Guides
Playbooks
Field notes for shipping with LLMs. English primary; Chinese pages under each card labeled 中文.
Hubs: Start here · Tools · Changelog · 中文
How tool schemas, transcripts, and multi-step agent loops inflate token usage — and a practical way to estimate overhead before you ship.
A practical privacy checklist for adopting AI coding assistants: training defaults, retention, secrets, repo scope, and vendor paperwork — without fake security scores.
How to set canonical URLs when you cross-post AI articles to Medium, newsletters, or company blogs — avoid duplicate-content confusion without myths.
Practical ways to estimate and count LLM tokens offline — heuristics, local tokenizers, and browser tools — before you burn API credits.
A practical decision matrix for choosing among Cursor, GitHub Copilot, and Claude Code — privacy, workflow, cost shape, and trial design. No fake scores.
Treat system prompts like versioned config: changelogs, owners, diffs, rollbacks, and review — so prompt edits do not become invisible production changes.
A practical worksheet to compare small/fast model costs per 1K tokens — inputs, outputs, retries, and when price alone misleads.
Production-oriented prompt patterns for JSON / schema-constrained LLM outputs: contracts, examples, validation loops, and failure modes — no fake benchmarks.
A practical framework for LLM feature unit economics in SaaS — cost per successful action, gross margin impact, pricing guardrails, and when to kill a feature.
A practical playbook for LLM rate limits: detect 429s, exponential backoff with jitter, budgets, queues, and user-facing degradation — without magical retry loops.
Why local tokenizers disagree with vendor usage fields, when the gap is harmless, and how to document the error band for budgeting and CI.
Build a pricing worksheet for OpenRouter versus direct OpenAI (or any gateway vs direct) — list price, routing fees, retries, and operational tradeoffs.
Estimate token and dollar cost for RAG features: chunk sizes, top-k, system overhead, retries, and how context windows blow monthly bills.
A practical playbook to shrink LLM prompts — cut noise, structure context, compress examples — while keeping the behaviors you actually need.
Design a retry budget for LLM-backed features — how many retries, when to fail open/closed, and how retries show up in token spend.
A lean AI tool stack for solopreneurs: writing, research, coding, publishing, and cost control — with selection criteria and a no-hype setup path.
What tokens are, how counting differs by model family, why estimators disagree, and how to budget LLM usage without treating vendor docs as mythology.
A 2026-ready AI SEO workflow: intent mapping, outline freeze, draft assist, on-page packs, internal links, and measurement — without black-hat tricks or fake test claims.
How to write system prompts for research agents: roles, tool use, citation rules, stop conditions, and anti-hallucination constraints that hold up in production.
Design a practical AI content pipeline: intake, outline freeze, section drafts, fact audit, SEO pack, publish — with tools, owners, and SLAs.
A practical comparison framework for ChatGPT, Claude, and Gemini on writing tasks — strengths, failure modes, and when to route work — without fake bake-off scores.
A practical playbook for capping LLM spend: token budgets, model routing, caching, retry policy, and monthly forecasting without a finance PhD.
Reusable prompt templates for briefs, outlines, drafts, SEO packs, and edits — designed for content teams that need consistency, not one-off chat magic.
How to write and maintain system prompts for product features: contracts, tool use, refusals, and change control.
A decision checklist for picking an AI coding assistant: privacy, IDE fit, context, pricing, and failure modes — skip the fake bake-offs.
A concrete pipeline for briefs → drafts → edits → publish using LLMs as assistants, with human gates where quality and liability live.
A practical method to forecast token spend: count inputs, model rates, retries, and caching — so your AI feature does not surprise finance.
A field checklist for writing prompts that hold up in production — roles, constraints, output schemas, and failure modes — without hype.
中文
2026 可用的 AI SEO 流程：意图映射、大纲冻结、辅助起草、On-page 打包、内链与更新节奏——不做黑帽，不编造实测。
中英双语静态站的信息架构、canonical、hreflang、sitemap 与 IndexNow/Search Console 提交清单，减少漏收与规范网址混乱。
把输入/输出 Token、重试、Agent 轮次和单价表拆开，按量计费才能算清；附可直接粘贴的估算步骤与上线检查清单。
把系统提示词当产品配置上线：职责边界、拒答规则、工具约定、版本与回滚，避免上线后行为漂移。
公众号用 AI 出初稿后的人工质检清单：事实、口径、合规、结构、SEO 与发布前终检，强调可勾选、可复用。
用 AI 辅助小红书初稿时，如何设计 brief、自动质检、双人审核与发布门禁，降低违规与虚假『实测』风险。
面向内容团队的可复制 Prompt 模板：Brief 扩写、大纲、分段起草、事实核对与 SEO 打包，强调人工把关与可复用。
给小团队的 LLM 费用控制手册：计量、预算、模型路由、压缩上下文、重试策略与月度估算，避免账单失控。
从选题到发布的可复用 AI 辅助流程：Brief、大纲、草稿、事实核对与 SEO 打包，强调人工把关。
面向实操者的 Prompt 清理方法与 Token/费用估算步骤，避免上线后账单失控。