Skip to content

README.md

Latest commit

 

History

History
148 lines (110 loc) · 2.82 KB

README.md

File metadata and controls

148 lines (110 loc) · 2.82 KB

Go Semantic Cache Example

A Go implementation of semantic caching for LLM responses using Valkey vector search.

Features

  • Semantic similarity matching using embeddings
  • Configurable similarity threshold
  • TTL-based cache expiration
  • HTTP REST API
  • Zero external dependencies (uses go-redis)

Quick Start

1. Start Valkey

cd ../../../deployment/docker
docker-compose up -d valkey-stack

2. Build and Run

go mod download
go run main.go

3. Configure (Optional)

export VALKEY_HOST=localhost
export VALKEY_PORT=6379
export SIMILARITY_THRESHOLD=0.92
export CACHE_TTL_SECONDS=86400
export PORT=8000

4. Test the Cache

# First query - cache miss
curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{"query": "What is machine learning?"}'

# Check stats
curl http://localhost:8000/cache/stats

API Endpoints

POST /query

{
  "query": "What is machine learning?",
  "skip_cache": false
}

GET /health

Health check endpoint.

GET /cache/stats

Get cache statistics.

POST /cache/clear

Clear all cache entries.

Project Structure

.
├── main.go   # Complete implementation
├── go.mod    # Go module file
└── README.md # This file

Configuration

Variable Default Description
VALKEY_HOST localhost Valkey server host
VALKEY_PORT 6379 Valkey server port
SIMILARITY_THRESHOLD 0.92 Cache hit threshold
CACHE_TTL_SECONDS 86400 Cache TTL (24 hours)
PORT 8000 API server port

Usage as a Library

package main

import (
    "context"
)

func main() {
    config := Config{
        ValkeyHost:          "localhost",
        ValkeyPort:          6379,
        SimilarityThreshold: 0.92,
        CacheTTLSeconds:     86400,
        EmbeddingDimensions: 1536,
    }

    cache := NewSemanticCache(config)
    ctx := context.Background()

    if err := cache.Initialize(ctx); err != nil {
        panic(err)
    }

    // Lookup
    embedding := generateEmbedding("What is AI?")
    result, _ := cache.Lookup(ctx, embedding)

    if result.Hit {
        fmt.Println("Cache hit!", result.Response)
    } else {
        // Store
        cache.Store(ctx, "What is AI?", "AI is...", "gpt-4", embedding)
    }
}

Note

This example uses mock embeddings for demonstration. In production, integrate with an embedding API like OpenAI:

import "github.com/sashabaranov/go-openai"

func getEmbedding(client *openai.Client, text string) ([]float32, error) {
    resp, err := client.CreateEmbeddings(context.Background(), openai.EmbeddingRequest{
        Input: []string{text},
        Model: openai.SmallEmbedding3,
    })
    if err != nil {
        return nil, err
    }
    return resp.Data[0].Embedding, nil
}