Caderno
Notas curtas e práticas sobre ferramentas de IA, modelos e como manter os gastos sob controle.
-
用人工智能起草求职信，把事实和预算留在自己手里
从拆解岗位要求、整理真实经历，到限制生成篇幅、核对事实和人工定稿，逐步写出可信、具体又不过度花费的求职信。
-
为学习小组搭建一个按需回答的 Discord 助手
从创建机器人、设置权限到接入模型和检查费用，逐步搭建一个只在被召唤时回答、帮助同学梳理解题思路的学习助手。
-
一步步为 Python 库生成可靠的单元测试
从一个金额计算函数开始，整理行为约定、生成测试清单、检查断言并验证测试效果，用小批量请求控制生成单元测试的成本。
-
把收藏夹变成能找回答案的个人片段搜索工具
从整理少量代码与笔记开始，逐步搭建个人片段搜索工具，用本地检索缩小范围，再按需调用模型解释结果，让搜索更好用，也让花费更可控。
-
把日记留在自己手里：逐步搭建每日摘要工具
从本地筛选、敏感信息替换到限额调用和人工核对，逐步搭建一个尽量少上传、少花钱、不过度解读的日记摘要工具。
-
从代码注释写出可用文档：一次小范围的整理实践
从一个函数开始，核对注释与实现，整理参数、边界和示例，再借助模型生成并验证文档，用较少的请求完成一次可靠的整理。
-
把复习资料做成闪卡：一步步搭建备考生成器
从一小节讲义开始，搭建能生成、校对、导出和安排复习的闪卡工具，并用明确的请求上限控制模型开销。
-
让人工智能帮你清洗杂乱表格，又不花冤枉钱
从保留原件、抽样诊断到制定规则、生成清洗程序与核对结果，逐步清理杂乱的逗号分隔值文件，让人工智能协助判断，让本地程序完成重复工作。
-
把会议记录自动变成可执行行动项
用固定输入、清晰提问和低成本模型，将杂乱会议记录整理为负责人明确、日期可追踪的行动项。
-
用课程笔记搭建一个小型检索问答演示
从整理笔记、切分段落到检索与核对引用，用本地检索减少模型调用，搭建一个预算可控的课程问答演示。
-
Automate Your Weekly Report Draft with a Script
Build a small script that turns weekly notes into a reviewable report draft, with practical checks to keep errors and spending under control.
-
Summarising Long PDFs on a Small Budget
A practical walkthrough for extracting PDF text, testing a low-cost workflow, and building a checked summary without repeated full-document requests.
-
Build an AI Telegram Bot Without Overspending
Build a personal AI Telegram bot step by step, with secure credentials, short conversations, spending limits, and practical testing.
-
A Command-Line Translator in Fifty Lines
Build a small terminal translator step by step, with clear inputs, predictable output, and practical controls for a student-sized budget.
-
Build a Study Assistant Without Overspending
Build a personal study assistant with a chat API, grounded answers, useful quizzes, and a budget you can control.
-
Call Several AI Providers Through One Endpoint
Connect once, test models from different providers, and track each request without turning a small project into a large bill.
-
Send Images to Vision Models, Step by Step
Learn how to prepare images, build a vision API request, check the response, and control spending through MyRelay.
-
Debug Request Logs Without Leaking Secrets
A practical walkthrough for logging AI requests safely, tracing failures, and spotting retries that quietly increase your spending.
-
Concurrency Limits: A Practical Walkthrough
Understand why concurrency limits exist, then build a simple queue, handle retries, and keep parallel AI requests within your budget.
-
Retry Flaky AI Requests Without Wasting Your Budget
A step-by-step walkthrough for handling network failures with bounded retries, sensible timeouts, backoff, and spending checks.
-
Count Tokens Before You Send a Request
A practical walkthrough for estimating input tokens, reserving output room, and checking request costs before spending your API budget.
-
Function Calling Through a Relay Endpoint
Build a small function calling loop through MyRelay, validate tool requests, return results, and keep token costs predictable.
-
Use the OpenAI Node.js SDK with a Custom Endpoint
Connect the OpenAI Node.js SDK to MyRelay, verify your first request, and keep retries, tokens, and testing costs under control.
-
Keep API Keys Out of Your Git History
A practical walkthrough for storing API keys safely, checking commits, and handling leaks without putting your development budget at risk.
-
Streaming AI Responses with Server-Sent Events
Build an SSE response flow step by step, handle interruptions, and track token costs without making your first streaming project expensive.
-
Handle Rate Limits Without Wasting Your Budget
A step-by-step walkthrough for pacing AI requests, retrying safely, saving progress, and keeping script costs predictable.
-
Your First API Call with cURL, Explained
Make your first AI API request with cURL, understand what each part does, and check the result without wasting your budget.
-
Hidden Image and File Costs: A Spending Walkthrough
Follow a practical walkthrough to uncover image and file input costs, trim unnecessary content, and measure spending before scaling.
-
Keep Your Side Project Under a Coffee a Day
A practical walkthrough for setting limits, choosing models, tracking usage, and keeping AI spending predictable with MyRelay.
-
When an AI Pass Beats Paying Per Token
A step-by-step way to compare an AI pass with token billing before you spend more than planned.
-
Understanding a Pass Multiplier on Your Monthly Quota
A practical walkthrough for estimating how a pass multiplier affects your monthly AI budget and planning usage calmly.
-
Track Project Spend with Separate API Keys
A practical way to separate AI spending by project, set limits, and review costs before a small budget becomes a surprise.
-
Cheap First, Expensive Second: A Model Cascade
Build a model cascade that handles routine requests cheaply and escalates difficult work while keeping quality and total spending visible.
-
How to Plan Around Weekly Pass Caps
A practical walkthrough for managing weekly passes, choosing models, and keeping AI spending predictable with MyRelay.
-
How a Small Overdraft Floor Limits Surprise Bills
A practical walkthrough of using a small negative-balance limit, request costs, and a personal budget to keep AI spending predictable.
-
The Real Cost of Long Chat Histories
A practical walkthrough for understanding, measuring, and controlling the cost of long AI chat histories.
-
How Cache Reads Are Billed and Why They Are Cheap
Follow a request from prompt to bill, learn why cached input costs less, and check whether caching is reducing your API spending.
-
提示词缓存如何降低账单：从整理输入到核对费用
按照挑选重复内容、整理提示词、确认缓存规则和核对账单的顺序，理解提示词缓存如何省钱，并用小规模测试判断是否值得开启。
-
按量付费还是月度通行证：一步步选出适合自己的方案
从使用频率、实际账单和权益限制出发，逐步比较按量付费与月度通行证，帮助个人开发者和学生找到更符合预算的选择。
-
开始写代码前，怎样估算这一轮会花多少钱
从划定任务、估算上下文和调用次数，到计算费用、预留余量与设置停止条件，一步步为个人编程会话做出可执行的预算。
-
个人开发者如何一步步设定每月人工智能预算
从可承受的月度上限出发，记录真实请求成本，拆分日常与疑难任务额度，并建立每周检查和接近上限时的处理规则。
-
令牌价格怎么算：从一次请求算到每月预算
用一组假设价格，逐步算清输入、输出、多轮对话和每月预算，再通过实际账单找出值得节省的部分。
-
论文文献综述提纲怎么选模型：从小样本试跑到预算内定稿
先整理可信文献，再用小样本比较模型，按结构质量与实际费用做选择。一步步完成可追溯的综述提纲，减少重复调用和返工。
-
把同一份缺陷报告交给三个模型，怎样测得公平又省钱
用搜索结果错位的缺陷练习三模型对比：固定报告、限制预算、统一验证，再按修复质量与总成本选择适合自己的模型。
-
推理模型与快速模型：一步步算清任务成本
从任务分类、小样本测试到失败升级，逐步比较推理模型与快速模型的费用、延迟和返工成本，让个人开发者与学生把预算花在真正需要的地方。
-
分类任务如何选模型：从小模型起步，按错误决定是否升级
用一套可复现的评测流程，比较小模型与大模型的分类质量、实际成本和处理速度，再决定是否采用分级处理。
-
Fable vs Opus: When the Premium Model Is Worth It
A practical walkthrough for testing Fable against Opus, measuring total cost, and deciding when a premium model earns its place.
-
初学者学数据库查询，怎样选到讲得清又省钱的模型
用一组小练习比较模型的查询正确性、解释能力和实际费用，逐步找到适合自己学习数据库查询的助手，避免为错误答案反复付费。
-
A Simple Rubric for Comparing Model Answers
Compare model answers with a small, repeatable rubric that checks correctness, usefulness, and cost before you spend more.
-
Compare Context Windows Without Wasting Tokens
A step-by-step way to compare model context limits, test useful recall, and choose enough room without paying for unnecessary tokens.
-
Choose a Model for Translating Documentation
A practical walkthrough for testing translation quality, preserving technical details, and keeping documentation costs predictable.
-
Pick the Right Model for Your Hackathon Weekend
A practical walkthrough for testing models, setting a spending limit, and shipping a hackathon demo without burning your budget.
-
GLM and MiniMax: Two Options Worth Testing
A budget-conscious walkthrough for testing GLM-5.3 and MiniMax M3 on real tasks, comparing results, and deciding when to use them.
-
A Practical Qwen Workflow for Multilingual Writing
Build a repeatable Qwen writing workflow with clear briefs, small tests, locale checks, and request-level cost tracking.
-
Grok for Brainstorming: A Practical Walkthrough
Use Grok to explore ideas, challenge assumptions, and choose a small experiment without letting brainstorming consume your budget.
-
DeepSeek V4 Reasoning on a Small Budget
A step-by-step walkthrough for testing DeepSeek V4, controlling token use, and checking reasoning results without stretching your budget.
-
Which Model Should Summarise Your Lecture Notes?
A practical walkthrough for choosing a model, testing summary quality and keeping lecture-note summarisation costs under control.
-
Gemini Flash for Fast Drafts and Quick Answers
A step-by-step walkthrough for using Gemini Flash to draft, ask focused questions, and manage your AI spending with MyRelay.
-
Haiku for Cheap Background Tasks: A Practical Walkthrough
Build a budget-conscious background workflow with Claude Haiku 4.5, clear task boundaries, validation, and request-level cost tracking.
-
GPT-6.1 Sol vs GPT-6 Astra: Pick One for Coding
A budget-conscious walkthrough for comparing GPT-6.1 Sol and GPT-6 Astra on real coding tasks before choosing a daily model.