The Hidden Bottleneck
Every routing engine has the same bottleneck — and most don’t optimize it
Aggregators obsess over routing algorithms and split strategies. But the dominant cost of every quote? Reading pool state. 40-60 pools × 1ms each = you’re spending 50ms just on data before routing even starts.
🔍
Pool Discovery
Find all pools that can route this token pair. AMMs, CLMMs, order books. Fast and well-optimized.
~3ms (cached graph)
⚡
Pool State Reads
Read reserves, tick data, fee tiers, sqrt prices from 40-60 pools. This is where Cachee wins. Every read is 1ms via RPC, 17ns via Cachee L1.
40-60ms (RPC/Redis)
🧮
Route & Quote
Run Dijkstra/split optimization across all viable paths. Return best execution to user.
~15ms (algorithmic)
The math that matters: A typical swap quote budget is 100-200ms. Pool state reads consume 40-60ms — up to 60% of your time. With Cachee: 0.7-0.9ms for all 50+ pools. That 50ms you recover lets you evaluate 50× more routing paths and find +4.2 bps better execution.
The Transformation
Same aggregator. Same algorithm. Different pool state layer.
Standard DEX Stack
~85ms
Request → Quote returned
Pool state reads40-60ms (RPC)
Pools evaluated40-60 pools
Routes scored200-500 routes
Price improvementBaseline
Stale state riskHigh (1s+ lag)
Cachee DEX Stack
~35ms
Request → Quote returned
Pool state reads0.9ms (L1 cache)
Pools evaluated200+ pools
Routes scored10,000+ routes
Price improvement+4.2 bps avg
Stale state riskNear-zero (17ns reads)
How It Works
Three layers of DEX-optimized caching
1
Predict Pool Access
AI watches swap patterns, volume distribution, and token pair popularity to predict which pool states will be queried in the next second.
Sub-second prediction
2
Pre-cache to L1 Memory
Predicted pool state — reserves, tick arrays, fee tiers, sqrt prices — loaded into in-process L1 memory. Direct pointer access, zero network hop.
17ns per pool read
3
Serve Every Route
Your routing engine reads all pool state from L1. 50× more routes evaluated. Better splits found. Tighter quotes returned. More volume captured.
59,000× vs Redis
DEX Use Cases
Every DEX operation is pool-state bound. Cachee unbinds them all.
🔀
Swap Routing
50+ pool reads per quote. Cachee serves all 50 in <0.9ms vs 50ms+ via RPC. 50× more routes explored.
📊
Multi-Hop Splits
Complex A→B→C→D routes need cascading pool reads. L1 cache makes 4-hop routes as fast as 1-hop.
🎯
Limit Orders
Continuous price monitoring across 10,000+ pools. Instant detection when fill conditions are met.
💰
LP Position Management
Real-time impermanent loss tracking across concentrated positions. 17ns state reads vs 1s+ RPC polling.
⚡
JIT Liquidity
Pre-cache incoming swap sizes and pool depths to provide just-in-time liquidity at optimal ranges.
🌐
Cross-Chain Routing
Cache pool state from Solana, Ethereum, Arbitrum, Base, Polygon — one L1 layer, cross-chain quotes.
The Value
What +4.2 bps better execution is worth
Better routes = better prices = more volume = more fees. The flywheel is simple: routing quality drives aggregator market share.
$18M+
Projected annual fee uplift — top-10 DEX aggregator on Cachee
+4.2 bps
Better execution per swap
$12M
Additional fee revenue
How we model it: Jupiter processes $2B+/day in swap volume. At 0.3% average fee, +4.2 bps execution improvement drives measurably higher win rates vs competitors → +22% volume capture → $12M incremental fees. Plus $6M/yr in RPC cost reduction (fewer full-node calls). The real flywheel: better prices → more users → more volume → more fees → wider moat.