Alchemy Cost Tuning

---
name: alchemy-cost-tuning
description: 'Optimize Alchemy API costs through CU budgeting, caching, and plan selection.

  Use when analyzing Alchemy billing, reducing Compute Unit consumption,

  or choosing the right plan for your dApp traffic.

  Trigger: "alchemy cost", "alchemy pricing", "alchemy CU budget",

  "alchemy billing optimization", "alchemy free tier limits".

  '
allowed-tools: Read, Write, Edit, Bash(npm:*)
version: 1.5.0
license: MIT
author: Jeremy Longshore <jeremy@intentsolutions.io>
tags:
- saas
- blockchain
- web3
- alchemy
- cost-optimization
compatibility: Designed for Claude Code
---
# Alchemy Cost Tuning

## Overview

Alchemy pricing is based on Compute Units (CU). Different API methods have different CU costs. Optimize by caching, batching, choosing cheaper methods, and right-sizing your plan.

## Plan Comparison

| Plan | CU/sec | Monthly CU | Price | Best For |
|------|--------|------------|-------|----------|
| Free | 330 | 300M | $0 | Dev/prototyping |
| Growth | 660 | 1.2B | $49/mo | Small dApps |
| Scale | Custom | Custom | Custom | High-traffic apps |

## CU Cost Reference (Top Methods)

| Method | CU | Optimization |
|--------|-----|-------------|
| `eth_blockNumber` | 10 | Cache 12s (1 block) |
| `eth_getBalance` | 19 | Cache 30s |
| `eth_call` | 26 | Cache based on use case |
| `getTokenBalances` | 50 | Cache 60s; batch addresses |
| `getNftsForOwner` | 50 | Cache 5 min |
| `getTokenMetadata` | 50 | Cache 24h (rarely changes) |
| `getAssetTransfers` | 150 | Cache aggressively; paginate |
| `getNftMetadataBatch` | 50 | Use batch over individual calls |

## Prerequisites

- An approved observation window with aggregate method counts and latency; do
  not record end-user wallet data merely to estimate API usage.
- Current plan and compute-unit limits confirmed in the organization’s Alchemy
  account, since commercial terms and method costs may change.
- A cache-invalidation policy that distinguishes safe metadata caching from
  time-sensitive balance, block, and transaction data.

## Instructions

### Step 1: CU Usage Monitor

```typescript
// src/cost/cu-monitor.ts
const CU_COSTS: Record<string, number> = {
  'eth_blockNumber': 10, 'eth_getBalance': 19, 'eth_call': 26,
  'getTokenBalances': 50, 'getNftsForOwner': 50, 'getTokenMetadata': 50,
  'getAssetTransfers': 150, 'getNftMetadataBatch': 50,
};

class CuMonitor {
  private usage: Array<{ method: string; cu: number; timestamp: number }> = [];

  record(method: string): void {
    this.usage.push({ method, cu: CU_COSTS[method] || 26, timestamp: Date.now() });
  }

  getHourlyReport(): { totalCu: number; byMethod: Record<string, number> } {
    const cutoff = Date.now() - 3600000;
    const recent = this.usage.filter(u => u.timestamp > cutoff);
    const byMethod: Record<string, number> = {};
    let totalCu = 0;

    for (const u of recent) {
      byMethod[u.method] = (byMethod[u.method] || 0) + u.cu;
      totalCu += u.cu;
    }

    return { totalCu, byMethod };
  }

  getMonthlyProjection(): { projectedMonthly: number; planRecommendation: string } {
    const hourly = this.getHourlyReport();
    const projectedMonthly = hourly.totalCu * 24 * 30;

    let recommendation = 'Free';
    if (projectedMonthly > 300_000_000) recommendation = 'Growth';
    if (projectedMonthly > 1_200_000_000) recommendation = 'Scale';

    return { projectedMonthly, planRecommendation: recommendation };
  }
}

export { CuMonitor };
```

### Step 2: Cost-Optimized Client Wrapper

```typescript
// src/cost/optimized-client.ts
import { Alchemy, Network } from 'alchemy-sdk';

const cache = new Map<string, { data: any; expiry: number }>();

// Cache token metadata aggressively (rarely changes)
async function getTokenMetadataCached(alchemy: Alchemy, contract: string) {
  const key = `metadata:${contract}`;
  const cached = cache.get(key);
  if (cached && cached.expiry > Date.now()) return cached.data;

  const data = await alchemy.core.getTokenMetadata(contract);
  cache.set(key, { data, expiry: Date.now() + 86400000 }); // 24h cache
  return data;
}

// Use batch instead of individual NFT metadata calls
// 1 batch call (50 CU) vs 100 individual calls (5000 CU)
async function getNftMetadataOptimized(
  alchemy: Alchemy,
  tokens: Array<{ contractAddress: string; tokenId: string }>
) {
  const BATCH_SIZE = 100;
  const results = [];
  for (let i = 0; i < tokens.length; i += BATCH_SIZE) {
    const batch = tokens.slice(i, i + BATCH_SIZE);
    const batchResults = await alchemy.nft.getNftMetadataBatch(batch);
    results.push(...batchResults);
  }
  return results;
}
```

### Step 3: Free Tier Optimization Checklist

```typescript
// For staying within Free tier (330 CU/sec, 300M CU/month):
// 1. Cache eth_blockNumber (saves 10 CU per redundant call)
// 2. Cache token metadata (saves 50 CU per redundant call)
// 3. Use getNftMetadataBatch instead of getNftMetadata (100x savings)
// 4. Avoid getAssetTransfers loops (150 CU each — cache results)
// 5. Use WebSockets instead of polling (one connection vs repeated calls)
// 6. Rate-limit user-facing endpoints to prevent CU bursts
```

## Output

- CU usage monitor with hourly reports and plan projections
- Cost-optimized client with aggressive caching
- Batch operations reducing CU consumption by 100x
- Free tier optimization checklist

## Examples

Collect one hour of aggregate development traffic, feed the counts into
`CuMonitor`, and identify the three methods responsible for the highest CU
projection. Add a 24-hour metadata cache and batched NFT metadata requests in
the test environment, then compare request counts and response correctness
against uncached fixtures. Promote only after cache hits never serve stale
time-sensitive balance or transfer data. If observed use approaches the
account limit or the monitor’s input is incomplete, throttle the noncritical
feature and obtain an updated account-limit report rather than guessing a
budget or silently dropping user-facing requests.

## Error Handling

| Failure | Response |
|---------|----------|
| Usage projection lacks complete observation data | Mark the estimate incomplete and collect a representative window before plan decisions. |
| Cache returns stale chain state | Invalidate the affected key and narrow its TTL or caching scope. |
| Account limit is approached | Apply bounded rate limiting and notify the account owner before service degradation. |
| Batch operation is partially rejected | Preserve successful items, retry only failed items within limits, and expose their unavailable state. |

## Resources

- [Alchemy Pricing](https://www.alchemy.com/pricing)
- [Alchemy Compute Units](https://www.alchemy.com/docs/reference/compute-unit-costs)

## Next Steps

For architecture design, see `alchemy-reference-architecture`.