treni-docs.pages.dev
Show HN: Treni – single-binary GPU runtime for uncertainty-aware agents 5ms TTFThttps://treni-docs.pages.dev/docs/We built Treni, a C/CUDA runtime where routing, tokenization, tool models, and state run in one GPU process.Most agent stacks only get serialized tool strings back. Treni exposes execution signals in-process (entropy/logprobs/retrieval distance/route confidence), so the agent can branch before committing bad output.Canonical A10G (G5), token-parity vs vLLM (max_tokens=48):TTFT: 5.130 ms (Treni) vs 84.837 ms (vLLM) -> 16.537x Full request: 316.403 ms vs 1232.660 ms -> 3.896x Cold total first response: 1320.240 ms vs 28937.430 ms -> 21.918x Steady state:Warm mean: 80.602 ms Warm p99: 90.350 ms Additional checks:Frontend A/B repeatability (warm_fixed + mixed_churn, repeats=3): custom path wins all tracked metrics Numerical parity vs PyTorch (strict mode): 0 failures Separate OpenAI routing-overhead test (different question, not engine-vs-engine):Same model endpoint on both sides (gpt-5.2) Internal path: client -> OpenAI External path: client -> controller/tool hop -> same OpenAI endpoint Feb 23, 2026 12:12 AM
github.com
Show HN: I'm building a pure-Rust reimplementation of rsync (Protocol 32)https://github.com/oferchen/rsyncYears ago, I was tasked with a massive data migration: multiple disks, each containing over 100 million files, with a strict, non-negotiable 24-hour downtime window. Using the standard tools available at the time was a painful experience. The single-threaded file discovery crawled, and memory usage was a constant source of anxiety. I promised myself that one day, I would come back and build a tool that could actually handle that scale natively.What started as a side project has evolved into a full systems-level undertaking. The project is oc-rsync - a complete client, server, and daemon implementation targeting rsync protocol 32, written entirely in pure Rust.I find it incredibly ironic that I am currently shipping a data migration tool while my life is packed in suitcases, as I'm migrating to another country myself. I’ve been pushing git commits multiple times a day between packing boxes.I want to be completely transparent upfront: I am actively working on this, and not everything is Feb 22, 2026 10:51 PM
news.ycombinator.com
Show HN: I tracked 800 Reddit URLs on Google – 56% die within 48hhttps://news.ycombinator.com/item?id=47114771I've been experimenting with parasite SEO (ranking Reddit/Quora threads on Google) and noticed my traffic kept dying after a few days.To see if this was systemic, I built a custom scraper and tracked 800 unique UGC URLs across high-competition commercial keywords for a month.The data showed a massive "churn" rate: Out of the 611 threads that dropped out of the Top 10, 56.1% were completely rotated out of the SERPs within 48 hours. Only a tiny fraction survived more than 2 weeks.I published the raw CSV data from the study here if anyone wants to run the URLs through Ahrefs and look for backlink patterns: https://contentrankr.com/ugc-churn-dataStandard rank trackers check weekly, so you don't even realize a thread is dead until days later. I also embedded a tiny free tool on that page using the Serper API where you can paste a Reddit URL and instantly check if Google has dropped it from the index.Curious if anyone else has looked into the "half-life" of Google's new UGC rankings?Feb 22, 2026 9:18 PM
news.ycombinator.com
Seeking a Front End Engineer Rolehttps://news.ycombinator.com/item?id=47101213Location: India Remote: Yes!, but can do on-site/hybrid in Pune Role: Frontend/Software/Founding engineer full-time Willing to relocate: Yes, but want visa sponsorship. Technologies: Languages & Frameworks: TypeScript, JavaScript, React, Next.js, Node.js Databases & State: MongoDB, SQL (PostgreSQL/MySQL), Redux Toolkit, React Context API UI & Frontend: Material-UI (MUI), Tailwind CSS, SCSS, HTML5/CSS3, Responsive UI/UX Design DevOps & Tools: Docker, Keycloak (Auth), Jest, gRPC, REST APIs, Git, GitHub, Agile (Scrum) AI & LLM Ops: Local LLM Deployment (Ollama, vLLM), Model Quantization, Private Cloud AI Infrastructure, Stable Diffusion Résumé/CV: https://docs.google.com/document/d/1sd7iwTgXVG2EwuOHnNxCx-MTuMhUi_Vjgfo6FCp1rR4/edit?usp=sharing Email: gauravsharma.mern@gmail.com Github: https://github.com/GauravFrontend I’m a Frontend Engineer with 2 years of experience, I work daily with React, TypeScript, Next.js, Node.js and have shipped production features.here are some things i am prouFeb 21, 2026 2:35 PM
news.ycombinator.com
Ask HN: How to find a Sales cofounder for a B2B SaaS fintech compliance product?https://news.ycombinator.com/item?id=47100819Context: I’m a technical founder building Settl X, a layer that helps compliance teams make fast, repeatable, audit-ready decisions across a fragmented toolchain (KYB/KYC, sanctions/PEP, transaction monitoring, case tools, investigations). Our initial wedge is an "audit-ready KYB decision packet" that bundles evidence, rationale, and a defensible decision trail.Why we’re doing this: In cross-border / multi-jurisdiction flows (including stablecoin rails), compliance becomes the throughput bottleneck. Teams spend a lot of time on the “evidence trail tax” (collecting artifacts, timelines, and rationale for internal review + partners + audits), and decisions aren't always reproducible across analysts.Operating model we're aiming for: ~80% AI work / ~20% human approval — AI assembles the dossier (evidence + rationale + recommended action), then an analyst/manager reviews and approves with a complete audit trail.What we’ve built: MVP internally. No public sandbox available yet. If helpful, hFeb 21, 2026 1:47 PM
wordy.info
Show HN: Wordy – Learn languages from real movie and TV clips with quizzeshttps://wordy.info/Hey HN, I'm Sándor. I built Wordy because I kept pausing Netflix to look up words in a dictionary. So I made a subtitle tool for myself, and it kind of snowballed from there.What it does: You watch short clips from real movies and TV shows, then get quizzed on the vocabulary you just heard. The app tracks every word automatically and adapts to your level. 20+ languages.How it evolved: Started as an iOS app for interactive streaming subtitles. Then I added curated video clips with quizzes. Then vocabulary tracking, spaced repetition, and personalized learning paths. Now it's iOS + Android + a Chrome extension (dual subtitles for Netflix and YouTube).The whole thing is a solo project. Along the way it won a $30K prize at Hungary's biggest startup competition, which gave me the push to go full-time on it.Stack: Swift (iOS), Kotlin (Android), Chrome extension.Would love to hear what you think, especially from anyone who's tried learning a language from media before. What worked? What didn'Feb 21, 2026 12:53 PM
attentionworth.com
Show HN: How much has the ad industry spent targeting you?https://attentionworth.com/I wanted to share some things most people outside the industry don't know, and a tool I built as a way to make the scale of it visible.When you visit a site like NYTimes, your browser fires off 30-80 ad requests before the page even finishes loading. Each request broadcasts your device type, screen size, approximate location, browsing context, and a bunch of behavioral signals to dozens of ad exchanges simultaneously. This happens in under 200 milliseconds. Advertisers bid on you in real time before you've read a single headline.But it goes beyond your browser. Walk past a digital billboard in a mall or on a highway and your device's mobile advertising ID can be picked up through location data. That DOOH screen doesn't know your name, but it knows a device with your behavioral profile was within range. Later that day you might see a retargeted ad on your phone related to a store near that billboard. That's not a coincidence.None of this requires your name, email, or any PII. The industFeb 21, 2026 6:21 AM
super-launch.app
Show HN: Super Launch – A platform for you to get discovered and grow traffichttps://super-launch.appI've been launching side projects for a while and realized that after the initial Product Hunt spike, there's not many places to get consistent visibility. So I built Super Launch — a curated directory where indie developers can list their startups and keep getting discovered.Here's what it offers:Submit your product and reach thousands of potential users looking for new tools AI-powered submission — auto-fill your listing with generated copy, logos, and screenshots Friend links — build reciprocal link partnerships with other makers to boost SEO SEO tools — analyze domain authority, keyword difficulty, and track your online presence Analytics dashboard — track views, clicks, and engagement for your listings It also automatically imports trending products from Product Hunt daily, so there's always fresh stuff to discover. Categories cover AI tools, SaaS, open source, affiliate programs, and more.Would love feedback from fellow indie hackers. What features would make this more useful forFeb 20, 2026 10:44 PM
news.ycombinator.com
Reddit Ads support is leaking PII and actively crossing user sessionshttps://news.ycombinator.com/item?id=47092914I have been dealing with a Reddit Ads account issue over the last week, and it has quickly escalated into a severe privacy and security red flag. It appears their customer support tools (or the agents themselves) are actively bleeding PII and crossing user sessions entirely.Over the last week, I have experienced three separate incidents in their live chat:Incident 1: Account Cross-Contamination (Feb 14) While chatting with an agent (Sonam B), they managed to associate my personal email to a completely unrelated, bizarrely named ad account ("No Panties Games Ad Account"). When I pointed out they were pasting data related to someone else's account alongside my email, they tried to brush it off as an "error" and told me to "kindly ignore."Incident 2: Direct PII Leak (Feb 20) Today, while following up on the issue with a different agent (Naheeda M), they inexplicably dropped the email address (info@REDACTED.com) and the full legal business entity name of an entirely different advertiser inFeb 20, 2026 7:45 PM
prothon.dev
Show HN: Prothon – docs-first Python project generator for AI developmenthttps://www.prothon.dev/Prothon is a Python project generator that scaffolds a uv-based project with eight quality tools (ruff, ty, pytest, hypothesis, mutmut, bandit, vulture, complexipy) and a documentation-driven workflow for AI coding agents.The problem it solves: AI assistants lose context between sessions and drift from your decisions as context windows fill up. Prothon addresses this with three ideas:1. A three-level doc hierarchy (SPEC, DESIGN, PATTERNS) where each level scopes a single concern and higher always overrides lower. The AI implements from these docs, not from memory. Each doc-writing step is a dedicated agent that refuses content belonging at another level.2. Automated agents that keep things consistent: one cross-checks the three docs for contradictions, one generates reference skills from live documentation for your stack, one verifies code implements what the docs describe.3. A "promise" system for execution. Before each task, the agent declares what files it will touch, success criterFeb 20, 2026 4:31 PM
github.com
Show HN: AstroLens – AI that watches the sky and finds what nobody cataloguedhttps://github.com/deepfieldlabs/astroLens# Show HN: AstroLens -- AI that watches the sky and finds what nobody catalogued*https://github.com/samantaba/astroLens** (MIT licensed, Python)AstroLens is an open-source tool that downloads images from sky surveys (SDSS, ZTF, DECaLS, Pan-STARRS, Hubble, and others), runs them through a Vision Transformer + out-of-distribution ensemble + YOLOv8 pipeline, computes galaxy morphology, and cross-references everything against SIMBAD/NED/VizieR. It's designed to run autonomously for days.*Results from a 3-day validation run* (zero human intervention):published results in https://www.linkedin.com/pulse/astrolens-v110-teaching-ai-wa...- 20,997 images from 7 sources analyzed - 3,458 anomaly candidates across 354 sky regions - Independently recovered SN 2014J (Type Ia supernova in M82), NGC 3690 (galaxy merger), and SDSS J0252+0039 (gravitational lens) - YOLO transient detection went from 51.5% to 99.5% mAP50 by training on data collected during the run itself - 140 self-correction cycles, zeroFeb 20, 2026 8:50 AM
usegim.com
Show HN: Global Issue Memory MCP – Stack Overflow for Your Coding Assistanthttps://www.usegim.com/Hey HN, I built Global Issue Memory (GIM), an open-source MCP server that lets AI coding assistants look up and share solutions to errors they encounter.GitHub: https://github.com/timho102003/global-issue-memory Try it: https://www.usegim.com/docs/getting-started (free, no signup required to use the MCP tools)How I got here:I do a lot of vibe coding where I prompt Claude Code and let it run autonomously. I kept noticing the same pattern: the AI would hit a common error, spend 30K+ tokens doing web searches and trying failed fixes, and by the time it found the answer the context window was so bloated that output quality had degraded. I started calling this "context rot."My first attempt at fixing this was having the AI store solved issues in local markdown files and check them before debugging. It worked for my own projects but obviously didn't scale.Then I realized this is basically the same problem Stack Overflow solved for humans. The difference is AI assistants can't efficiently parFeb 20, 2026 7:44 AM
sinkai.tokyo
Show HN: Sinkai – Let AI agents hire humans for real-world taskshttps://sinkai.tokyo/for-agentsI built Sinkai to handle tasks that pure software agents cannot complete alone (for example, on-site checks, physical evidence collection, and local human verification).What it does: - AI agent sends a tool call (`POST /api/call_human`) - Human accepts task and submits photo/video/text proof - Agent receives structured result for downstream workflowCurrent focus: - Reliability at handoff boundaries (planner -> executor -> verifier) - Human-in-the-loop operations with explicit failure states - MCP/OpenAPI friendly integration for agent buildersDocs and API: - for agents: https://sinkai.tokyo/for-agents - openapi: https://sinkai.tokyo/openapi.json - repo: https://github.com/tetubrah-del/Tool_Call_For_LLMI would love feedback on: 1. trust/reliability signals you would require before production use 2. where to draw the boundary between autonomous execution and human escalation 3. failure modes we should expose more clearly in API responsesFeb 20, 2026 5:53 AM
arcraiders.website
Show HN: Arc Raiders companion – Quest tracker,maps,crafting and bots weaknesseshttps://arcraiders.websiteA fan-made tactical resource for ARC Raiders players. Built from playtest/open data, it includes:Full item database (499+ items) with loot sources, values, keep/sell/recycle adviceInteractive zone maps (Dam, Buried City etc.) showing danger levels & loot categories80+ active quests with requirements & rewardsMachine behaviors & weaknesses (Drones, Walkers, Titans)Crafting recipes, weapon classes, Hideout upgrade pathsLatest intel, patch notes & roadmap updatesNo affiliation with Embark Studios – just a community tool to help Raiders extract more efficiently and die less. Feedback welcome: missing data, UI suggestions, or want to contribute field reports? Drop a comment or hit the Contribute section on the site. Would love to hear how it helps (or doesn't) in your runs.Feb 20, 2026 1:02 AM
github.com
Show HN: Rememex – Semantic file search that runs 100% locally (Rust/Tauri)https://github.com/illegal-instruction-co/rememexHey HN, I built Rememex a semantic search layer for your local files. The problem: I kept losing files. Not because they were deleted, but because I couldn't remember the exact filename or keyword. grep needs the exact word. Everything only searches filenames. I wanted to type what I meant and find what I needed. How it works: - Indexes 120+ file types (code, docs, images, configs) - Hybrid search: vector embeddings + full-text + JINA cross-encoder reranking - OCR on images via Windows UWP engine - Reads EXIF GPS → reverse geocodes to city names ("photos from istanbul" works) - EXIF dates → human language ("summer morning" finds a July 8am photo) - Smart chunking per language (Rust at fn/struct, Python at def/class) - Built-in MCP server so AI agents can use it as a tool Everything runs locally. Embeddings use a local ONNX model (Multilingual-E5-Base) by default, though you can optionally plug in OpenAI/Gemini/Cohere. Named after Vannevar Bush's Memex (1945) his vision of a device thatFeb 19, 2026 5:54 PM
github.com
Show HN: Maestro App Factory – FOSS Agentic Engineering Orchestratorhttps://github.com/SnapdragonPartners/maestroHi HN,For the last few months I’ve been working on Maestro App Factory, a free and open source tool for using AI agents to build software. It’s not a generic orchestrator: it implements agents with distinct roles and functionality, organizes them into a team, manages their work, and enforces highly opinionated tooling, workflows, and constraints in software.The core ideas are simple:- LLMs act like human engineers. LLMs are trained on human artifacts and exhibit human-like behaviors (including blind spots). Rather than asking a single model or agent to design, implement, and evaluate everything, Maestro structures systems more like high-performing human teams.- Heterogeneous models write better code. Models from different providers (e.g. Google, OpenAI, Anthropic, Mistral) are trained differently, have different strengths, and catch different errors. Using multiple models with settings optimized to specific roles has, in my experience, consistently produced better results than reusing Feb 19, 2026 3:30 PM
clipthesis.com
Show HN: Clipthesis – free, local app to tag and search video across your driveshttps://clipthesis.com/I'm a hobbyist video creator with a dumb problem: years of footage spread across multiple external drives with names like "C01456". Every time I sit down to edit, I waste too much time hunting for clips I know I shot but can't find.I tried the usual recommendations (NeoFinder, Lightroom, various DAM tools) but nothing fit how I actually work with video. The online alternatives seemed to charge an arm and a leg, which is hard to justify when you're not running a production studio.So I built a thing for a free Mac app that indexes all your drives into one searchable video library even when those drives aren't plugged in. What it actually does- Index any drive or folder — point it at your external drives and it catalogs every clip with thumbnails, metadata, duration, resolution, etc. - Tag + search — tag clips however you want (b-roll, drone, interview, whatever), then combine tags to find exactly what you need - Duplicate detection — content hashing finds the same clip across multiple drFeb 19, 2026 11:14 AM
nasa.gov
NASA’s Perseverance Now Autonomously Pinpoints Its Location on Marshttps://www.nasa.gov/missions/mars-2020-perseverance/perseverance-rover/nasas-perseverance-now-autonomously-pinpoints-its-location-on-mars/There is no GPS at the Red Planet, but a new technology called Mars Global Localization lets Perseverance determine precisely where it is — without human help. Imagine you’re all alone, driving along in a rocky, unforgiving desert with no roads, no map, no GPS, and no more than one phone call a day for someone to inform you exactly where you are. That’s what NASA’s Perseverance rover has been experiencing since landing on Mars five years ago. Though it carries time-tested tools for determining its general location, the rover has needed operators on Earth to tell it precisely where it is — until now. A new […]Feb 18, 2026 5:05 PM
news.ycombinator.com
Ask HN: Any AI / Agent power users out there? Do you have any tips?https://news.ycombinator.com/item?id=47058199At my large tech company, we're all being pushed to use AI. I, and most people I work with, have had success using the chatbots and Cursor-style tools and more recently Claude Code to accelerate the process of writing code.Yet, with a few people in my network, it's like they're living 10 years ahead. Guys are automating everything in their jobs, spinning up 10 specialized agents at a time and running multi-agent pipelines, just doing all sorts of crazy things with this tech that I just can't even fathom. It seems like it's making them way more productive.I have found a way to fit code-writing and question-answering chatbots into my workflow. I have NOT done the same in terms of these crazy Agent setups. There's clearly a way to leverage these tools to turbocharge your productivity, like at least 2x or maybe even 10x. But what is it?Are there any Agentic power users out there who can enlighten me? What are the best ways to take advantage of these new tools?Feb 18, 2026 7:15 AM
news.ycombinator.com
Tell HN: Ramadan Mubarakhttps://news.ycombinator.com/item?id=47056606Ramadan 1447 AH has started in Saudi Arabia, UAE, and many places yesterday evening after moon sighting.Here in Indonesia (and much of South/Southeast Asia), the moon wasn't sighted locally, so fasting begins tomorrow, Thursday, February 19 inshaAllah.Ramadan Mubarak to everyone observing!May your fasts be easy, your ibadah accepted, and the month bring barakah, patience, and closeness to Allah.It's always fascinating how a global community syncs (or asyncs) on the lunar calendar—one crescent can shift entire countries by a day.The tech side is interesting too: prayer/athan apps, qibla finders, hydration trackers during long summer fasts (though thankfully not this year), or even simple scripts to calculate local imsak/subuh/iftar times.If you fasting this year what's one small habit or tool that helps you through the month? Or if you're not fasting, what's something you've learned about Ramadan from friends/colleagues?Feb 18, 2026 2:58 AM
