メニュー
Smart LLM cost tracking and caching for Python
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | ClawCache Free |
| description | Smart LLM cost tracking and caching for Python |
| version | 0.2.0 |
| author | ClawCache Team |
| category | Developer Tools |
| license | MIT |
ClawCache is a production-ready Python library that helps you track every penny spent on LLM APIs and automatically cache responses to slash costs.
Based on comprehensive simulation with 48 API calls across 4 common use cases:
| Metric | Value |
|---|---|
| Cache Hit Rate | 58.3% |
| Total Cost | $0.0062 |
| API Calls Saved | 28 out of 48 |
| Scenarios Tested | Code Review, Data Analysis, Content Generation, QA Support |
| Scenario | Calls | Cache Hits | Hit Rate |
|---|---|---|---|
| Code Review | 12 | 7 | 58.3% |
| Data Analysis | 12 | 8 | 66.7% |
| Content Generation | 12 | 7 | 58.3% |
| QA Support | 12 | 6 | 50.0% |
pip install clawcache
from clawcache.free.cost import async_monitor_cost
from clawcache.free.cache_basic import BasicCache
# Initialize cache
cache = BasicCache()
# Decorate your LLM function
@async_monitor_cost
async def my_llm_call(prompt, model="gpt-4-turbo"):
# Check cache first
cached = await cache.aget(prompt, model=model)
if cached:
return cached.content
# Make actual API call
response = await openai.ChatCompletion.acreate(
model=model,
messages=[{"role": "user", "content": prompt}]
)
# Cache the response
await cache.aset(prompt, response, model=model)
return response
# Use it
result = await my_llm_call("Explain quantum computing")
ClawCache automatically tracks all your LLM spending:
# See today's detailed cost report
clawcache --report
# Output shows:
# - Money spent today
# - Money saved via cache
# - Total API calls
# - Cache hit rate
# - Efficiency metrics
tiktoken when availableClawCache takes security seriously:
Customize ClawCache behavior via environment variables:
export CLAWCACHE_HOME=/path/to/cache # Default: ~/.clawcache
ClawCache supports composite cache keys for better accuracy:
# Cache by prompt + model + temperature
await cache.aset(
prompt,
response,
model="gpt-4-turbo",
temperature=0.7
)
| Model | Input ($/1M tokens) | Output ($/1M tokens) |
|---|---|---|
| GPT-4 Turbo | $10.00 | $30.00 |
| GPT-3.5 Turbo | $0.50 | $1.50 |
| Claude 3.5 Sonnet | $3.00 | $15.00 |
| Claude 3 Haiku | $0.25 | $1.25 |
@async_monitor_cost
async def review_code(code_snippet):
prompt = f"Review this code for bugs: {code_snippet}"
return await llm_call(prompt, model="gpt-4-turbo")
@async_monitor_cost
async def analyze_data(dataset):
prompt = f"Analyze this dataset: {dataset}"
return await llm_call(prompt, model="claude-3-5-sonnet")
@async_monitor_cost
async def generate_content(topic):
prompt = f"Write a blog post about: {topic}"
return await llm_call(prompt, model="gpt-3.5-turbo")
Based on typical usage patterns:
Want even more savings and insights? ClawCache Pro will include:
Free: Cost tracking with CLI reports + exact-match caching
Pro: Adds social sharing with charts + semantic caching + advanced analytics
Contributions welcome! Please:
MIT License - see LICENSE for details
Made with ❤️ for the AI community
Save money. Track costs. Build better.