convcontext

ExampleManager_tokenSavings demonstrates the token savings from context pruning

package main

import (
	"fmt"

	"github.com/2389-research/hex/internal/convcontext"
	"github.com/2389-research/hex/internal/core"
)

func main() {
	// Simulate a long conversation (500 messages with substantial content)
	messages := make([]core.Message, 500)
	messages[0] = core.Message{Role: "system", Content: "You are a helpful coding assistant with extensive knowledge of Go, Python, JavaScript, and software architecture."}

	for i := 1; i < 500; i++ {
		if i%2 == 1 {
			messages[i] = core.Message{
				Role: "user",
				Content: fmt.Sprintf("This is user message #%d asking a detailed question about coding in Go and Python. "+
					"I need help understanding how to implement a feature with proper error handling, testing, and documentation. "+
					"Can you provide a complete example with best practices?", i),
			}
		} else {
			messages[i] = core.Message{
				Role: "assistant",
				Content: fmt.Sprintf("This is assistant response #%d providing a detailed explanation of code concepts with examples. "+
					"Here's how you implement that feature: First, you need to set up the proper structure, then handle errors appropriately, "+
					"write comprehensive tests, and document your code thoroughly. Let me show you a complete example...", i),
			}
		}
	}

	// Without context management
	tokensWithoutPruning := convcontext.EstimateMessagesTokens(messages)

	// With context management (1000 token limit for demonstration)
	manager := convcontext.NewManager(1000)
	prunedMessages := manager.Prune(messages)
	tokensWithPruning := convcontext.EstimateMessagesTokens(prunedMessages)

	// Calculate savings
	tokensSaved := tokensWithoutPruning - tokensWithPruning
	percentSaved := float64(tokensSaved) / float64(tokensWithoutPruning) * 100

	// Cost calculation (Sonnet 4.5 pricing: $3/million input tokens)
	costWithout := float64(tokensWithoutPruning) * 3.0 / 1_000_000
	costWith := float64(tokensWithPruning) * 3.0 / 1_000_000
	costSaved := costWithout - costWith

	fmt.Printf("Original conversation: %d messages, ~%d tokens\n", len(messages), tokensWithoutPruning)
	fmt.Printf("After pruning: %d messages, ~%d tokens\n", len(prunedMessages), tokensWithPruning)
	fmt.Printf("Tokens saved: %d (%.1f%%)\n", tokensSaved, percentSaved)
	fmt.Printf("Cost saved per request: $%.4f\n", costSaved)
	fmt.Printf("Cost saved over 100 requests: $%.2f\n", costSaved*100)

}

Output:
Original conversation: 500 messages, ~37400 tokens
After pruning: 14 messages, ~995 tokens
Tokens saved: 36405 (97.3%)
Cost saved per request: $0.1092
Cost saved over 100 requests: $10.92