慣性聚合 高效追讀感興趣之博客、新聞、科技資訊
閱原文 以慣性聚合開啟

推薦訂閱源

aimingoo的专栏
aimingoo的专栏
Y
Y Combinator Blog
云风的 BLOG
云风的 BLOG
Microsoft Azure Blog
Microsoft Azure Blog
腾讯CDC
T
The Blog of Author Tim Ferriss
P
Proofpoint News Feed
Hugging Face - Blog
Hugging Face - Blog
博客园_首页
小众软件
小众软件
美团技术团队
Martin Fowler
Martin Fowler
爱范儿
爱范儿
有赞技术团队
有赞技术团队
博客园 - 【当耐特】
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
Microsoft Security Blog
Microsoft Security Blog
宝玉的分享
宝玉的分享
J
Java Code Geeks
B
Blog
V
V2EX
Stack Overflow Blog
Stack Overflow Blog
B
Blog RSS Feed
博客园 - Franky

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
I tried to build a SaaS. I'm shipping tiny libraries inst...
nuan xiang · 2026-05-24 · via DEV Community

nuan xiang

I tried to build a SaaS. I'm shipping tiny libraries instead.

For 7 days I poured energy into ChatProof — a testing framework for AI chat UIs. The problem was real: every team building Claude/GPT wrappers hits the same streaming-render bugs, and existing e2e tools weren't built for non-deterministic LLM output. I had a competitive analysis. I had a PRD. I even had a name.

Then I admitted something to myself: product/market work just isn't my strength.

I'm a builder. I love writing TypeScript. I love designing APIs. I love npm test going green. I do not love writing cold DMs. I do not love figuring out which subreddit my ICPs hang out in. I do not love crafting positioning documents.

Once I named it, the pivot was obvious.

The pivot

Instead of one ambitious SaaS, I'd ship tiny zero-deps libraries that solve specific Anthropic SDK pain points I personally hit. Each one:

  • < 500 lines of TypeScript
  • Zero runtime dependencies
  • MIT licensed
  • Published to npm
  • 1–2 weeks of work max

The goal isn't to build a unicorn. It's to leave a GitHub trail that demonstrates I can ship real code, get organic distribution through npm + search, and let the work compound over months.

The 5 days

Day 1 — claude-stream-collector

Consume Anthropic Messages streaming events into a typed CollectedResult. Handles tool_use input JSON delta accumulation (the SDK leaves this to you) and merges cache token usage across message_start + message_delta (the SDK overwrites). ~120 lines, 6 tests.

Day 3 — claude-retry

Smart retry for Anthropic API calls. Respects retry-after headers (which the SDK's built-in retry doesn't always honor), exponential backoff with jitter, custom retry predicates, AbortSignal support. ~180 lines, 25 tests.

Day 5 — claude-pricing

Calculate Claude API cost from token usage. Bundled pricing table with smart model-alias resolution (strips date suffixes), batch API discount, session CostTracker with budget alerts. ~180 lines, 19 tests.

Three packages, ~480 lines of code, 50 unit tests. All public. All on npm.

What I'm learning

1. Match the work to who you are, not who you think you should be.

Indie Hackers and Twitter are full of advice assuming you're a generalist who loves every part of building a company. Many great engineers aren't, and that's fine. Tiny libraries are a perfectly valid path if you optimize for shipping over selling.

2. Zero-deps is a feature, not a constraint.

Every dependency is a future maintenance burden, a potential security issue, and a reason for someone to skip your package. For 100–500 line utilities, you almost never need anything but Node's standard lib.

3. Series narratives compound.

Each package on its own is a small thing. But three packages with shared types (TokenUsage flows from claude-stream-collectorclaude-pricing cleanly) form a tiny ecosystem. Each new package strengthens the others.

4. Build in public ≠ build for engagement.

I posted about each release on X and 即刻. Engagement so far: zero likes, zero replies. Downloads on the first package: 129 in 7 days. People install packages they discover via search and never engage with the announcement. That's fine — the artifacts persist longer than any tweet.

5. Pivoting hurts less when you preserve learning.

ChatProof isn't dead. The painpoint research, competitive analysis, and architecture thinking from those 7 days informed every package since. Especially: I knew the streaming pain because I'd done the homework.

What's next

More small packages: claude-cache-helper, claude-tools-typed, maybe claude-prompt-toolkit. Same shape — small, focused, zero-deps. I'll keep posting about them but won't measure success in likes.

If you're stuck on an indie project trying to be a "real founder," maybe try shipping a single utility you'd actually use. You might find you ship four more before you'd have written a single landing page.


Three packages:

Building in public: ship-log-2026