🛠️ All DevTools

Global Warming at 3 °C by 2050? What's Behind the New German Climate Warning

Data centers have hiked electricity prices on the public by $23B

Solving 20 Erdős Problems with 20 Codex Accounts Running in Parallel

Show HN: Tilion – Stealth Browser Infrastructure for Agents

Financing the AI boom: from cash flows to debt [pdf]

Show HN: An Interactive Guide for Castlevania II: Simon's Quest I&#x27;ll start by copying the comment I left on the Reddit post where I originally shared this project and then go into some more detail for this audience. (<a href="https:&#x2F;&#x2F;www.reddit.com&#x2F;r&#x2F;nes&#x2F;comments&#x2F;1umrnd5&#x2F;i_made_an_interactive_guide_for_castlevania_ii&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.reddit.com&#x2F;r&#x2F;nes&#x2F;comments&#x2F;1umrnd5&#x2F;i_made_an_inte...</a>)<p>&quot;&quot;&quot; Castlevania II has always held a special place in my heart. I didn’t own it myself as a kid but I had a family member who did. This meant that I only got a few chances to actually play it, but that left a big impression on me anyway. I recently beat it for the first time and along the way I decided I wanted to build the most over-the-top guide you could imagine as an ode to the game. This is that guide. I hope you all enjoy.<p>Source code: <a href="https:&#x2F;&#x2F;github.com&#x2F;baclap&#x2F;the-ultimate-simons-quest-guide" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;baclap&#x2F;the-ultimate-simons-quest-guide</a> &quot;&quot;&quot;<p>This is a Progressive Web App with full offline support. There are no static images or audio files used to create the guide. What it&#x27;s actually doing is taking real binary data extracted from the ROM at build time, processing it at runtime in WebAssembly, and then rendering it using WebGL2. Similarly the game music is played from ROM-derived data through a Web Audio AudioWorklet that recreates the NES audio channels. There are additional curated JSON files that describe how all this data fits together on the map, the text from NPCs, the secret information, and so on.<p>This was really just a labor of love and something I&#x27;m proud to share. Looking forward to your feedback. Thanks for taking a look!

Show HN: Flashbang – DuckDuckGo bangs resolved locally with a Service Worker I like to use DuckDuckGo-style bangs and snaps, they are fast and efficient shortcuts.<p>However, neither Kagi nor DuckDuckGo resolves them as quickly as I would like and subjectively Google has better search results than DuckDuckGo.<p>After trying a few local alternatives eg. unduck, unduckified, I wasn&#x27;t satisfied, the ones I tried briefly loaded a page before redirecting causing visible page flickering, still took time to resolve the actual redirect and lacked advanced features (address-bar autocomplete). Flashbang avoids that by handling the redirect in a Service Worker, before the browser renders anything. On my machine, the added ove rhead is around 0.14ms.<p>It has 14,470 bangs, custom shortcuts, address-bar suggestions, and works offline once installed (except for suggestions). No runtime dependencies.<p>Try it: <a href="https:&#x2F;&#x2F;flashbang-dyr.pages.dev" rel="nofollow">https:&#x2F;&#x2F;flashbang-dyr.pages.dev</a> Code: <a href="https:&#x2F;&#x2F;github.com&#x2F;ph1losof&#x2F;flashbang" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;ph1losof&#x2F;flashbang</a>

Show HN: Pullrun – Run One OCI Image as a Container, Firecracker, or Apple VM

Launch HN: Agnost AI (YC S26) – Extract user feedback from agent conversations Hey HN, we’re Shubham &amp; Parth, childhood friends building Agnost AI (<a href="https:&#x2F;&#x2F;agnost.ai">https:&#x2F;&#x2F;agnost.ai</a>), product analytics for teams building chat and voice agents.<p>We read production conversations and find behavioral failures like users rageprompting (cursing at the agent), repeatedly rephrasing the same request, correcting the agent, asking for missing features, or leaving after an answer that was technically successful.<p>We have an interactive demo with no signup here: <a href="https:&#x2F;&#x2F;app.agnost.ai?demo=true">https:&#x2F;&#x2F;app.agnost.ai?demo=true</a><p>Here&#x27;s a demo video: <a href="https:&#x2F;&#x2F;www.tella.tv&#x2F;video&#x2F;agnost-ai-launch-hn-demo-9haa" rel="nofollow">https:&#x2F;&#x2F;www.tella.tv&#x2F;video&#x2F;agnost-ai-launch-hn-demo-9haa</a><p>The core problem is that chat and voice products do not have the same metrics as web apps. When the product interface is language, clicks and funnels become much less useful. Users also rarely give explicit feedback, and when they do it&#x27;s usually sugarcoated. I barely type &#x2F;feedback in Claude or Codex myself. Most users just curse, ask again, correct the agent, or leave. So product engineers get technical visibility from latency, errors, and traces, but still have to guess whether users got what they wanted.<p>We got here after building around agents for the last year and got a couple of founders asking for something like a PostHog for conversations for the AI assistants they were building.<p>We are not trying to be in the observability or evals space. Observability tells you what happened technically. Evals validate cases you already know. We&#x27;re more on the discovery side like what users wanted, where they got frustrated, what they asked for repeatedly, and what new evals should exist.<p>Teams send us agent conversation messages through SDKs or OTel, optionally with metadata like account, plan, source, organization, etc. We cluster conversations into product-specific intents. Feature requests and bugs are default categories; most other clusters are created dynamically from the customer’s data and evolve over time. You can create your own cluster in plain English. If a cluster gets too broad, we split it. If a new pattern appears, we suggest it.<p>One AI video editor company used Agnost AI to find feature requests hidden inside chat. The biggest one was that around 70 users wanted auto-subtitles, but users said it as “add this text in this frame” 12x in a single session, “can you caption it”, “give me transcript of audio” and variations across languages. The team later built the feature.<p>Doing this over millions of messages without sending everything to an LLM was the hard part initially. In ClickHouse, “fetch the last 50 events by time across conversations” and “fetch all events in this conversation” want different sort orders, so we had to iterate a lot on sorting keys, partitions, materialized views, and projections.<p>For finding new clusters, sending everything through an LLM was too slow and expensive. HDBSCAN-style embedding clustering also gets painful at scale because of pairwise comparisons. We first split conversations into segments based on cosine drift, run BIRCH to compress the candidate space, and then use HDBSCAN-like clustering on the smaller set. For matching existing clusters, we use embeddings, smaller classifiers&#x2F;BERT-style models, and LLMs only as fallback for ambiguous cases.<p>We’re live with multiple companies and ingesting ~1M chat and voice messages per day. Pricing is public: Starter is free, Pro is $499&#x2F;month, and Enterprise is for higher volume, security, retention needs. We use each customer’s data only for that customer. We are SOC 2 Type 1 compliant, Type 2 is in progress, and our SDKs are on PyPI and npm.<p>We’d love feedback from the HN community and people building chat or voice agents: how do you detect these signals today, what feedback methods have worked, and what would block you from trying this? Happy to answer questions and take criticism.

👩🏿‍💻👨🏾‍💻👩🏼‍💻👨🏽‍💻👩🏻‍💻中国独立开发者项目列表 -- 分享大家都在做什么

Show HN: Opening lines of famous literary works This came from an idea that had been knocking around in my head for several years. I had been collecting opening lines of famous works and thought it would be cool to see one everyday as I opened the browser. I tried different styles but landed on the simple background with the text, let the words speak for themselves. Over time i&#x27;ve added more quotes I believe now there are close to 60, so hopefully you can refresh a few times and get a fresh one every time. I hope you guys like it, enjoy!

Show HN: Cascade Chat – A Hackable IRCv3 Client for macOS, Windows, and Linux Hello HN! I&#x27;m Matt and today I&#x27;d like to show you Cascade Chat.<p>One of my earliest internet experiences was with mIRC. I always admired its straightforward, pleasant UI and the way it wove a hackable core into the code. The way you could build on the visual and API layers of the underlying IRC client to me was fascinating software machinery. It was truly a client that you could build on top of.<p>As my career has progressed, I moved away from Windows and adopted Linux as my daily driver. That was where I found HexChat, the closest thing I could find to mIRC many years later. While I really enjoyed HexChat, it wasn&#x27;t quite what mIRC offered. I eventually found myself on macOS with no clear analogue to either. That&#x27;s why I built Cascade Chat.<p>Cascade is a modern IRCv3 client. It supports persistent local history with full-text search, network management, replies, typing indicators, link previews, pinned messages, native notifications, SASL authentication, server-time, chathistory, account and away tracking, and the ratified IRCv3 capability set.<p>I also wanted Cascade to be hackable, so I built in two fundamentally different layers:<p>- Scripting powered by Go scripts for personal automation, event handlers, and timers. Scripts run in-process with no access to the standard library, filesystem, or network. - Plugins that communicate over JSON-RPC and can be written in any language as external processes.<p>Stack-wise, it&#x27;s a Go application built on top of Wails v3, which leverages the OS&#x27;s native WebView to render modern web tech frontends as desktop applications. The result is an Electron-like experience without packaging a separate Chromium runtime.<p>Full disclosure, since this is HN: I built Cascade using agentic engineering. I made the product and UX decisions, designed the architecture and code interfaces, and I reviewed and dogfooded the resulting work. Coding agents implemented much of the space between those decisions.<p>To further ensure consistent quality, I built gates around that process rather than treating generated code as finished. I focused on unit and integration tests, full-stack end-to-end tests against a real IRCv3 server (Ergo), automated release candidates, and regular dogfooding of the prerelease channel.<p>Cascade is open source under the BSD 3-Clause license. Prebuilt packages are available for macOS, Windows, and Linux. The current builds are unsigned, so macOS and Windows require a first-run confirmation step that I&#x27;ve documented in the README.<p>I&#x27;d love to know if you&#x27;d make Cascade your daily IRC client, and if not, what that&#x27;d take! Feedback and PRs welcome.

[Other] Show HN: I built a smart proxy so your coding agent can run loose The only way to go fast is full YOLO mode in your coding agent. I&#x27;ve got the local sandbox figured out (pro tip: Incus VMs work great) but I wanted to keep my agents from doing things like inadvertently blowing up my cloud services or chasing a prompt to POST to some random website. I struggle most with this on my side projects where my permission model isn&#x27;t quite as robust as it is at the office.<p>I started with a firewall on the Incus container but every time the agent needed access to something new, I was poking more holes in it - and it didn&#x27;t differentiate between HTTP verbs.<p>I&#x27;ve been using Claude Code auto mode too but I wanted to be able to guide the LLM, not just hand over control completely. I also want some flat allow and deny rules.<p>So, I built Trollbridge: an HTTP&#x2F;S proxy that runs in front of your coding agent and lets you set up allow or deny lists, realtime approve or deny in the TUI, or wire up an LLM that makes calls based on your prior decisions.<p>Would love to hear what you think - give it a shot, leave a comment, file a bug. Thanks!<p>Website: <a href="https:&#x2F;&#x2F;trollbridge.dev" rel="nofollow">https:&#x2F;&#x2F;trollbridge.dev</a> Github: <a href="https:&#x2F;&#x2F;github.com&#x2F;dandriscoll&#x2F;trollbridge" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;dandriscoll&#x2F;trollbridge</a>

[Other] Show HN: Low-latency local LLM runner via OpenJDK Panama FFM (Java 22) I wanted to run AI from inside the JVM. I started out with the standard REST sidecar, ripped that out to use Project Panama (Foreign Function &amp; Memory API) in the new JDK versions to interface directly with llama.cpp. I still wasn&#x27;t happy with how that functioned, so I built libargus.cc to get a clean ABI to expose a structured API up in the JVM landscape. It still uses Project Panama to interface directly with llama.cpp, whisper.cpp, and ggml compute graphs.<p>I have zero-allocation on the hot paths, memory segments for prompts and tokens are allocated once inside confined Arenas. Raw pointers pass straight through down to the low C level. This avoids primitive array cloning and heap churn.<p>I mapped out the native structures from llama.cpp and whisper.cpp while matching the compiler&#x27;s padding to maintain safe memory access.<p>I bundle pre-compiled native binaries in the jar for easy deployment.<p>This execution engine provides the foundation I need for work I&#x27;m doing on a spatio-temporal memory layer (L-TABB) to replace RAGs. I&#x27;d love to get technical feedback to polish any issues while I continue working on the next layer. Deep-dives from anyone hacking on Project Panama or low-latency systems in modern JDK would be very appreciated!<p>I&#x27;m much better with code than prose, so I&#x27;ll let the code do most of the talking.<p>Happy Hacking! &#x2F;David<p>Code: <a href="https:&#x2F;&#x2F;libargus.cc" rel="nofollow">https:&#x2F;&#x2F;libargus.cc</a> Project Landing Page: <a href="https:&#x2F;&#x2F;projectargus.cc" rel="nofollow">https:&#x2F;&#x2F;projectargus.cc</a>

Show HN: Oodle.ai – $10 per million agent traces Hi HN, we&#x27;re Kiran and Vijay!<p>Over the past two years, we have built a columnar storage engine for observability: logs, metrics, and traces. Today, it&#x27;s exciting for us to show what we&#x27;ve built on top of that foundation: LLM Agent Observability.<p>Given how non-deterministic agents are, storing all traces without sampling was critical for us. But these traces tend to be in the MBs, sometimes GBs - we needed to store them inexpensively. We also needed the queries and analyses to be fast. To meet both these goals, we store them in S3 in our own parquet-like file format, and query them using AWS Lambda.<p>Since we process each span of every trace, instead of running LLM-based evals on each, we first analyze them using deterministic techniques. We detect tool failures, retries, loops, abnormal token usage, latency regressions, schema violations, sentiment, and other production signals. We&#x27;ve written more about the approach here: <a href="https:&#x2F;&#x2F;blog.oodle.ai&#x2F;you-cant-sample-your-way-to-reliable-agents&#x2F;" rel="nofollow">https:&#x2F;&#x2F;blog.oodle.ai&#x2F;you-cant-sample-your-way-to-reliable-a...</a><p>The combination of our own engine, no sampling, and deterministic processing before LLM-for-evals allows us to price at $10 per million traces, provide sub-second p99 query latency, and have healthy margins. Before building this, we used Langfuse for our own agent observability, which was 6x more expensive.<p>Still super early, and rough around some edges, we would love your questions and feedback!

The Conservationist Who Turned 40 Terabytes of Public Data into a Video Game

Show HN: Town – Discord in a pixel town where the NPCs have skills I built this thing called Town. Think Discord, but you walk around a pixel town where the NPCs have skills.<p>Agents have become my initial brainstorm partner for a lot of stuff — sub-agents to counter my own thinking, role-playing a specific person, grounding them in a specific knowledge so they refer instead of guess. I started turning my claude-code into this pattern and it worked well enough that I wanted to see if I turn into a fun shareable thing where other people can also talk to.<p>It&#x27;s a small pixel map of buildings. You walk in, step into any building, and talk to the npcs who live there.<p>Each NPC is an AI agent with its own personality (voice, backstory, what they care about) and a skillset (tools like memory, web search, writing docs). The personality makes them fun to talk to; the tools make them able to actually do something.<p>Multiplayer&#x2F;Group discussions: Real people wander the same map. If someone else is in your town you see them walking around and can chat live. Step into a building together and press G to open a shared chat between the humans and AI avatars in that room.<p>The engineering bit I like most: each NPC is a separate agent with scoped tool permissions. The author decides which tools each one gets — a shopkeeper can hand out inventory, a detective can issue a Case Closed card, a barista just talks.<p>A few towns I&#x27;ve built:<p>- AI Startup Town: get in a room with NPCs of famous investors — Paul Graham, Garry Tan, Michael Seibel, Dalton Caldwell. Pitch your idea, stress-test your plan, build your startup doc.<p>- Murder Mystery Town: Iris Bell is dead. Talk to the NPCs, catch someone in a contradiction, figure out who did it, and report back to Detective Reeve to close the case.<p>- Roast Town: get roasted, or learn the fundamentals of roasting.<p>- CORE Town: the sampler — one town that shows off everything. Pitch Garry Tan at YC House, stand trial at Trial of You, get worked over by a noir detective, get roasted. Best place to start.<p>You can build your own town too, and it&#x27;s open source. A town is just a JSON + MDX folder — a buildings list, one .mdx per AI avatar (frontmatter = identity, body = system prompt) — then town deploy from the CLI. There&#x27;s also a claude plugin that builds the whole thing in one shot from a rough description.<p>Guest mode, no signup:<p>Startup: <a href="https:&#x2F;&#x2F;town.getcore.me&#x2F;startup?invite_code=7EQMH0">https:&#x2F;&#x2F;town.getcore.me&#x2F;startup?invite_code=7EQMH0</a><p>Murder Mystery: <a href="https:&#x2F;&#x2F;town.getcore.me&#x2F;murder-mystery?invite_code=S85S24">https:&#x2F;&#x2F;town.getcore.me&#x2F;murder-mystery?invite_code=S85S24</a><p>Roast: <a href="https:&#x2F;&#x2F;town.getcore.me&#x2F;roast-town?invite_code=0MXN93">https:&#x2F;&#x2F;town.getcore.me&#x2F;roast-town?invite_code=0MXN93</a><p>CORE Town: <a href="https:&#x2F;&#x2F;town.getcore.me&#x2F;core-town?invite_code=H4C0TZ">https:&#x2F;&#x2F;town.getcore.me&#x2F;core-town?invite_code=H4C0TZ</a><p>Full list: <a href="https:&#x2F;&#x2F;town.getcore.me&#x2F;explore">https:&#x2F;&#x2F;town.getcore.me&#x2F;explore</a><p>Source: <a href="https:&#x2F;&#x2F;github.com&#x2F;RedPlanetHQ&#x2F;town" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;RedPlanetHQ&#x2F;town</a><p>Rather fun project overall. Honestly I have been thinking a lot about how community engagement is changing with agents and this felt like a good way. It&#x27;s a very raw shot. Kindly share your thoughts and feedback.

Show HN: Beautiful Type Erasure with C++26 Reflection Try it on Compiler Explorer: <a href="https:&#x2F;&#x2F;godbolt.org&#x2F;z&#x2F;91dj5jeGW" rel="nofollow">https:&#x2F;&#x2F;godbolt.org&#x2F;z&#x2F;91dj5jeGW</a><p>Check out the source code: <a href="https:&#x2F;&#x2F;github.com&#x2F;RyanJK5&#x2F;rjk-duck" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;RyanJK5&#x2F;rjk-duck</a>

Coding agents think ahead of time

面向 Grok Build、Grok Web 与 Grok Console 的多账号 API 网关