| name | gemini-api-caching |
| description | Best practices for Gemini API caching strategy including cache versioning, entity-stable keys, and cache busting for WescoBar |
Gemini API Caching
Purpose
Implement robust caching strategies for Gemini API calls to reduce cost, improve latency, and provide better user experience while supporting cache invalidation when needed.
When to Use
- Implementing any Gemini API feature
- Generating character portraits or world images
- Caching AI-generated content
- Reducing API quota usage
- Improving app performance
Caching Strategy
From AGENTS.md Section 3: "A robust, multi-layered local storage cache"
1. Cache Versioning
2. Entity-Stable Keys
3. Cache Busting
Instructions
Step 1: Define Cache Version
// Global cache version constant
const CACHE_VERSION = 'v2';
// Prepend to all cache keys
const getCacheKey = (entity: string, id: string) => {
return `${CACHE_VERSION}-${entity}:${id}`;
};
// Example
const portraitKey = getCacheKey('character-portrait', 'core-pablo');
// Result: "v2-character-portrait:core-pablo"
Purpose: Instant global cache invalidation by bumping version
When to bump:
- Prompt template changes
- Image generation model updates
- Quality improvements needed
- Global regeneration required
Step 2: Use Entity-Stable Keys
// ✅ CORRECT: Entity-based key (stable)
const cacheKey = `${CACHE_VERSION}-character-portrait:${character.id}`;
// ❌ WRONG: Prompt-based key (invalidates on prompt changes)
const cacheKey = `${CACHE_VERSION}-${fullPromptText}`;
Why entity-stable:
- Same character = same cache key
- Minor prompt tweaks don't invalidate cache
- Consistent across sessions
- Reduces unnecessary regeneration
Entity Types:
character-portrait:<id>- Character portraitsworld-scene:<id>- World/location imagesstory-illustration:<id>- Story illustrationsui-element:<name>- UI-generated images
Step 3: Implement Cache Read
async function getCharacterPortrait(
character: Character,
options?: { forceRebuild?: boolean }
): Promise<string> {
// Check force rebuild flag
if (options?.forceRebuild) {
return await generateNewPortrait(character);
}
// Build cache key
const cacheKey = `${CACHE_VERSION}-character-portrait:${character.id}`;
// Check localStorage cache
const cached = localStorage.getItem(cacheKey);
if (cached) {
console.log(`✅ Cache hit: ${cacheKey}`);
return cached; // Return cached URL
}
console.log(`⚠️ Cache miss: ${cacheKey}`);
// Generate new image
const imageUrl = await generateNewPortrait(character);
// Store in cache
localStorage.setItem(cacheKey, imageUrl);
return imageUrl;
}
Step 4: Implement Cache Write
async function generateAndCachePortrait(
character: Character
): Promise<string> {
// Generate image via Gemini API
const imageUrl = await geminiService.generateImage({
prompt: buildCharacterPrompt(character),
// ... other options
});
// Cache the result
const cacheKey = `${CACHE_VERSION}-character-portrait:${character.id}`;
localStorage.setItem(cacheKey, imageUrl);
console.log(`💾 Cached: ${cacheKey}`);
return imageUrl;
}
Step 5: Implement Cache Busting
// Option 1: Force rebuild via flag
async function regeneratePortrait(character: Character) {
const imageUrl = await getCharacterPortrait(character, {
forceRebuild: true // Bypass cache
});
return imageUrl;
}
// Option 2: Clear specific cache entry
function clearPortraitCache(character: Character) {
const cacheKey = `${CACHE_VERSION}-character-portrait:${character.id}`;
localStorage.removeItem(cacheKey);
console.log(`🗑️ Cleared cache: ${cacheKey}`);
}
// Option 3: Clear all portraits
function clearAllPortraits() {
const prefix = `${CACHE_VERSION}-character-portrait:`;
Object.keys(localStorage)
.filter(key => key.startsWith(prefix))
.forEach(key => localStorage.removeItem(key));
console.log(`🗑️ Cleared all portrait cache`);
}
// Option 4: Bump global version
// Change CACHE_VERSION = 'v2' → 'v3'
// All old caches become stale automatically
Step 6: Handle Cache Expiration (Optional)
interface CachedImage {
url: string;
timestamp: number;
version: string;
}
function getCachedImage(key: string, maxAge: number = 7 * 24 * 60 * 60 * 1000): string | null {
const cached = localStorage.getItem(key);
if (!cached) return null;
try {
const data: CachedImage = JSON.parse(cached);
// Check version
if (data.version !== CACHE_VERSION) {
localStorage.removeItem(key);
return null;
}
// Check age
const age = Date.now() - data.timestamp;
if (age > maxAge) {
localStorage.removeItem(key);
return null;
}
return data.url;
} catch {
// Invalid format - clear cache
localStorage.removeItem(key);
return null;
}
}
function setCachedImage(key: string, url: string) {
const data: CachedImage = {
url,
timestamp: Date.now(),
version: CACHE_VERSION
};
localStorage.setItem(key, JSON.stringify(data));
}
Integration Patterns
With Rate Limiting Skill
Combine with gemini-api-rate-limiting skill:
async function generateImagesSequentially(characters: Character[]) {
for (const character of characters) {
// Check cache first (from this skill)
const cached = await getCharacterPortrait(character);
if (cached) {
updateCharacterImage(character.id, cached);
continue; // Skip API call
}
// Cache miss - generate with rate limiting
// (from gemini-api-rate-limiting skill)
try {
const imageUrl = await generateWithTimeout(character, 30000);
updateCharacterImage(character.id, imageUrl);
} catch (error) {
handleGenerationError(character.id, error);
}
// Delay between calls (rate limiting)
await new Promise(resolve => setTimeout(resolve, 2000));
}
}
UI Integration
// In React component
function CharacterCard({ character }: { character: Character }) {
const [imageUrl, setImageUrl] = useState<string | null>(null);
const [isRegenerating, setIsRegenerating] = useState(false);
useEffect(() => {
// Load from cache or generate
getCharacterPortrait(character).then(setImageUrl);
}, [character.id]);
const handleRegenerate = async () => {
setIsRegenerating(true);
// Force rebuild (cache busting)
const newUrl = await regeneratePortrait(character);
setImageUrl(newUrl);
setIsRegenerating(false);
};
return (
<div>
<img src={imageUrl || placeholderImage} alt={character.name} />
<button onClick={handleRegenerate} disabled={isRegenerating}>
{isRegenerating ? 'Regenerating...' : 'Regenerate Portrait'}
</button>
</div>
);
}
Cache Management UI
// Admin/settings panel for cache management
function CacheManagement() {
const [stats, setStats] = useState({ portraits: 0, scenes: 0, total: 0 });
useEffect(() => {
// Calculate cache stats
const portraitKeys = Object.keys(localStorage)
.filter(key => key.includes('character-portrait'));
setStats({
portraits: portraitKeys.length,
scenes: 0, // Add scene counting
total: localStorage.length
});
}, []);
const clearAllCache = () => {
const confirmed = window.confirm('Clear all cached images? This will regenerate on next load.');
if (confirmed) {
Object.keys(localStorage)
.filter(key => key.startsWith(CACHE_VERSION))
.forEach(key => localStorage.removeItem(key));
alert('Cache cleared!');
window.location.reload();
}
};
return (
<div>
<h3>Gemini API Cache</h3>
<p>Cached portraits: {stats.portraits}</p>
<p>Total cached items: {stats.total}</p>
<button onClick={clearAllCache}>Clear All Cache</button>
</div>
);
}
Best Practices
- Always check cache first - Reduce API calls
- Use entity-stable keys - Don't invalidate on prompt changes
- Version globally - Easy bulk invalidation
- Provide regenerate option - Let users force refresh
- Consider expiration - For time-sensitive content
- Monitor cache size - localStorage has limits (~5-10MB)
- Handle cache errors - Fallback to generation
Related Skills
gemini-api-rate-limiting- Combine cache with rate limitinggemini-api-error-handling- Handle cache failures
Storage Limits
localStorage Limits
- Size: ~5-10MB depending on browser
- Monitor usage: Track total cache size
- Cleanup strategy: Remove oldest entries when full
function getCacheSize(): number {
let total = 0;
for (const key in localStorage) {
if (localStorage.hasOwnProperty(key)) {
total += localStorage[key].length + key.length;
}
}
return total; // bytes
}
// If approaching limit, clear old entries
if (getCacheSize() > 5 * 1024 * 1024) { // 5MB
clearOldestEntries();
}
Notes
- Caching reduces API costs significantly
- Entity-stable keys prevent unnecessary cache invalidation
- Global version bumping enables instant cache refresh
- Force rebuild option gives users control
- Combine with rate limiting for complete API management