How I cut AI API costs by 80% with caching and smart routing

The Problem

If you're building with OpenAI or Claude, you're
probably overpaying by 60-80% on every API call.

Here's why:

Most AI apps call GPT-4 for every single request —
even when they already have the answer cached from
a previous call. Same question, 100 different users,
100 full-price API calls.

I got tired of seeing this problem everywhere,
so I built VibeCore to fix it automatically.

What is VibeCore?

VibeCore is a middleware layer that sits between
your app and any AI API. It automatically:

• Caches repeated prompts (zero cost on duplicates)
• Understands similar prompts (semantic caching)
• Routes simple queries to free models
• Tracks your savings on every request

How it works

Layer 1 — Exact Cache

When the same prompt is asked again, VibeCore
returns the cached response instantly.

Cost: Rs.0
Speed: ~5ms

Layer 2 — Semantic Cache

When a similar prompt is asked (e.g. "capital of
France?" vs "What is France's capital?"), VibeCore
finds the closest cached response using embeddings.

Cost: Rs.0
Speed: ~30ms

Layer 3 — Smart Routing

Simple prompts (under 20 words, no complex keywords)
are routed to free local models like Groq's llama.

Cost: Rs.0
Speed: ~500ms

Integration

Install the npm package:

npm install @aadi0001/vibecore

Use it in your app:

const VibeCore = require('@aadi0001/vibecore')

const vc = new VibeCore('YOUR_API_KEY')

const result = await vc.generate('What is photosynthesis?')

console.log(result.response)
console.log('Saved: Rs.' + result.saved)
console.log('Source:', result.source)

For Python:

import requests

response = requests.post(
'https://vibecore-07n6.onrender.com/generate',
json={'prompt': 'What is photosynthesis?'},
headers={'x-api-key': 'YOUR_API_KEY'}
)

print(response.json()['response'])
print('Saved:', response.json()['saved'])

Response format

Every response includes cost data:

{
"response": "Photosynthesis is...",
"cached": false,
"source": "groq",
"saved": 0.012,
"total_saved": 0.024
}

Real results

In testing with 10 requests:

• 6 cache hits (60% cache rate)
• 4 groq calls (free model)
• 0 paid API calls
• Total saved: Rs.0.08

At scale with 10,000 requests/day:

• Estimated savings: Rs.800/day
• Monthly savings: Rs.24,000

The dashboard

Every user gets a personal dashboard showing:

• Total requests made
• Total money saved
• Cache hit rate
• Live request log

Get started free

1. Get your free API key (1000 requests, no credit card):
   https://vibecore-07n6.onrender.com

2. Install:
   npm install @aadi0001/vibecore

3. Replace your AI calls — savings start immediately.

Tech stack

• FastAPI (Python backend)
• Redis (caching)
• Groq API (free AI model)
• Sentence Transformers (semantic similarity)
• Node.js SDK (npm package)
• Render (deployment)

Built this in 48 hours. Would love your feedback
in the comments!

What other AI cost optimizations have you tried?