No self-reporting. Only code. Built a platform that tries to infer what you know from your actual work. Comments URL: https://news.ycombinator.com/item?id=47815869 Points: 1 # Comments: 0
I built LogsGo as a learning project to explore log ingestion, querying, and storage tradeoffs. It’s a small Go-based system where logs come in over gRPC, land in memory first, then flush into local storage and optionally S3-compatible object storage. I also added a simple query language plus a small UI to inspect log occurrences over time. This wasn’t built because I think the world needed “another logging system” or because I’m an expert here. I mostly wanted to learn by building something end to end: ingestion paths, storage layering, querying, retention, auth/TLS, and some UI work. Repo: https://github.com/Saumya40-codes/LogsGO I’d genuinely appreciate feedback, including “this design is wrong for X reason” type feedback. If parts of it feel overengineered / naive / badly thought through, that’s useful for me too. Comments URL: https://news.ycombinator.com/item?id=47815402 Points: 1 # Comments: 0
Article URL: https://github.com/AssafWoo/homebrew-pandafilter Comments URL: https://news.ycombinator.com/item?id=47814614 Points: 3 # Comments: 0
Built InGaming as a frontend MVP for the iGaming space, with two main parts: 1. a back-office admin for casino / gambling operations 2. a multi-brand storefront system for casino sites The starting point was simple: in all the years I’ve worked around iGaming, I haven’t seen a single admin panel I’d call genuinely good. Most of them are either bloated, awkward to use, or patched together from disconnected tools that don’t really fit how teams work day to day. So I decided to build the kind of admin I actually wish existed. On the admin side, the MVP includes flows for players, payments and transactions, casino and sportsbook operations, content and static pages, reporting, and website configuration. On the storefront side, I built a separate frontend repo to validate a multi-brand setup on shared foundations. Right now that includes two demo brands: BetStake and ShipBet. Important caveat: this is frontend work and product validation, not a finished end-to-end platform. I’m treating it as an expanded MVP that is already strong enough to be a serious starting point, rather than just another rough concept. What I’m really trying to figure out now is whether this is already valuable as a project someone could buy as a head start. My view is that it could save a lot of time for a team that would otherwise need to build the admin, storefront layer, and overall product structure from scratch. Happy to answer product or frontend architecture questions. Comments URL: https://news.ycombinator.com/item?id=47813386 Points: 1 # Comments: 0
I built this to run OpenClaw safely. The problem: every sandbox I tried still handed the real API token to the agent as an env var. nilbox never gives the agent the real token. It gets a fake placeholder instead (ANTHROPIC_API_KEY=ANTHROPIC_API_KEY). nilbox intercepts outbound API calls and swaps in the real token at the network layer. So if the agent leaks the "token" — attacker gets a useless string. That's it. Also ships a managed Linux runtime (consistent across mac/win/linux) and a Store for one-click agent app installs. Full shell access too. Available for macOS, Windows, and Linux https://nilbox.run Curious how others are thinking about token security when running agents locally. Comments URL: https://news.ycombinator.com/item?id=47812193 Points: 3 # Comments: 0
Built this around one simple idea: the workflow that wants to execute should not be the same place that decides whether execution may continue. This project puts an external allow/deny boundary before action. Public entry points: * live pilot * commercial request * private deployment There is also a GitHub Marketplace action install surface, but the main point is the boundary model itself: decision stays outside the workflow that is asking to proceed. Looking for feedback from people working on CI/CD, security controls, approval boundaries, and automated execution. Comments URL: https://news.ycombinator.com/item?id=47811161 Points: 2 # Comments: 0
I built this tool a while back when I accidentally deleted thousands of my PDFs. I found the existing ext4magick and similar solutions to be cumbersome and complicated to use, and wanted something similar that just did PDFs. As a bonus, because it only handles PDF documents, the pattern recognision is super simple, allowing this program to scan through a disk at high speed, like the maximum read speed for your disk. Hope people find it useful. Mirror: https://github.com/seanhly/recover-pdfs Comments URL: https://news.ycombinator.com/item?id=47810848 Points: 3 # Comments: 0
Cerebral Valley Built with Opus 4.7: a Claude Code hackathon Join us again for Built with Opus 4.7: a Claude Code virtual hackathon! Opus 4.7 is our most capable model yet, and we’re excited for you to push the limits ... | Tuesday, April 21, 2026 We’ll select 500 participants and give each one $500 in Claude API credits to build for a week. Compete to win $100k in Claude API credits. https://x.com/claudeai/status/2045248224659644654 2 个帖子 - 2 位参与者 阅读完整话题
We built AI Subroutines in rtrvr.ai. Record a browser task once, save it as a callable tool, replay it at: zero token cost, zero LLM inference delay, and zero mistakes. The subroutine itself is a deterministic script composed of discovered network calls hitting the site's backend as well as page interactions like click/type/find. The key architectural decision: the script executes inside the webpage itself, not through a proxy, not in a headless worker, not out of process. The script dispatches requests from the tab's execution context, so auth, CSRF, TLS session, and signed headers get added to all requests and propagate for free. No certificate installation, no TLS fingerprint modification, no separate auth stack to maintain. During recording, the extension intercepts network requests (MAIN-world fetch/XHR patch + webRequest fallback). We score and trim ~300 requests down to ~5 based on method, timing relative to DOM events, and origin. Volatile GraphQL operation IDs are detected and force a DOM-only fallback before they break silently on the next run. The generated code combines network calls with DOM actions (click, type, find) in the same function via an rtrvr.* helper namespace. Point the agent at a spreadsheet of 500 rows and with just one LLM call parameters are assigned and 500 Subroutines kicked off. Key use cases: - record sending IG DM, then have reusable and callable routine to send DMs at zero token cost - create routine getting latest products in site catalog, call it to get thousands of products via direct graphql queries - setup routine to file EHR form based on parameters to the tool, AI infers parameters from current page context and calls tool - reuse routine daily to sync outbound messages on LinkedIn/Slack/Gmail to a CRM using a MCP server We see the fundamental reason that browser agents haven't taken off is that for repetitive tasks going through the inference loop is unnecessary. Better to just record once, and get the LLM to generate a script leveraging all the possible ways to interact with a site and the wider web like directly calling backed API's, interacting with the DOM, and calling 3P tools/APIs/MCP servers. Comments URL: https://news.ycombinator.com/item?id=47810533 Points: 5 # Comments: 1
Built this code art project as a commentary on the recent wave of using LOC as a primary measure of engineering productivity It's now up to 350M lines of useless (but hopefully funny) code, constantly growing using GitHub actions and https://github.com/jshchnz/codemaxxing Unfortunately, GitHub has now sent me a notice to remove the repo within the next 14 days, so enjoy it while you can Comments URL: https://news.ycombinator.com/item?id=47808458 Points: 1 # Comments: 0
Article URL: https://kwverdit.com Comments URL: https://news.ycombinator.com/item?id=47807541 Points: 1 # Comments: 0
I built a mock server that reads an OpenAPI spec and serves realistic, deterministic responses — no mock definitions, no config files. ``` mimikos start petstore.yaml ``` That's the entire setup. Mimikos parses your spec, classifies each endpoint (create, fetch, list, update, delete), and generates schema-valid responses with realistic data. Same request always returns the same response, so it's safe for snapshot tests and CI. What's different from Prism, WireMock, etc.: Most mock servers either (a) require you to hand-write every response, or (b) generate random data that changes on every request. When the spec changes, your mocks break — or worse, silently become wrong. Mimikos sits in an unoccupied spot: zero config + high-quality responses. The key piece is behavioral inference — a three-layer heuristic classifier that determines what each endpoint does (not just what it returns) from the HTTP method, path structure, response schema, and operationId. It gets this right 96% of the time across a 344-endpoint corpus (Stripe, GitHub, Asana, Spotify specs). Features: 1. Deterministic responses via request fingerprinting + per-field sub-seeding (adding a field to your schema doesn't change existing field values) 2. Stateful mode (`--mode stateful`) for full CRUD workflows — POST creates, GET retrieves, DELETE removes 3. 120+ semantic field mappings — `email` gets an email, `name` gets a name, `url` gets a URL 4. Spec-authored `example` values used directly when present 5. RFC 7807 error responses with field-level diagnostics for invalid requests 6. Single Go binary, no runtime dependencies, <50ms responses Technical details: The architecture separates startup from runtime: spec parsing, schema compilation, and behavior classification happen once at startup. At request time, it's routing + deterministic generation from a pre-computed seed. No AI in the request path — no LLM calls, no network dependencies. I explored using LLMs for behavior classification twice (GPT-4.1, Claude) and they scored worse than the heuristic both times. The heuristic wins because it can exploit structural signals (path patterns, schema shapes) that LLMs tend to ignore in favor of semantic reasoning. Written in Go, single binary. OpenAPI 3.0 and 3.1. Apache 2.0 licensed. - Get started: https://mimikos.dev/getting-started/ - Repo: https://github.com/mimikos-io/mimikos Comments URL: https://news.ycombinator.com/item?id=47806801 Points: 2 # Comments: 0
Hi everyone, I built a small tool called GitShrink to solve a simple problem: making videos small enough (<10MB) to upload to GitHub. It runs entirely in the browser, so nothing is uploaded anywhere. Website: https://igtumt.github.io/gitshrink/ GitHub: https://github.com/igtumt/gitshrink It’s a small, local-first tool with no accounts, no tracking, and no backend. Use cases: README demo videos Small product demos Screen recordings for GitHub Feedback is welcome. Comments URL: https://news.ycombinator.com/item?id=47806514 Points: 2 # Comments: 1
Hi HN, I built NotchPrompter because I needed a simple way to read notes while looking at the camera during calls, without heavy or paid software. - 100% free & open-source - native macOS (SwiftUI) - minimalist - focuses on the essentials. Feedback and contributions are more than welcome! PS No, I didn't use AI for it. I always wanted to play with SwiftUI and this is my 6th approach to this. Previous projects were too complex for my beginner skills. I'm mainly a Java developer. It took me ~5 months to build this during free weekends. Comments URL: https://news.ycombinator.com/item?id=47804818 Points: 1 # Comments: 0
I built a small CLI after running into UI/UX problems on my own site. It runs quick checks on a URL (layout overflow, tap targets, basic a11y via axe, console/network errors) and outputs a JSON report plus screenshots. Early stage and feedback welcome Comments URL: https://news.ycombinator.com/item?id=47803926 Points: 1 # Comments: 0
Hi HN, I built a small macOS utility to solve something that kept bothering me: external drives not being safely ejected when a Mac goes to sleep, leading to “Disk Not Ejected Properly” warnings and potential data issues. macOS doesn’t always give volumes enough time to unmount, especially with SSDs, SD cards, or disk images. Manually ejecting works, but it’s easy to forget. Ejectify automates this by unmounting selected volumes right before sleep (or display off) and mounting them again on wake. In Ejectify 2, I focused on making (un)mounting more reliable, with an optional helper for better consistency, improved sleep handling to give operations more time to complete, broader volume support, and the ability to suppress system warnings if they still occur. To celebrate the launch, it’s available for €4.99 (instead of €6.99) with the code EJECTIFY2. Best, Niels. Comments URL: https://news.ycombinator.com/item?id=47803513 Points: 2 # Comments: 0
I've always felt like Twitch was missing a social aspect, and I'm trying to address it. I know they had Twitch Pulse, and they decided to shut it down, but I'm hoping this can be a better solution. Comments URL: https://news.ycombinator.com/item?id=47803313 Points: 1 # Comments: 1
Article URL: https://github.com/Jay9185/PIREP Comments URL: https://news.ycombinator.com/item?id=47803303 Points: 1 # Comments: 0
Hey HN, I built a service that merges user identity across websites, mobile apps, and in-app WebViews into one profile. The problem: if you have an app and a website, the same person shows up as 2-4 different "users" in your analytics. Your user counts are inflated, your funnels are broken, and you're retargeting people who already converted. I built SDKs (Web, Android, iOS) that handle this server-side. The bit I'm most proud of is a WebView bridge — when someone opens a link in your app's in-app browser, the SDKs silently link the sessions without needing a login. Live demo: https://crosstrack-demo.onrender.com Is this a problem you've actually hit? Curious what you'd need to try something like this. https://crosstrackdata.com Comments URL: https://news.ycombinator.com/item?id=47803204 Points: 2 # Comments: 0
I built *valuepulse.ai* out of a frustration I kept running into: everything is scattered across different tools. If I want to search internal knowledge, I use something like Glean. If I want to analyze data, I switch to a BI tool or a data agent. And if I need external context, I’m back to Google or Perplexity. Even with all the progress in generative AI, it still feels pretty disconnected. So I wanted to see if it’s possible to bring search (both web and enterprise) and analytics into one place. The core idea behind Valuepulse is a *context layer* that models a company’s business domain (customers, revenue, operations, etc.) and sits across different data sources (structured and unstructured). On top of that, I built: * natural language → dashboard generation (similar idea to tools like Replit/Lovable, but for BI), using agents like text-to-SQL and visualization selection * a data agent that can query databases and warehouses in plain English * unified search across internal docs and the web using vector search It’s still early (beta), but I’d really appreciate feedback—especially on whether this feels useful for real business users (non-technical) or where it might break. Comments URL: https://news.ycombinator.com/item?id=47802054 Points: 2 # Comments: 1