📦 astaxie/tokenhub

TokenHub

TokenHub is the private enterprise AI gateway that finally puts smart governance and cost control into a single private deployment. By acting as a central bridge between your team and upstream models like OpenAI, Anthropic, or local tools like Ollama, it helps you manage keys, set quotas, and track exact costs by department. Unlike complex cloud gateways, it uses a simple SQLite setup that you can spin up locally in minutes. If you want to give your team safe access to different AI providers without losing track of your budget, this is the perfect self-hosted solution to try today.

📰 https://news.ycombinator.com/item?id=48995071

🆔 @hackernewsgithubprojects

📦 razzant/claudexor

Claudexor

Claudexor is the local-first control plane that finally lets you run and coordinate the AI coding agents you already pay for right from your machine. Instead of fighting rate limits, it sits on top of tools like Claude Code, Codex, and Cursor, letting you register multiple subscriptions side by side and automatically rotating spent accounts out of the way when they hit vendor quotas. It can even run competitive races between different models to find the cleanest patch, using independent AI reviewers to grade the results before anything touches your code. It is an incredibly clever way to orchestrate local AI development without any external telemetry.

📰 https://news.ycombinator.com/item?id=48988194

🆔 @hackernewsgithubprojects

📦 autowarefoundation/auto_e2e

AutoE2E

Build a self-driving system using only cameras and zero HD-maps. This repository delivers a fully open-source end-to-end AI model designed to navigate complex city streets and highways. By taking raw feeds from seven surround cameras and a basic map tile, it predicts exactly where a vehicle should steer and how it should accelerate over the next six seconds. It even comes with a production dashboard to play synchronized video scenes, visualize predicted trajectories, and check the model's reasoning. Grab your camera data, run the setup, and start training your own autonomous driving model today.

🆔 @hackernewsgithubprojects

📦 duartesantos8/opengym

openGym

You can host your own complete workout and body-weight tracker on your home server with just a single command. The project, called opengym, lets you plan routines, run guided workouts with animated demos, and track your personal records, all while keeping your data entirely in your own hands. Instead of typing passwords, you log in securely using your phone's built-in fingerprint or face scanner. Because everything runs on your own hardware, you never have to worry about paid subscriptions, ads, or your fitness data being harvested. It is the perfect, private companion to help you stay consistent and hit your fitness goals.

🆔 @hackernewsgithubprojects

📦 cursor/minisqlite

minisqlite

minisqlite is the incredible Rust reimplementation of SQLite that reads and writes actual SQLite database files with zero unsafe code. It is an incredibly polished library that packs a query planner, a B-tree engine, and transactions into a tiny public API with just four methods. The coolest part is its flawless on-disk compatibility; you can create a database in standard SQLite, modify it with minisqlite, and open it back up in SQLite without a single error. It is a masterclass in clean architecture, making it the perfect playground to see how a real database works under the hood.

📰 https://news.ycombinator.com/item?id=48982628

🆔 @hackernewsgithubprojects

📦 pancreations/halo-mcc-vr

Halo-MCC-VR

Halo-MCC-VR is the OpenXR mod that finally brings true virtual reality to Halo 3 inside the Master Chief Collection. It delivers full stereoscopic 3D, natural head tracking, and motion controller support directly into the campaign without patching game files. What makes this project incredibly fascinating is its creation process: every single line of C++ was written by artificial intelligence coding assistants, working under the guidance of a non-programmer modder who reverse-engineered the engine and tested the results in a headset. You get fully tracked virtual arms, physical weapon aiming, and smooth locomotion in an iconic campaign. Grab the files, run the launcher, and experience the fight like never before.

🆔 @hackernewsgithubprojects

📦 sniperunder123/okf-knowledge

okf-knowledge

Using a massive, thousands-of-lines README file to explain your codebase to AI agents can burn through ninety percent more tokens than necessary and cause constant hallucinations. The okf-knowledge project solves this by bringing the Open Knowledge Format directly into your Claude Code terminal as a portable skill. Instead of one giant text file, it organizes your system into atomic, interconnected Markdown files where each concept has a tiny structured header. AI agents can easily navigate this lightweight web of files, loading only what they need. It includes built-in python scripts to automatically build, validate, and even visually map your knowledge graph.

🆔 @hackernewsgithubprojects

📦 xen0xys/flowy

Flowy

Flowy is the self-hosted finance platform that finally takes back control of your family's money from chaotic spreadsheets and disconnected banking apps. Built as a collaborative household workspace, it brings checking, savings, investments, and cash accounts into one single dashboard. What makes it special is that you host it yourself on your own server, meaning your private financial data is never shared, harvested, or locked behind a subscription. You can build a customized catalog of categories to track real money movement with your partner in real time. It is a secure, permanent home for your long-term financial history.

🆔 @hackernewsgithubprojects

📦 perfectgf/lora-dataset-studio

lora-dataset-studio

lora-dataset-studio is the self-hosted workbench that finally makes training custom AI faces and characters feel like a breeze instead of a chore. If you have ever tried building a clean dataset for training, you know the absolute headache of cropping, captioning, and fixing watermarks across dozens of separate tools. This clever app pulls everything into a single, gorgeous dashboard. You can generate multiple angles from just one reference photo, auto-clean watermarks with a smart brush, auto-caption everything, and run your training sessions seamlessly. It is the ultimate playground for getting your custom image models looking exactly how you want.

🆔 @hackernewsgithubprojects

📦 runvendo/vendo

Vendo

Give your customers the power to build their own custom features and mini-apps directly inside your existing product. Vendo places a secure AI agent inside your software, allowing users to safely automate tasks, design custom views, and link up their tools without breaking your application. The coolest part is how easily you can set it up; you just drop a short instruction into your coding assistant, and it automatically configures the integrations, runs a system check, and hooks everything up safely. It is like giving your application a flexible, customizable interface that adapts to exactly what each client needs. Go grab the package on npm and let your users build...

📰 https://news.ycombinator.com/item?id=48926618

🆔 @hackernewsgithubprojects

📦 mobai-app/simslim

simslim

The iOS simulator manager that slashes memory footprint by seventy-five percent so you can run four times as many virtual devices on a single Mac. Every time you boot up a fresh iOS simulator, Apple fires up about one hundred and eighty background services like Siri, Spotlight, and photo indexing. While those are great on a phone, they completely choke your machine during automated testing or app development. Simslim lets you target specific simulators and permanently disable those power-hungry background daemons. You can go from a laptop that starts chugging after five simulators to running nearly twenty side-by-side without swapping.

🆔 @hackernewsgithubprojects

📦 extra-org/extra

Extra

You can now add a smart, domain-aware agent layer to any application using just a single YAML file. Extra is a lightweight engine that turns your declarative configuration into a running system of focused AI agents, each scoped to its own tools, prompts, and data. Instead of writing complex routing logic, you describe your agents and let the engine automatically direct user requests to the right specialist. This strict scoping keeps answers grounded and prevents different parts of your business logic from bleeding together. It is an incredibly clean way to build production-ready agentic features in days rather than months.

🆔 @hackernewsgithubprojects

📦 matheuzsecurity/furtex

furtex

Linux security systems are completely blind to this. The furtex toolkit executes complex post-exploitation and evasion techniques by using raw kernel operations instead of standard system calls, which entirely bypasses traditional endpoint detection. Instead of relying on common system call tables that security agents monitor, this project uses io_uring workqueues and custom BPF map tools to read files, establish network connections, and inject code silently. It is a fascinating playground for researchers trying to understand the next generation of stealth techniques, demonstrating how modern kernel features can be repurposed. You can compile it directly to test how your own defenses hold up against these quiet tactics.

📰 https://news.ycombinator.com/item?id=48999323

🆔 @hackernewsgithubprojects

📦 akitaonrails/ai-memory

AI Memory

AI Memory is the developer tool that keeps your coding agents in sync across different terminal sessions and providers. If you have ever been frustrated by an AI assistant losing all context the second you close the window, this utility solves it by automatically capturing session milestones and writing them to a shared git-backed markdown wiki. You can literally shut down Claude Code mid-task, fire up Codex or Cursor in the exact same workspace, and the new agent immediately receives a clear handoff of where you left off. It is the ultimate persistent brain for your developer agents.

🆔 @hackernewsgithubprojects

📦 marcelroed/gigatoken

Gigatoken

Slash your language model tokenization times from hours to mere seconds by switching to gigatoken. This clever project delivers a drop-in replacement for Hugging Face and tiktoken that runs up to a thousand times faster, letting you process text data at multiple gigabytes per second. Instead of getting bogged down by slow regex engines during pre-tokenization, it uses highly optimized rust code, custom SIMD operations, and a smart caching system to fly through text. It is a game-changer for anyone dealing with massive datasets, allowing you to tokenize billions of words on a standard CPU over your morning coffee break.

🆔 @hackernewsgithubprojects

📦 justvugg/nanoeuler

nanoeuler

You can train a hundred and sixteen million parameter language model on a single graphics card using code written entirely from scratch in C and CUDA. A brilliant developer built nanoeuler to bypass heavy frameworks like PyTorch completely, hand-writing everything from the tokenizers to the actual math equations that run the neural network. By coding the forward and backward training passes by hand, they created an educational masterpiece that pretrains on classics and web text, then fine-tunes into a chat assistant. It is a beautiful, raw demonstration of pure engineering that shows exactly how these massive AI systems compute under the hood.

📰 https://news.ycombinator.com/item?id=48710778

🆔 @hackernewsgithubprojects

📦 postgresql-tools/migra

migra

Compare two PostgreSQL database schemas and instantly generate the SQL migration script needed to transition from one to the other. Instead of manually writing tedious alter table commands, you just point this tool at your production database and your development branch. It compares tables, constraints, indexes, views, and security policies, then outputs the clean SQL to sync them up. If you prefer not to connect to a live database, you can simply feed it schema dump files instead. It also features optional smart AI breakdowns that explain migration risks, generate rollbacks, and advise on potential locking issues before you deploy.

📰 https://news.ycombinator.com/item?id=48328343

🆔 @hackernewsgithubprojects

📦 chenglou/freerange

Freerange

Catch layout bugs and math errors in your TypeScript code at compile time without ever opening a browser. Freerange analyzes ordinary TypeScript code to track the exact ranges of your numbers, instantly flagging potential division by zero, out-of-bounds array indexes, or layout calculations that might result in NaN or Infinity. Instead of introducing a complex new language or messy annotations, it lets you use simple, static console assertions to set up guardrails for your UI code. It is a incredibly fast tool that works directly with the official TypeScript API to guarantee your calculations are completely safe before your app even runs.

📰 https://news.ycombinator.com/item?id=48987176

🆔 @hackernewsgithubprojects

📦 crawlseo/crawlseo

CrawlSEO

CrawlSEO is the self-hosted dashboard that finally makes SEO monitoring painless for founders who do not want to pay hundreds of dollars a month for bloated agency tools. It brings Google Search Console data, Core Web Vitals, and a built-in site crawler together in one free dashboard. You can run concurrent audits on up to two thousand pages, map performance scores, and track keywords with direct delta comparisons. Best of all, it has an integrated Model Context Protocol server, meaning you can hook up AI agents like Claude to query your SEO data or trigger site crawls directly from your developer terminal.

🆔 @hackernewsgithubprojects

📦 sqliteai/adam

Adam

The entire AI agent loop now fits into a single C include file. That is the magic of adam, an embeddable, cross-platform library that acts like the SQLite of AI frameworks. Instead of dragging in massive python environments, this tiny C library compiles anywhere, from iOS to WebAssembly, giving you tool calling, long-term memory, and full audio pipelines. It even runs local models right on your device. You can even embed it as a SQL extension in SQLite or Postgres so your database can literally query itself using natural language. It is a incredibly smart, self-contained way to build AI apps.

📰 https://news.ycombinator.com/item?id=48035677

🆔 @hackernewsgithubprojects