🛠️ All DevTools

[Other] Show HN: MCP Gateway – Unifying Access to MCP Servers Without N×M Integrations Many teams connecting LLMs to external tools eventually encounter the same architectural issue: as more tools and agents are added, the integration pattern becomes an N×M mesh of direct connections. Each agent implements its own auth, retries, rate limiting, and logging; each tool needs credentials distributed to multiple places and observability becomes fragmented.<p>We built LLM gateway with this goal to provide a single place to manage authentication, authorization, routing, and observability for MCP servers, with a path toward a more general agent-gateway architecture in the future.<p>The system includes a central MCP registry, support for OAuth2&#x2F;DCR integration, Virtual MCP Servers for curated toolsets, and a playground for experimenting with tool calls.<p>Resources -<p>Architecture Blog – Covers the N×M problem, gateway motivation, design choices, auth layers, Virtual MCP Servers, and the overall model.<p><a href="https:&#x2F;&#x2F;www.truefoundry.com&#x2F;blog&#x2F;introducing-truefoundry-mcp-gateway" rel="nofollow">https:&#x2F;&#x2F;www.truefoundry.com&#x2F;blog&#x2F;introducing-truefoundry-mcp...</a><p>Tutorial – Step-by-step guide to writing an MCP server, adding Okta-based OAuth, and integrating it with the Gateway.<p><a href="https:&#x2F;&#x2F;docs.truefoundry.com&#x2F;docs&#x2F;ai-gateway&#x2F;mcp-server-oauth-okta" rel="nofollow">https:&#x2F;&#x2F;docs.truefoundry.com&#x2F;docs&#x2F;ai-gateway&#x2F;mcp-server-oaut...</a><p>Feedback on gaps and edge cases is welcome.<p><a href="https:&#x2F;&#x2F;www.truefoundry.com&#x2F;mcp-gateway" rel="nofollow">https:&#x2F;&#x2F;www.truefoundry.com&#x2F;mcp-gateway</a>

[Other] Building a Toast Component

[IDE/Editor] Show HN: Fresh – A new terminal editor built in Rust I built Fresh to challenge the status quo that terminal editing must require a steep learning curve or endless configuration. My goal was to create a fast, resource-efficient TUI editor with the usability and features of a modern GUI editor (like a command palette, mouse support, and LSP integration).<p>Core Philosophy:<p>- <i>Ease-of-Use:</i> Fundamentally non-modal. Prioritizes standard keybindings and a minimal learning curve.<p>- <i>Efficiency:</i> Uses a lazy-loading piece tree to avoid loading huge files into RAM - reads only what&#x27;s needed for user interactions. Coded in Rust.<p>- <i>Extensibility:</i> Uses TypeScript (via Deno) for plugins, making it accessible to a large developer base.<p>The Performance Challenge:<p>I focused on resource consumption and speed with large file support as a core feature. I did a quick benchmark loading a 2GB log file with ANSI color codes. Here is the comparison against other popular editors:<p><pre><code> - Fresh: Load Time: *~600ms* | Memory: *~36 MB* - Neovim: Load Time: ~6.5 seconds | Memory: ~2 GB - Emacs: Load Time: ~10 seconds | Memory: ~2 GB - VS Code: Load Time: ~20 seconds | Memory: OOM Killed (~4.3 GB available) </code></pre> (Only Fresh rendered the ansi colors.)<p>Development process:<p>I embraced Claude Code and made an effort to get good mileage out of it. I gave it strong specific directions, especially in architecture &#x2F; code structure &#x2F; UX-sensitive areas. It required constant supervision and re-alignment, especially in the performance critical areas. Added very extensive tests (compared to my normal standards) to keep it aligned as the code grows. Especially, focused on end-to-end testing where I could easily enforce a specific behavior or user flow.<p>Fresh is an open-source project (GPL-2) seeking early adopters. You&#x27;re welcome to send feedback, feature requests, and bug reports.<p>Website: <a href="https:&#x2F;&#x2F;sinelaw.github.io&#x2F;fresh&#x2F;" rel="nofollow">https:&#x2F;&#x2F;sinelaw.github.io&#x2F;fresh&#x2F;</a><p>GitHub Repository: <a href="https:&#x2F;&#x2F;github.com&#x2F;sinelaw&#x2F;fresh" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;sinelaw&#x2F;fresh</a>

[Other] Show HN: The Taka Programming Language Hi HN! I created a small stack-based programming language, which I&#x27;m using to solve Advent of Code problems. I think the forward Polish notation works pretty nicely.

[CLI Tool] Fine-tuning, RL, and inference in one CLI Fine-tune LLMs and ML models with automatic dataset conversion, hyperparameter sweeps, and custom RL environments - monostate/aitraining

[Other] Inline insights with C/C++ that shows CPU, memory, energy beLow automatically analyzes your C and C++ embedded code to identify performance bottlenecks and generate optimized code tailored to your target hardware. Slash execution time, reduce energy consumption, and accelerate time to market. Designed for developers building in automotive, aerospace, robotics, and other performance-critical systems, beLow simplifies the complex work of embedded code optimization so teams can focus on innovation, not fine-tuning.

[API/SDK] Show HN: Golang Client Library for Gradium.ai TTS/STT API

[Other] Client-side GPU load balancing with Redis and Lua

[Other] Detecting AV1-encoded videos with Python

[Other] Show HN: RunMat – runtime with auto CPU/GPU routing for dense math Hi, I’m Nabeel. In August I released RunMat as an open-source runtime for MATLAB code that was already much faster than GNU Octave on the workloads I tried. <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=44972919">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=44972919</a><p>Since then, I’ve taken it further with RunMat Accelerate: the runtime now automatically fuses operations and routes work between CPU and GPU. You write MATLAB-style code, and RunMat runs your computation across CPUs and GPUs for speed. No CUDA, no kernel code.<p>Under the hood, it builds a graph of your array math, fuses long chains into a few kernels, keeps data on the GPU when that helps, and falls back to CPU JIT &#x2F; BLAS for small cases.<p>On an Apple M2 Max (32 GB), here are some current benchmarks (median of several runs):<p>* 5M-path Monte Carlo * RunMat ≈ 0.61 s * PyTorch ≈ 1.70 s * NumPy ≈ 79.9 s → ~2.8× faster than PyTorch and ~130× faster than NumPy on this test.<p>* 64 × 4K image preprocessing pipeline (mean&#x2F;std, normalize, gain&#x2F;bias, gamma, MSE) * RunMat ≈ 0.68 s * PyTorch ≈ 1.20 s * NumPy ≈ 7.0 s → ~1.8× faster than PyTorch and ~10× faster than NumPy.<p>* 1B-point elementwise chain (sin &#x2F; exp &#x2F; cos &#x2F; tanh mix) * RunMat ≈ 0.14 s * PyTorch ≈ 20.8 s * NumPy ≈ 11.9 s → ~140× faster than PyTorch and ~80× faster than NumPy.<p>If you want more detail on how the fusion and CPU&#x2F;GPU routing work, I wrote up a longer post here: <a href="https:&#x2F;&#x2F;runmat.org&#x2F;blog&#x2F;runmat-accel-intro-blog" rel="nofollow">https:&#x2F;&#x2F;runmat.org&#x2F;blog&#x2F;runmat-accel-intro-blog</a><p>You can run the same benchmarks yourself from the GitHub repo in the main HN link. Feedback, bug reports, and “here’s where it breaks or is slow” examples are very welcome.

[Other] Show HN: Marmot – Single-binary data catalog (no Kafka, no Elasticsearch)

[Other] Nixtml: Static website and blog generator written in Nix

[Other] A deep dive into QEMU: The Tiny Code Generator (TCG), part 1 (2021)

[API/SDK] The Frontend Database API Gateway vibecode-db rethinks how frontend apps talk to the backend. Define your schema once and query any backend - SQLite, Supabase, PostgreSQL, REST APIs — using one unified, type-safe interface. No rewrites, no SDK switching, no lock-in. Swap databases anytime, build offline-first apps, and integrate any backend through CustomAdapter. Write once. Use anywhere.

[IDE/Editor] Xcode Simulator Supercharged SimCockpit is the native control center for your Xcode Simulator. We built it to replace the friction of terminal commands and scattered tools with a single floating window. Toggle network conditions, mock GPS locations, edit User Defaults, and record demos without breaking your flow. It is the missing companion designed to let you focus on building great apps, not managing your environment.

[Other] Show HN: Webclone.js – A simple tool to clone websites I needed a lightweight way to archive documentation from a website. wget and similar tools failed to clone the site reliably (missing assets, broken links, etc.), so I ended up building a full website-cloning tool using Node.js + Puppeteer.<p>Repo: <a href="https:&#x2F;&#x2F;github.com&#x2F;jademsee&#x2F;webclone" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;jademsee&#x2F;webclone</a><p>Feedback, issues, and PRs are very welcome.

[Other] ULID: Universally Unique Lexicographically Sortable Identifier

[Code Quality] Show HN: Cm-colors –I got tired of manually fixing wcag contrast,so I made this I usually look up palettes and the UI comes out nice except some color pairs don&#x27;t pass wcag color contrast and well - just isn&#x27;t readable Ended up writing a tiny library that automatically nudges your text color just enough to pass AA or AAA, while keeping it visually similar to the original along with a color contrast linter so we can use it in CI Thought you guys might find it useful too, it&#x27;s foss :&gt;

[Other] Durin is a library for reading and writing the Dwarf debugging format

[Other] Show HN: An AI zettelkasten that extracts ideas from articles, videos, and PDFs Hey HN! Over the weekend (leaning heavily on Opus 4.5) I wrote Jargon - an AI-managed zettelkasten that reads articles, papers, and YouTube videos, extracts the key ideas, and automatically links related concepts together.<p>Demo video: <a href="https:&#x2F;&#x2F;youtu.be&#x2F;W7ejMqZ6EUQ" rel="nofollow">https:&#x2F;&#x2F;youtu.be&#x2F;W7ejMqZ6EUQ</a><p>Repo: <a href="https:&#x2F;&#x2F;github.com&#x2F;schoblaska&#x2F;jargon" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;schoblaska&#x2F;jargon</a><p>You can paste an article, PDF link, or YouTube video to parse, or ask questions directly and it&#x27;ll find its own content. Sources get summarized, broken into insight cards, and embedded for semantic search. Similar ideas automatically cluster together. Each insight can spawn research threads - questions that trigger web searches to pull in related content, which flows through the same pipeline.<p>You can explore the graph of linked ideas directly, or ask questions and it&#x27;ll RAG over your whole library plus fresh web results.<p>Jargon uses Rails + Hotwire with Falcon for async processing, pgvector for embeddings, Exa for neural web search, crawl4ai as a fallback scraper, and pdftotext for academic papers.