- Published on
在搭建 AI 会话 API 的场景中, 可以结合两者优势, 使用 `HTTP + 流式返回` 处理初始响应,快速获取第一段回复, 使用 `WebSocket` 维护后续的对话,实现更顺畅的交互体验.
Ai-companion
All Posts
- nextjs (15)
- express (12)
ai-companion (11)
- best-practices (10)
- react (9)
- web-scraping (8)
- leetcode (7)
- hono (6)
- elasticsearch (6)
- scrapy (5)
- interview (5)
- supabase (4)
- aws (4)
- websocket (4)
- authjs (4)
- k8s (4)
- git (4)
- cloudflare (3)
- undetected-chromedriver (3)
- vercel (3)
- react-query (3)
- prisma (3)
- protobuf (3)
- shadcn (3)
- curl_cffi (2)
- clerk (2)
- authentication (2)
- deployment (2)
- drizzle (2)
- neon (2)
- mongodb (2)
- redis (2)
- tailwind-css (2)
- tiptap (2)
- zod (2)
- zustand (2)
- pino (2)
- browser-automation (1)
- cache-control (1)
- free-tier (1)
- cloud-databases (1)
- sql (1)
- nosql (1)
- planetscale (1)
- docker (1)
- postgresql (1)
- flyio (1)
- giscus (1)
- isr (1)
- jotai (1)
- newsletter (1)
- mailerlite (1)
- redux (1)
- reford (1)
- english (1)
- ai-integration (1)
- sonner (1)
- swr (1)
- s3 (1)
- test (1)
- vscode (1)
- trie-树 (1)
- token (1)
- tts (1)
- oauth (1)
- credential (1)
- eks (1)
- cicd (1)
- apidoc (1)
- pm2 (1)
- cluster (1)
- kafka (1)
- streaming (1)
- superjson (1)
- leetcode_javascript (1)
- lib (1)
- mongoose (1)
- Published on
使用 pino 记录日志并将其写入到 Amazon S3 后通过 Athena 进行分析- Published on
在构建 AI 平台(如 ChatGPT、DeepSeek、Qwen 等)时,Token 控制的优化对成本管理、性能提升和用户体验至关重要- Published on
将敏感词构建成 Trie 树,支持快速前缀匹配,效率高于循环或 includes(), 适用于大规模敏感词库.- Published on
由于大模型存在 Token 限制(如 GPT-4 上限约 128k tokens),长对话会因 Token 超限而丢失早期对话内容.`滑动窗口 + 摘要压缩 + 长期记忆持久化 (如 Redis、MongoDB)`, 不仅能够在 GPT 的 Token 限制内高效保留对话信息,还能最大化上下文的完整度,是在多轮对话管理中值得推荐的解决方案.