Acervo Benchmark Report

Every message grows your context window. Acervo replaces growing history with a constant-size knowledge graph. Here's the proof.

76%
Avg Token Savings
360
Turns Tested
94%
Context Hit Rate
165
Entities Extracted
0
Phantom Entities

2026-03-27 07:45 — v0.2.2-2 — 6 scenarios

The Problem

Traditional conversation memory has three fundamental flaws.

📈

Linear Token Growth

Every message adds to the context window. By turn 50, you're sending thousands of tokens of old history with every request.

💸

Exploding Costs

API costs scale with token count. A 100-turn conversation costs 10x more than the first 10 turns combined.

🔍

Lost Context

Truncation discards old messages. The AI forgets what the user said 30 turns ago even though it might be critical now.

Acervo vs Full History: Token Usage Over Time

The red line (full history) grows with every turn. The green line (Acervo) stays flat because the knowledge graph compresses context.

Scenario Case Studies

Real conversation patterns tested end-to-end. Each scenario runs through the full Acervo pipeline with a real LLM.

Context Resurrection
Cost Crossover
Peak Compression
Graph Milestone
Small Talk Resilience

What This Means For Your Wallet

Projected savings per 1,000 conversations based on real API pricing.

Based on published API pricing (per 1M tokens). Output estimated at 30% of input. Actual costs depend on response length.