ai: add free model fallback chain (deepseek + llama) with 429 handling and modelUsed tagging

Author: gitcoder89431

src/lib/stores/ruixen.ts

@@ -106,6 +106,12 @@ class RateLimiter {
 class Ruixen {
 	private apiKey = '';
 	private model = 'openai/gpt-oss-120b:free';
+	private fallbackModels: string[] = [
+		'openai/gpt-oss-120b:free',
+		'deepseek/deepseek-chat-v3.1:free',
+		'deepseek/deepseek-r1:free',
+		'meta-llama/llama-3.3-70b-instruct:free',
+	];
 	private analyzer: AIAnalyzer | null = null;
 	private limiter = new RateLimiter();
 	private queue: QueueTask[] = [];
@@ -136,24 +142,11 @@ class Ruixen {
 				return;
 			}
 
-			// Try immediate if allowed
+			// Try immediate if allowed (with fallback model sequence on 429)
 			if (this.limiter.canRequest(this.model)) {
 				this.limiter.record(this.model);
 				this.dailyUsage.set(loadDailyCount().count);
-				this.analyzer
-					.analyzeJournalEntry(pet, fullEntry)
-					.then(resolve)
-					.catch((e) => {
-						console.error('Ruixen immediate analysis failed:', e);
-						const msg = String((e as Error)?.message || e);
-						if (msg.includes('429') || /Rate limit exceeded/i.test(msg)) {
-							const today = new Date().toDateString();
-							setExhausted(today);
-							saveDailyCount({ date: today, count: DAILY_FREE_LIMIT });
-							this.dailyUsage.set(DAILY_FREE_LIMIT);
-						}
-						resolve(this.offlineHeuristic(pet, fullEntry));
-					});
+				this.tryModelsSequentially(fullEntry, pet).then(resolve);
 				return;
 			}
 
@@ -225,26 +218,11 @@ class Ruixen {
 
 				this.limiter.record(this.model);
 				this.dailyUsage.set(loadDailyCount().count);
-				try {
-					const res = await this.analyzer.analyzeJournalEntry(
-						next.pet,
-						this.toJournalEntry(next.entry, next.pet)
-					);
-					next.resolve(res);
-				} catch (e) {
-					console.error('Ruixen queued analysis failed:', e);
-					const msg = String((e as Error)?.message || e);
-					if (msg.includes('429') || /Rate limit exceeded/i.test(msg)) {
-						const today = new Date().toDateString();
-						setExhausted(today);
-						saveDailyCount({ date: today, count: DAILY_FREE_LIMIT });
-						this.dailyUsage.set(DAILY_FREE_LIMIT);
-					}
-					next.resolve(this.offlineHeuristic(next.pet, this.toJournalEntry(next.entry, next.pet)));
-				} finally {
-					this.queue.shift();
-					this.queueSize.set(this.queue.length);
-				}
+				const journalEntry = this.toJournalEntry(next.entry, next.pet);
+				const res = await this.tryModelsSequentially(journalEntry, next.pet);
+				next.resolve(res);
+				this.queue.shift();
+				this.queueSize.set(this.queue.length);
 			}
 			this.running = false;
 		};
@@ -417,6 +395,41 @@ class Ruixen {
 			},
 		];
 	}
+
+	/**
+	 * Attempt analysis across fallback free models sequentially.
+	 * On 429 from a model, tries the next. Other errors short-circuit to offline heuristic.
+	 * If all fail with 429, mark day exhausted and return offline heuristic.
+	 */
+	private async tryModelsSequentially(
+		entry: JournalEntry,
+		pet: PetPanelData
+	): Promise<AnalysisResult> {
+		if (!this.analyzer) return this.offlineHeuristic(pet, entry);
+		for (let i = 0; i < this.fallbackModels.length; i++) {
+			const model = this.fallbackModels[i];
+			try {
+				const res = await this.analyzer.analyzeJournalEntry(pet, entry, model);
+				return res;
+			} catch (e) {
+				const msg = String((e as Error)?.message || e);
+				const is429 = /429|Rate limit exceeded/i.test(msg);
+				if (!is429) {
+					console.warn('Model attempt failed (non-429), aborting fallbacks:', model, e);
+					return this.offlineHeuristic(pet, entry);
+				}
+				console.warn('Model rate-limited, trying next free model:', model);
+				if (i === this.fallbackModels.length - 1) {
+					const today = new Date().toDateString();
+					setExhausted(today);
+					saveDailyCount({ date: today, count: DAILY_FREE_LIMIT });
+					this.dailyUsage.set(DAILY_FREE_LIMIT);
+					return this.offlineHeuristic(pet, entry);
+				}
+			}
+		}
+		return this.offlineHeuristic(pet, entry);
+	}
 }
 
 export const ruixen = new Ruixen();

src/lib/utils/ai-analysis.ts

@@ -8,6 +8,8 @@ export interface AnalysisResult {
 	healthConcerns: string[];
 	recommendations: string[];
 	nextCheckupSuggestion?: string;
+	/** Which model actually produced this analysis (for fallback visibility) */
+	modelUsed?: string;
 }
 
 export class AIAnalyzer {
@@ -17,12 +19,16 @@ export class AIAnalyzer {
 
 	constructor(apiKey: string, _model: string = 'openai/gpt-oss-120b:free') {
 		this.apiKey = apiKey;
-		// Hard-enforce the chosen free model for all calls
-		this.model = 'openai/gpt-oss-120b:free';
+		this.model = _model || 'openai/gpt-oss-120b:free';
 	}
 
-	async analyzeJournalEntry(pet: PetPanelData, entry: JournalEntry): Promise<AnalysisResult> {
+	async analyzeJournalEntry(
+		pet: PetPanelData,
+		entry: JournalEntry,
+		modelOverride?: string
+	): Promise<AnalysisResult> {
 		const prompt = this.buildAnalysisPrompt(pet, entry);
+		const model = modelOverride || this.model;
 		const referer = typeof window !== 'undefined' ? window.location.origin : undefined;
 
 		if (typeof window !== 'undefined' && (import.meta as any)?.env?.DEV) {
@@ -41,7 +47,7 @@ export class AIAnalyzer {
 				},
 				body: JSON.stringify({
 					apiKey: this.apiKey,
-					model: this.model,
+					model,
 					messages: [
 						{
 							role: 'system',
@@ -71,14 +77,16 @@ export class AIAnalyzer {
 				console.debug('[Ruixen] Response (journal) preview', String(content).slice(0, 300));
 			}
 			const parsed = this.parseAnalysisResponse(content);
-			return this.sanitizeAnalysis(parsed, pet, entry);
+			const clean = this.sanitizeAnalysis(parsed, pet, entry);
+			clean.modelUsed = model;
+			return clean;
 		} catch (error) {
 			console.error('AI Analysis Error:', error);
 			throw error;
 		}
 	}
 
-	async analyzeWeeklySummary(pet: PetPanelData): Promise<string> {
+	async analyzeWeeklySummary(pet: PetPanelData, modelOverride?: string): Promise<string> {
 		const sevenDays =
 			(pet.journalEntries || [])
 				.slice()
@@ -98,14 +106,15 @@ PET: ${pet.name} (${pet.breed || pet.species || 'pet'}, ${pet.age ?? 'unknown'}y
 LAST 7 DAYS:\n${sevenDays}`;
 
 		const referer = typeof window !== 'undefined' ? window.location.origin : undefined;
+		const model = modelOverride || this.model;
 		const response = await fetch(this.baseUrl, {
 			method: 'POST',
 			headers: {
 				'Content-Type': 'application/json',
 			},
 			body: JSON.stringify({
 				apiKey: this.apiKey,
-				model: this.model,
+				model,
 				messages: [
 					{
 						role: 'system',