@serafimcloud — 730+ shadcn/ui-style production components in one library
I spent 70 days full-time curating the ultimate library of @shadcn/ui-like components.
And today, I'm launching it publicly.
Here's what it is: • 730+ production-ready components from 50+ top design engineers • Each component is yours to own - just like shadcn/ui • Install
That's a serious curation effort. The "yours to own" model is what makes shadcn/ui work in the first place — you get source code, not a black-box dependency you can't touch. 730 components across 50 contributors means the variance in quality is probably high, but if even 10% are solid and save you from building the same date picker a fifth time, it's worth exploring.
@LangChain — Khoj, a self-hostable AI second brain
Khoj - your AI second brain
Self-hostable. Get answers from the web or your docs. Build custom agents, schedule automations, do deep research.
Turn any online or local LLM into your personal, autonomous AI (gpt, claude, qwen, mistral)
https://github.com/khoj-ai/khoj
khoj-ai/khoj hits a specific niche: the people who want the Perplexity or NotebookLM experience but aren't willing to feed their private docs to a third-party API. Self-hosted, runs against any LLM backend (including local ones via Ollama), and supports scheduled automations. The deep research mode is the interesting part — it's not just RAG over your notes, it chains searches and reasoning steps. Worth a closer look if you're building internal knowledge tools or just tired of re-finding the same Notion pages.
@rauchg — smoothStream({ chunking: 'line' }) in the AI SDK
smoothStream({ chunking: 'line' })
Short post, but the API surface it's announcing is worth noting. Vercel's AI SDK added smoothStream with a chunking: 'line' option — instead of forwarding raw token chunks (which causes that jittery character-by-character streaming effect), it buffers until a full line is ready and then flushes. The result is much smoother for code output specifically. One function call, cleaner UX. That's the kind of wrapper that used to require 30 lines of manual buffering logic.
@nutlope — building something from Anthropic's agent article
Working on something new based on Anthropic's great agent article!
No details yet — just a teaser with a screenshot — but Hassan's track record of shipping fast and in public makes this worth watching. The Anthropic agents article it references is the one on building effective agents with clear workflow patterns (prompt chaining, routing, parallelization, orchestrators). If he's building a working demo around those patterns, it'll be a useful reference point.
@LangChain — Tavily Company Researcher on LangGraph
Tavily Company Researcher
An open-source AI tool that delivers real-time company research reports with unmatched accuracy. Built on LangGraph, it combines intelligent search and structured workflows to transform complex company research into automated, reliable insights.
LangGraph as the orchestration layer for structured research tasks makes sense — the graph-based state machine approach handles branching search strategies better than a linear chain. "Unmatched accuracy" is marketing, but the architecture idea is solid: you define a workflow graph, plug in Tavily for search, and get a structured report out. Useful template for anyone building internal research pipelines.
@marclou — what customers actually care about
Your customers don't care about:
❌ SQL vs. NoSQL ❌ The trendy UI library ❌ The number of React hooks you use
Your customers care about:
✅ Your headline (is it worth my time?) ✅ The problem you solve (what do I get?) ✅ Your pricing table (is it worth my money?)
A reminder that hits differently at the start of a new year when everyone's planning what to build next. The list of things customers don't care about reads like a retrospective of every bikeshedding thread I've sat through. The flip side: if your headline is weak, your pricing is confusing, or the problem statement is buried — you're losing before anyone runs your code.
