写给真正
交付技术栈的人
关于 React、Node.js、TypeScript 与 AI 的原创文章，来自与我们操作层软件同一套工程实践。正文为英文。
标签：llm-costs
在 Ubuntu 24.04 上用 Qdrant 为 LLM 应用做语义缓存
按含义而不是按文本缓存回答，压低 LLM API 开销：Ubuntu 24.04 上的 Qdrant 部署、租户隔离、按路由设置相似度阈值，以及 IOPS 成本。
788 词阅读Cutting LLM Costs: Caching, Model Routing and Prompt Trimming
Production LLM bills grow fast. Use semantic caching, model routing, leaner prompts, RAG reranking, output caps and batching to spend less without losing quality.
1824 词阅读Where Our LLM API Bill Went and the Architecture Changes That Cut It
Step-level token attribution, model tiering, prompt trimming and backoff with jitter cut a coding agent's API costs; includes cost formulas and common optimization mistakes.
1861 词阅读Cut Repeated Prompt Costs by 90% With One Line of Prompt Caching
Turn on automatic prompt caching with a single field, verify hits in the usage block, and avoid the prefix mistakes and minimum-length traps that silently erase savings.
914 词阅读Why Hermes Agent Burns Tokens and How to Cut Its Prompt to 4K
A one-word hello to Hermes Agent can trigger 30 requests and 1.5M tokens. See where the 16K prompt comes from and which CLI commands trim it below 4,000 tokens.
877 词阅读