, avgLatencyMs: 1000, maxTokens: 1000000, supportsStructured: true, supportsVision: true, supportsStreaming: true, qualityScore: 0.85, capabilities: ["chat", "code", "analysis"] }
+];
+
+const INTENT_TIER_MAP: Partial<Record<Intent, ModelTier>> = {
+  brainstorm: "economy",
+  spec: "standard",
+  architecture: "premium",
+  plan: "standard",
+  generate_code: "standard",
+  test: "economy",
+  fix: "standard",
+  optimize: "premium",
+  document: "economy",
+  benchmark: "economy"
+};
+
+export class ModelStrategyEngine {
+  private profiles: Map<string, ModelProfile> = new Map();
+  private providers: Map<string, IModelProvider> = new Map();
+  private metrics: Map<string, ProviderMetrics> = new Map();
+  private selectionHistory: Array<{ timestamp: number; provider: string; model: string; success: boolean; latency: number; cost: number }> = [];
+
+  constructor(
+    providerInstances: Record<string, IModelProvider>,
+    customProfiles?: ModelProfile[]
+  ) {
+    const catalog = customProfiles ?? MODEL_CATALOG;
+    for (const profile of catalog) {
+      const key = `${profile.provider}:${profile.model}`;
+      this.profiles.set(key, profile);
+    }
+
+    for (const [name, provider] of Object.entries(providerInstances)) {
+      this.providers.set(name, provider);
+      this.metrics.set(name, {
+        provider: name,
+        totalCalls: 0,
+        successRate: 1,
+        avgLatencyMs: 0,
+        totalCostUsd: 0
+      });
+    }
+  }
+
+  select(context: SelectionContext): SelectionResult {
+    const candidates = this.getCandidates(context);
+    
+    if (candidates.length === 0) {
+      const fallback = this.getFallbackProvider();
+      return {
+        provider: fallback.provider,
+        profile: fallback.profile,
+        fallbackChain: [],
+        estimatedCost: 0,
+        estimatedLatency: 1000,
+        reasoning: "No suitable model found, using fallback"
+      };
+    }
+
+    const scored = candidates.map(c => ({
+      ...c,
+      score: this.scoreCandidate(c, context)
+    }));
+
+    scored.sort((a, b) => b.score - a.score);
+
+    const selected = scored[0];
+    const fallbackChain = scored.slice(1, 4);
+
+    const estimatedTokens = context.estimatedTokens ?? this.estimateTokens(context.prompt);
+    const estimatedCost = this.calculateCost(selected.profile, estimatedTokens);
+
+    return {
+      provider: selected.provider,
+      profile: selected.profile,
+      fallbackChain: fallbackChain.map(f => f.profile),
+      estimatedCost,
+      estimatedLatency: selected.profile.avgLatencyMs,
+      reasoning: this.generateReasoning(selected, context)
+    };
+  }
+
+  async callWithStrategy<T>(
+    context: SelectionContext,
+    prompt: string,
+    schema?: object,
+    options?: ProviderCallOptions
+  ): Promise<ModelResponse<T>> {
+    const selection = this.select(context);
+    const fallbackProviders = selection.fallbackChain
+      .map(p => this.providers.get(p.provider))
+      .filter((p): p is IModelProvider => p !== undefined);
+
+    const startTime = Date.now();
+    let lastError: unknown;
+
+    // Try primary
+    try {
+      const result = schema
+        ? await selection.provider.callStructured<T>(prompt, schema, options)
+        : await selection.provider.callModel(prompt, options) as ModelResponse<T>;
+
+      this.recordSuccess(selection.profile, Date.now() - startTime, result.costUsd ?? 0);
+      return result;
+    } catch (err) {
+      lastError = err;
+      this.recordFailure(selection.profile, err);
+    }
+
+    // Try fallbacks
+    for (const fallbackProvider of fallbackProviders) {
+      const fallbackProfile = selection.fallbackChain.find(p => p.provider === this.getProviderName(fallbackProvider));
+      if (!fallbackProfile) continue;
+
+      try {
+        const result = schema
+          ? await fallbackProvider.callStructured<T>(prompt, schema, options)
+          : await fallbackProvider.callModel(prompt, options) as ModelResponse<T>;
+
+        this.recordSuccess(fallbackProfile, Date.now() - startTime, result.costUsd ?? 0);
+        return result;
+      } catch (err) {
+        lastError = err;
+        if (fallbackProfile) {
+          this.recordFailure(fallbackProfile, err);
+        }
+      }
+    }
+
+    throw new Error(`All providers failed. Last error: ${lastError instanceof Error ? lastError.message : String(lastError)}`);
+  }
+
+  private getCandidates(context: SelectionContext): Array<{ provider: IModelProvider; profile: ModelProfile }> {
+    const candidates: Array<{ provider: IModelProvider; profile: ModelProfile }> = [];
+
+    for (const [key, profile] of this.profiles) {
+      if (!this.meetsRequirements(profile, context)) continue;
+
+      const provider = this.providers.get(profile.provider);
+      if (!provider) continue;
+
+      candidates.push({ provider, profile });
+    }
+
+    return candidates;
+  }
+
+  private meetsRequirements(profile: ModelProfile, context: SelectionContext): boolean {
+    if (context.requiresStructured && !profile.supportsStructured) return false;
+    if (context.requiresVision && !profile.supportsVision) return false;
+    if (context.requiresStreaming && !profile.supportsStreaming) return false;
+
+    if (context.maxCostUsd !== undefined) {
+      const estimatedTokens = context.estimatedTokens ?? this.estimateTokens(context.prompt);
+      const estimatedCost = this.calculateCost(profile, estimatedTokens);
+      if (estimatedCost > context.maxCostUsd) return false;
+    }
+
+    if (context.maxLatencyMs !== undefined && profile.avgLatencyMs > context.maxLatencyMs) return false;
+
+    if (context.preferredTier && profile.tier !== context.preferredTier) {
+      const tierPriority: ModelTier[] = ["economy", "standard", "premium", "reasoning"];
+      const preferredIdx = tierPriority.indexOf(context.preferredTier);
+      const profileIdx = tierPriority.indexOf(profile.tier);
+      if (profileIdx > preferredIdx + 1) return false;
+    }
+
+    return true;
+  }
+
+  private scoreCandidate(candidate: { profile: ModelProfile }, context: SelectionContext): number {
+    const criteria = context.criteria ?? "balanced";
+    let score = 0;
+
+    const metrics = this.metrics.get(candidate.profile.provider);
+    const successRate = metrics?.successRate ?? 1;
+
+    switch (criteria) {
+      case "cost":
+        score = this.scoreByCost(candidate.profile, context);
+        break;
+      case "latency":
+        score = this.scoreByLatency(candidate.profile);
+        break;
+      case "quality":
+        score = candidate.profile.qualityScore * 100;
+        break;
+      case "balanced":
+      default:
+        score = (
+          this.scoreByCost(candidate.profile, context) * 0.3 +
+          this.scoreByLatency(candidate.profile) * 0.3 +
+          candidate.profile.qualityScore * 30 +
+          successRate * 10
+        );
+    }
+
+    if (context.intent) {
+      const preferredTier = INTENT_TIER_MAP[context.intent];
+      if (preferredTier && candidate.profile.tier === preferredTier) {
+        score += 10;
+      }
+    }
+
+    return score;
+  }
+
+  private scoreByCost(profile: ModelProfile, context: SelectionContext): number {
+    const tokens = context.estimatedTokens ?? this.estimateTokens(context.prompt);
+    const cost = this.calculateCost(profile, tokens);
+    return Math.max(0, 100 - cost * 1000);
+  }
+
+  private scoreByLatency(profile: ModelProfile): number {
+    return Math.max(0, 100 - profile.avgLatencyMs / 50);
+  }
+
+  private calculateCost(profile: ModelProfile, tokens: number): number {
+    const inputTokens = Math.ceil(tokens * 0.7);
+    const outputTokens = Math.ceil(tokens * 0.3);
+    return (inputTokens * profile.costPer1kTokens.input + outputTokens * profile.costPer1kTokens.output) / 1000;
+  }
+
+  private estimateTokens(prompt: string): number {
+    return Math.ceil(prompt.split(/\s+/).length * 1.3);
+  }
+
+  private generateReasoning(selected: { profile: ModelProfile; score: number }, context: SelectionContext): string {
+    const criteria = context.criteria ?? "balanced";
+    return `Selected ${selected.profile.provider}/${selected.profile.model} (tier: ${selected.profile.tier}) ` +
+           `based on ${criteria} criteria with score ${selected.score.toFixed(1)}. ` +
+           `Est. latency: ${selected.profile.avgLatencyMs}ms, quality: ${(selected.profile.qualityScore * 100).toFixed(0)}%`;
+  }
+
+  private getFallbackProvider(): { provider: IModelProvider; profile: ModelProfile } {
+    const hostProvider = this.providers.get("host");
+    const hostProfile = Array.from(this.profiles.values()).find(p => p.provider === "host");
+
+    if (hostProvider && hostProfile) {
+      return { provider: hostProvider, profile: hostProfile };
+    }
+
+    const firstProvider = this.providers.values().next().value!;
+    const firstProfile = this.profiles.values().next().value!;
+    return { provider: firstProvider, profile: firstProfile };
+  }
+
+  private getProviderName(provider: IModelProvider): string {
+    for (const [name, p] of this.providers) {
+      if (p === provider) return name;
+    }
+    return "unknown";
+  }
+
+  private recordSuccess(profile: ModelProfile, latency: number, cost: number): void {
+    const metrics = this.metrics.get(profile.provider);
+    if (!metrics) return;
+
+    metrics.totalCalls++;
+    metrics.successRate = (metrics.successRate * (metrics.totalCalls - 1) + 1) / metrics.totalCalls;
+    metrics.avgLatencyMs = (metrics.avgLatencyMs * (metrics.totalCalls - 1) + latency) / metrics.totalCalls;
+    metrics.totalCostUsd += cost;
+
+    this.selectionHistory.push({
+      timestamp: Date.now(),
+      provider: profile.provider,
+      model: profile.model,
+      success: true,
+      latency,
+      cost
+    });
+  }
+
+  private recordFailure(profile: ModelProfile, error: unknown): void {
+    const metrics = this.metrics.get(profile.provider);
+    if (!metrics) return;
+
+    metrics.totalCalls++;
+    metrics.successRate = (metrics.successRate * (metrics.totalCalls - 1)) / metrics.totalCalls;
+    metrics.lastError = error instanceof Error ? error.message : String(error);
+    metrics.lastErrorTime = Date.now();
+
+    this.selectionHistory.push({
+      timestamp: Date.now(),
+      provider: profile.provider,
+      model: profile.model,
+      success: false,
+      latency: 0,
+      cost: 0
+    });
+  }
+
+  getMetrics(): Record<string, ProviderMetrics> {
+    return Object.fromEntries(this.metrics);
+  }
+
+  getSelectionHistory(limit: number = 100): typeof this.selectionHistory {
+    return this.selectionHistory.slice(-limit);
+  }
+
+  getProfiles(): ModelProfile[] {
+    return Array.from(this.profiles.values());
+  }
+
+  addProfile(profile: ModelProfile): void {
+    const key = `${profile.provider}:${profile.model}`;
+    this.profiles.set(key, profile);
+  }
+
+  getIntentTier(intent: Intent): ModelTier {
+    return INTENT_TIER_MAP[intent] ?? "standard";
+  }
+
+  recommendForIntent(intent: Intent): ModelProfile[] {
+    const tier = this.getIntentTier(intent);
+    return Array.from(this.profiles.values())
+      .filter(p => p.tier === tier)
+      .sort((a, b) => b.qualityScore - a.qualityScore);
+  }
+}
+
+export async function smartFallbackCall(
+  providers: IModelProvider[],
+  prompt: string,
+  options: {
+    maxRetriesPerProvider?: number;
+    retryDelayMs?: number;
+    callOptions?: ProviderCallOptions;
+    onFallback?: (from: string, to: string, reason: string) => void;
+  } = {}
+): Promise<ModelResponse<string>> {
+  const { maxRetriesPerProvider = 1, retryDelayMs = 500, callOptions, onFallback } = options;
+  const errors: string[] = [];
+  let lastProvider = "";
+
+  for (const provider of providers) {
+    const providerName = (provider as any).constructor?.name ?? "unknown";
+    const healthy = await provider.healthCheck().catch(() => false);
+    if (!healthy) {
+      if (onFallback && lastProvider) {
+        onFallback(lastProvider, providerName, "health_check_failed");
+      }
+      continue;
+    }
+
+    let attempt = 0;
+    while (attempt <= maxRetriesPerProvider) {
+      try {
+        return await provider.callModel(prompt, callOptions);
+      } catch (err) {
+        const errorMsg = err instanceof Error ? err.message : String(err);
+        errors.push(`${providerName}: ${errorMsg}`);
+
+        if (err instanceof RaxProviderError && err.isRetryable && attempt < maxRetriesPerProvider) {
+          await new Promise(r => setTimeout(r, retryDelayMs * (attempt + 1)));
+          attempt++;
+          continue;
+        }
+
+        if (onFallback) {
+          const nextProvider = providers[providers.indexOf(provider) + 1];
+          if (nextProvider) {
+            const nextName = (nextProvider as any).constructor?.name ?? "unknown";
+            onFallback(providerName, nextName, errorMsg);
+          }
+        }
+        break;
+      }
+    }
+    lastProvider = providerName;
+  }
+
+  throw new Error(`smart_fallback_exhausted: ${errors.join("; ")}`);
+}