2026년 1월 23일의 Show HN

Hey HN, I'm Marc. I run Numia, a blockchain data company serving real-time APIs at near-petabyte scale. ObsessionDB is managed ClickHouse we built for ourselves that costs roughly 50% less than other managed options. Now we're launching it as a standalone service.

We had the classic split: ClickHouse for real-time analytics APIs, BigQuery for data warehouse workloads. Data replicated between both, costs adding up. Managed ClickHouse worked but was too expensive to put our warehouse workload into. Self-hosted meant replica nodes doubling storage and waiting days to copy 25TB to new replicas.

What we really needed was separated storage and compute. SharedMergeTree does this, but it's proprietary to ClickHouse Cloud. So we built our own version. Migrated everything in, both the real-time APIs and warehouse workloads. No more replicating between systems. Been running it in production across 5 projects.

The main wins: compute scales without replication tax, separate endpoints for ingest/API/ad-hoc so they can't interfere, no ZooKeeper or replica management on your end.

Honest tradeoffs: we're early, EU only right now. Small workloads on a single node with local NVMe will be faster, the separation adds a bit of latency. If you're running something small and static, a self-hosted node is probably simpler. Where this shines is large datasets and elastic workloads where you don't want to babysit infra.

Check it out: https://obsessiondb.com, happy to go deep on architecture, merges, failover, whatever.

I started to use AI agents for coding and quickly ran into a frustrating limitation – there is no easy way to share my development environment logs with AI agents. So that's what is Teemux. A simple CLI program that aggregates logs, makes them available to you as a developer (in a pretty UI), and makes them available to your AI coding agents using MCP.

There is one implementation detail that I geek out about:

It is zero config and has built-in leader nomination for running the web server and MCP server. When you start one `teemux` instance, it starts web server, .. when you start second and third instances, they join the first server and start merging logs. If you were to kill the first instance, a new leader is nominated. This design allows to seamless add/remove nodes that share logs (a process that historically would have taken a central log aggregator).

A super quick demo:

npx teemux -- curl -N https://teemux.com/random-logs

Hi HN,

I’m the creator of C/C++ Cheatsheet — a modern, practical reference for both C and C++ developers. It includes concise snippet-style explanations of core language features, advanced topics like coroutines and constexpr, system programming sections, debugging tools, and useful project setups. You can explore it online at https://cppcheatsheet.com/.

I built this to help both beginners and experienced engineers quickly find clear examples and explanations without digging through fragmented blogs or outdated docs. It’s open source, regularly maintained, and contributions are welcome on GitHub.

If you’ve ever wanted a lightweight, example-focused guide to: - Modern C++ (templates, lambdas, concepts) - C fundamentals and memory handling - System programming - Debugging & profiling …this site aims to be that resource.

Any feedback is welcome. Thank you.

Hey all, Justin here. I previously built Phind, the AI search engine for developers.

One of the biggest problems we had there was figuring out what went wrong with bad searches. We had tons of searches per day, but less than 1% of users gave any explicit feedback. So we were either manually digging through searches or making general system improvements and hoping they helped.

This problem gets harder with agents. Traces are longer and more complex. It takes more effort to review them, so I'm building a tool that lets you analyze LLM outputs directly to help developers of LLM apps and agents understand where things are breaking and why.

I've put together a demo using browser-use agent traces (gpt-5): https://trails-red.vercel.app/viewer

It's early, but I have lots of ideas - live querying of past failures for currently-running agents, preference models to expand sparse signal data.

Would love feedback on the demo. Also if you're building agents and have 10k+ traces per day that you're not looking at but would like to, I'd love to talk.

Hi HN, I'm building Startups.in, a work-in-progress startup intelligence platform that aggregates global startup profiles with funding, sector, and location data in one place. It's still rough around the edges and in active development, but I wanted to share the public version for feedback.

This started as a personal project because I wanted a clean, searchable dataset of startups across regions without jumping between multiple sources or dealing with noise I didn't want :).

The product is still very much a work in progress, but it's in a usable state and open to feedback.

What it currently does: + Browse startup profiles with funding and basic company metadata + Search and filter by industry, geography, etc. + View simple ecosystem trends

No signup required to try it though you're welcome to sign-up to use watchlists etc.

How I built it: It's backed by a custom crawler (for data I need) and enrichment pipeline using n8n workflows, with a lightweight web UI focused on fast querying and filtering.

What I'm trying to learn from HN: + What data points would make this genuinely useful to you? + Would an API be valuable? + Does the UX get in the way of exploration?

I'm actively iterating on it and happy to discuss further. Thanks.

Hi HN, I've been building Thalo, a plain-text format for structured knowledge. It's designed to be human-readable and version-controlled, while giving tools and AI just enough structure to work with it. It's inspired by plain-text accounting tools such as Beancount.

The format is simple: you define your entities (e.g. opinions, book reviews, facts) including type definitions for metadata fields. Then you write entries that the CLI validates against your schema. The value is the feedback loop: LLMs can easily extract information from unstructured text, but they need validation to make the data high quality. Thalo's "compiler" provides these rules.

Example:

  2026-01-08T14:30Z create book-review "Designing Data-Intensive Applications" ^ddia #book 
    rating: "5" 
    author: "Martin Kleppmann"

    # Summary
    The definitive guide to distributed systems and data architecture. Dense but essential.

(In this example, ^ddia is a stable link identifier that can be used to reference this entry)

It's really up to the user how to use it. Some ways I'm using it:

- End of day journaling: blurt out thoughts, let AI extract insights, facts.

- Organizing my knowledge: using previous "content" I've written (blog posts, websites, my CV) to extract insights.

- Processing commits I've made to keep a living record of my work.

Because it's all plain-text, it allows you to open your knowledge base in any text editor or in Claude Code to do agentic search and analysis. I hooked up a simple Telegram bot to answer questions.

The tooling includes a CLI for validation, an LSP with completions and go-to-definition, a VS Code extension, and a Prettier plugin. There's also a simple scripting API that allows you to loop over entries and use the visitor pattern to traverse your knowledge base or write new rules.

I'd love to hear what use cases people come up with. Let me know what you think!

- Code (MIT): https://github.com/rejot-dev/thalo

- Landing Page: https://thalo.rejot.dev

I built this small tool for my own use to reduce the number of tokens I send to LLMs (Claude Code, etc.). It’s just a simple utility to filter command outputs before they hit the context.

Here is what I’m getting with it so far:

I’m putting it out there in case it's useful to anyone else. It's written in Rust.

P.S. This is just a tool I built for my own needs and decided to share. If you have constructive feedback on the Rust code or the logic, I'd love to hear it. If it's not for you, that's totally fine too—no need for "angry" comments, just trying to be helpful!

Hi HN, I built mermaidtui, a lightweight TypeScript engine that renders Mermaid flowcharts directly in your terminal as clean Unicode or ASCII boxes.

Visualizing Mermaid diagrams usually requires a heavy setup: a headless browser (Puppeteer/Playwright), SVG-to-image conversion, or a web preview. That's fine for documentation sites, but it's overkill for TUI apps, CI logs, or quick terminal previews.

The Solution is a small engine (<= 1000 LOC) that uses a deterministic grid-based layout to render diagrams using box-drawing characters. Key Features

- Intelligent Routing: It uses corner characters (┌, ┐, └, ┘) for orthogonal paths.

- Topological Layering: Attempts a readable, structured layout.

- Support for Chained Edges: A --> B --> C works out of the box.

- Zero Heavy Dependencies: No Mermaid internals, no Chromium, just pure TypeScript/JavaScript. With commander for the CLI, not the MermaidTUI library

I wanted a way to see high-quality diagrams in my CLI tools quickly, it’s great for SSH sessions where you can’t easily open an SVG. I was initially embedding this within a cli tool I’m working on and figured I’d extract out a library for others to use. I also initially used regex to parse, but now I made the parser a bit more robust. I'd love to hear your thoughts on the layout engine or any specific Mermaid syntax you'd like to see supported next!

GitHub: https://github.com/tariqshams/mermaidtui

npm i mermaidtui

We used Claude Agent SDK to make Claude Tutor. It's main goal is increase human knowledge, understanding, and agency.

It's an email and CLI agent to help people level up their software engineering skills.

We think there's too much focus on AI agency right now and not enough on human agency.

Open sourced, and curious for feedback! This is v0.1 so it's hella early.

Ps. Next step is to get this working on Open Agent SDK and explore other interfaces.

Specification: https://github.com/fedi-e2ee/public-key-directory-specificat...

I've been working on solving this problem so we can have end-to-end encryption for DMs on ActivityPub (and therefore Mastodon).

I thought I'd share it here in case anyone was interested in this work.

I kept copy-pasting terminal output into Claude Code. Built this instead.

Wake spawns your shell in a PTY, captures commands and output via shell hooks, stores everything in SQLite, and exposes it through an MCP server. Claude Code can then query your terminal history directly.

  - `wake shell` to start a session                                                                                                                                                                                                          
  - Work normally                                                                                                                                                                                                                            
  - Claude sees what happened                                                                                                                                                                                                                
                                                                                                                                                                                                                                             

Written in Rust. Zsh/bash supported. All data stays in ~/.wake/

Free AI creative suite. Upscale anime, manga & photos by 2x or 4x with Waifu2x. Generate stunning videos with Kling AI, Veo 3 & Sora 2. No download required.

After my annual checkup showed LDL 4.4 mmol/L (170 mg/dL) and triglycerides 2.0 mmol/L at 33, I tried tracking with ChatGPT (lost data when context got too big), then spreadsheets (too tedious).

Built a simple tracker focused on cholesterol. Log meals, see lipid breakdown, track trends. I believe snacks and sugar were my main issue.

Stack: Angular 17 + NestJS + Supabase

Started January 1st, already lost 3kg. Same breakfast daily (psyllium, oats, chia, skyr, whey, berries), cut sugar from daily to once per week.

Free during beta. Looking for feedback on whether strict diet cutting or 80/20 approach is more sustainable long-term.

I absolutely hate what the internet does to my brain, to my capacity for focus, and to how little I retain information.

I've been slowly migrating away from reading online to offline. Started a few years ago with a reMarkable tablet, then an Onyx Boox for colored papers. But newsletters always fell off my radar. There's just too much going on in my inbox.

So I built a tool that takes any RSS feed (Substack, Ghost, etc.) and formats it as a printable newspaper.

Tech: Single HTML file, vanilla JS, no backend. Fetches feeds via CORS proxy (Cloudflare Worker). Renders into a 3-column A4 layout.

I don't own a printer, though, haha. I just send the PDF to my reMarkable. But I'm curious if there's interest in getting a personalized newspaper like this actually delivered to your door.

Would love feedback on the layout, and curious how others handle the "too many newsletters" problem.

Hi HN,

I built CleanAF because my Windows Desktop kept turning into a dumping ground for downloads and screenshots.

CleanAF is a tiny one-click tool that:

keeps system icons intact

moves everything else into a timestamped “Current Desktop” folder

auto-sorts files by type

requires no install, no internet, no background service

It’s intentionally simple and does one thing only.

Source + download: https://github.com/TheZemmouri/CleanAF

I’m considering adding undo/restore, scheduling, and exclusion rules if people find it useful.

Feedback welcome.

In going against the trend of everything viral, I decided to make a little app to help discover some youtube video gems that show early promise by way of their "views to Likes" ratio. And combined with some keywords this throws up some interesting videos that haven't (yet) gainned any traction, but show early signs of doing so.My thinking is that such videos/channels validates interest and confirms high quality content thats relevant and therefore desirable.Of course this would logically extend to getting an algorithim boost from YT and therefore further exposure/likes etc ? I will caveat all of that with I have no idea how YT algo works though:) Anyways, appreciate any feedback on this and whether it could be useful to anyone who could make an angle from this.

E2B has been my favorite Sandbox environment for the AI agents I build. On actively using it for the last few months, CLI-ing every time is a lot of work. So here is a VSCode Extension