Show HN: Dashboard tracking all GitHub PRs and analyzing Code Agent activity
Show HN (score: 5)Description
We are researchers from ETH Zurich interested in the real-world adoption and impact of Code Agents.
To measure this, we built a dashboard, scraping all public PRs on GitHub, analyzing which are created by different code agents (Codex, Jules, Copilot, Devin, etc.), and measuring their merge rates, sliced by various repository and PR characteristics.
Since mid-May, we've analyzed over 10 million PRs and already found some interesting trends:
Usage is high, but shallow. Agents submit ~7% of all PRs overall, but only ~1–2% on popular repos. Most activity is in low-star or experimental projects.
Merge rates vary drastically. On low-traffic repos, some agents get 90%+ of their PRs merged. On popular projects, that can drop to <25%.
Pre-review helps. Agents that require human-in-the-loop review (e.g., Jules, Codex) have 30–50% higher merge rates than Copilot-style fire-and-forget PRs.
Bias toward new code. Agent PRs mostly add code. Refactorings and deletions are rare.
If you have ideas for what other characteristics we should look at let us know or play with the code yourself
More from Show
Show HN: We developed an AI tool to diagnose car problems
Show HN: We developed an AI tool to diagnose car problems Hey HN,<p>We built AutoAI – an AI tool that tells you what's wrong with your car in plain English.<p>Just enter:<p>Your car’s make/model/year<p>The OBD2 error codes (optional) (like P0420, P0171, etc.)<p>Any symptoms you're noticing (e.g. “rough idle” or “weird sound when starting”)<p>And we’ll tell you:<p>The most likely issue<p>How to verify it yourself<p>Whether it’s a DIY fix or shop-worthy<p>No more endless Googling or forum-hopping. Built for car owners, tinkerers, and pros who want fast, reliable answers. Powered by a repair-trained AI using real-world automotive data.<p>We’re trying to make diagnostics smarter, not replace your mechanic – just make you way more informed before spending money.<p>Would love feedback or crazy edge-case inputs to improve it.
Show HN: FluidAudio – Swift Speaker Diarization on CoreML
Show HN: FluidAudio – Swift Speaker Diarization on CoreML We needed a speaker diarization solution that could run every few seconds alongside transcription on iOS and macOS. But native Swift support was either limited or locked behind paid licenses. Since diarization is a common need in speech-to-text workflows, we decided to open source our work and give back to the community.<p>We initially tried sherpa-onnx, which works, but running both diarization and transcription models slowed down older devices. CPU-only inference just isn’t ideal for near real-time workloads, so we wanted the option to offload segmentation and speaker embedding to the GPU or ANE. Supporting M1 Macs in particular meant pushing more of the workload to the ANE.<p>Instead of shoehorning the ONNX model into CoreML with C++, we converted the original PyTorch models directly to CoreML. This approach required some monkey-patching in the PyTorch and pyannote code, but the initial benchmarks look promising.<p>We’d love feedback! We're currently working on adding VAD and integrating Parakeet for transcription, but still wrestling with CoreML model conversion.
Show HN: NodeLoop – Hub for electronics design knowledge and tools
Show HN: NodeLoop – Hub for electronics design knowledge and tools Building a free web toolbox for hardware engineers: harness cable diagram generator, connector pinout tools (M.2, JTAG...), microcontroller serial monitor, and various other small utilities.<p>No sign-up required. Designed from my own needs. Feedback and feature suggestions are welcome.
Show HN: Pangolin – Open source alternative to Cloudflare Tunnels
Show HN: Pangolin – Open source alternative to Cloudflare Tunnels Pangolin is an open source self-hosted tunneled reverse proxy management server with identity and access control, designed to securely expose private resources through encrypted WireGuard tunnels running in user space.<p>We made Pangolin so you retain full control over your infrastructure while providing a user-friendly and feature-rich solution for managing proxies, authentication, and access, all with a clean and simple dashboard web UI.<p>GitHub: <a href="https://github.com/fosrl/pangolin">https://github.com/fosrl/pangolin</a><p>Deployment takes about 5 minutes on a VPS: <a href="https://docs.fossorial.io/Getting%20Started/quick-install" rel="nofollow">https://docs.fossorial.io/Getting%20Started/quick-install</a><p>Demo by Lawrence Systems (YouTube): <a href="https://youtu.be/g5qOpxhhS7M?si=M1XTWLGLUZW0WzTv&t=723" rel="nofollow">https://youtu.be/g5qOpxhhS7M?si=M1XTWLGLUZW0WzTv&t=723</a><p>Some use cases:<p><pre><code> - Grant users access to your apps from anywhere using just a web-browser - Proxy behind CGNAT - One application load balancer across multiple clouds and on-premises - Easily expose services on IoT and edge devices for field monitoring - Bring localhost online for easy access </code></pre> A few key features:<p><pre><code> - No port forwarding and hide your public IP for self-hosting - Create proxies to multiple different private networks - OAuth2/OIDC identity providers - Role-based access control - Raw TCP and UDP support - Resource-specific pin codes, passwords, email OTP - Self-destructing shareable links - API for automation - WAF with CrowdSec and Geoblocking</code></pre>
Show HN: Snub – A fast, lightweight file search CLI for Windows (written in C)
Show HN: Snub – A fast, lightweight file search CLI for Windows (written in C) built snub, a fast and lightweight recursive file search tool for Windows, written in modern C (C17). File Explorer and PowerShell often crawl on large codebases — so I wrote something that doesn’t.<p>snub is: - Fully multithreaded - Avoids indexing entirely - Supports globbing, size/date filters, result limiting, and JSON output - Packaged as a single small executable<p>No dependencies. Just raw C and the WinAPI.<p>GitHub: <a href="https://github.com/seeyebe/snub">https://github.com/seeyebe/snub</a><p>Would love feedback on C idioms, WinAPI practices, or ideas for portability and better build flow.
Show HN: Track the AI-generated code in your repo
Show HN: Track the AI-generated code in your repo
Show HN: I built a single API to post on all social platforms
Show HN: I built a single API to post on all social platforms Building integrations for TikTok, X, Facebook, Instagram, YouTube, and LinkedIn got frustrating with how different every platform is with auth setup, rate limits, and media requirements. So we built Post for Me: one REST API that handles scheduling and publishing posts for TikTok, Facebook, Instagram, X, YouTube, LinkedIn, Pinterest, Bluesky, and Threads.<p>It's built for developers. You bring your own app credentials from the social platform, so you're in control of your data. There's no limitations on account connections, and pricing is based on usage (not some arbitrary limitations). We wanted the API we wish we had when we first started.<p>What it does - A single POST replaces a dozen different calls - Pay-per-post metered pricing (volume discounts) and a free tier (no seat fees) - Cascading configs: set captions/media globally, then override per platform or account<p>Try it at <a href="https://www.postform.dev" rel="nofollow">https://www.postform.dev</a>. Would love feedback, bug reports, or architecture questions!
Show HN: Interactive pinout for the Raspberry Pi Pico 2
Show HN: Interactive pinout for the Raspberry Pi Pico 2 I've been trying to make accessible and beautiful GPIO pinouts since I started one for the Raspberry Pi in 2013 [1]. I've since given the Raspberry Pi Pico [2] and Pico 2 [3] microcontrollers the same treatment when they launched.<p>Recently I've updated these with a new "Upside-down" view to complement the rear view, giving a pinout in the right orientation to match your project.<p>The Pico sites are all hand-coded single HTML pages with supporting CSS and minimal JS. They are set up to optionally install as a "Desktop" web app. They also degrade into a somewhat usable table in lieu of CSS and use vector graphics (for the board itself) to be viewable and printable at any size.<p>Finally, hidden behind "Advanced" is a pinout of the test pads and special function pins!<p>[1] - <a href="https://web.archive.org/web/20130505194305/pi.gadgetoid.com/pinout" rel="nofollow">https://web.archive.org/web/20130505194305/pi.gadgetoid.com/...</a> [2] - <a href="https://pico.pinout.xyz" rel="nofollow">https://pico.pinout.xyz</a> [3] - <a href="https://pico2.pinout.xyz" rel="nofollow">https://pico2.pinout.xyz</a>
Show HN: I built a website to summarize Tech Twitter each day
Show HN: I built a website to summarize Tech Twitter each day Hey folks, I built this as a nightly job that pulls the tweets from a burner account I made, then passes it to GPT-4.1 to summarize alongside some recent headlines: <a href="https://github.com/neelsomani/tweet-insight-daily/blob/main/scripts/lambda_function.py">https://github.com/neelsomani/tweet-insight-daily/blob/main/...</a>
Show HN: Simple wrapper for Chrome's built-in local LLM (Gemini Nano)
Show HN: Simple wrapper for Chrome's built-in local LLM (Gemini Nano) Chrome now includes a native on-device LLM (Gemini Nano) starting in version 138. I've been building with it since it was in origin trials, it's powerful but the official Prompt API is still a bit awkward:<p>- Enforces sessions even for basic usage<p>- Requires user-triggered downloads<p>- Lacks type safety or structured error handling<p>So I open-sourced a small TypeScript wrapper I originally built for other projects to smooth over the rough edges:<p>github: <a href="https://github.com/kstonekuan/simple-chromium-ai">https://github.com/kstonekuan/simple-chromium-ai</a><p>npm: <a href="https://www.npmjs.com/package/simple-chromium-ai" rel="nofollow">https://www.npmjs.com/package/simple-chromium-ai</a><p>- Stateless prompt() method inspired by Anthropic's SDK<p>- Built-in error handling and Result-based .Safe.* variants with neverthrow<p>- Token usage checks<p>- Simple initialization that provides a helper to trigger downloads (must be triggered by user action)<p>It’s intentionally minimal for hacking and prototyping. If you need fine-grained control (e.g. streaming, memory control), use the native API directly:<p><a href="https://developer.chrome.com/docs/ai/prompt-api" rel="nofollow">https://developer.chrome.com/docs/ai/prompt-api</a><p>Would love to hear what people build with it or any feedback!
Show HN: GraphFlow – A lightweight Rust framework for multi-agent orchestration
Show HN: GraphFlow – A lightweight Rust framework for multi-agent orchestration It all started with a conversation among friends about limitations in current multi-agent orchestration frameworks. We faced issues like limited control over agent memory and state, complicated persistence, scaling problems, and lack of type safety in Python-based tools. These challenges inspired us to try something different. The result was GraphFlow, a Rust-based lean framework for orchestrating multi-agent workflows that's simple, scalable, and robust. Its key features include: Graph-based orchestration: Easily define workflows using nodes and edges. Lean Execution Engine: A minimal and efficient graph executor / state machine implementation. Clear Memory Management: Direct and transparent handling of agent states. Simple DB Schema: Easy-to-understand schema for persistence and state tracking. High Performance: Native Rust performance with low overhead and easy scaling. Type Safety: Rust's type system reduces runtime errors. GraphFlow is open-source ofc and aims to solve real-world problems we've experienced firsthand. We'd love your feedback!
Show HN: ParsePoint – AI OCR that pipes any invoice straight into Excel
Show HN: ParsePoint – AI OCR that pipes any invoice straight into Excel Hi HN,<p>I run a small ecommerce shop and, until recently, spent way too many evenings copy-pasting supplier invoices into Excel so my books stayed clean and my expense tracking was granular. It finally hit me that I’d rather invest that time in code than in Ctrl-C/Ctrl-V, so I built ParsePoint.app.<p>Why I built it • Manual invoice entry was swallowing 4 hours every month. • Accounting software APIs felt heavyweight for my use-case; I really just wanted clean rows in a spreadsheet. • I love tinkering with AI models and needed an excuse to spin one up in production.<p>What it does 1. Upload a PDF (or forward an email). 2. ParsePoint’s AI OCR extracts line-items, amounts, tax, dates, etc. 3. Download a ready-to-pivot Excel/CSV file or hit the API to drop the data wherever you like.<p>Under the hood • Frontend: React • API: .NET 8, PostgreSQL • AI layer: an open-source VLLM model fine-tuned for document layouts<p>Outcome so far • My own workload: from 4 hours/month to less than 10 minutes. • Early beta testers (other solo-store owners) report similar time savings and fewer bookkeeping errors. • The pay-as-you-go credit system means no subscriptions or lock-ins—use it only when you need it.<p>I’d love feedback on the tech approach, pricing model, or anything that looks off. All comments welcome, and I’m here to answer every question.<p>Thanks for reading! Marcin – maker of ParsePoint.app
Show HN: Built email parsing for booking confirmations for my travel app – Aruko
Show HN: Built email parsing for booking confirmations for my travel app – Aruko <p><pre><code> A few months ago I shared my travel app here. Today I'm back with something that solved a real technical challenge I was facing. The problem: Parsing booking confirmation emails accurately. Built a parsing system that: - Distinguishes between connection hubs and actual destinations - Captures all segments (flights, hotels, trains) in the right order - Handles different booking site email formats - Creates complete itineraries automatically </code></pre> Happy to discuss more if anyone's interested :)
Show HN: Terminal in Browser
Show HN: Terminal in Browser I wrote a small golang project to render terminal on browser. It is a single binary that can be deployed on any server and you can access the terminal over web.
Show HN: A tool to benchmark LLM APIs (OpenAI, Claude, local/self-hosted)
Show HN: A tool to benchmark LLM APIs (OpenAI, Claude, local/self-hosted) I recently built a small open-source tool to benchmark different LLM API endpoints — including OpenAI, Claude, and self-hosted models (like llama.cpp).<p>It runs a configurable number of test requests and reports two key metrics: • First-token latency (ms): How long it takes for the first token to appear • Output speed (tokens/sec): Overall output fluency<p>Demo: <a href="https://llmapitest.com/" rel="nofollow">https://llmapitest.com/</a> Code: <a href="https://github.com/qjr87/llm-api-test">https://github.com/qjr87/llm-api-test</a><p>The goal is to provide a simple, visual, and reproducible way to evaluate performance across different LLM providers, including the growing number of third-party “proxy” or “cheap LLM API” services.<p>It supports: • OpenAI-compatible APIs (official + proxies) • Claude (via Anthropic) • Local endpoints (custom/self-hosted)<p>You can also self-host it with docker-compose. Config is clean, adding a new provider only requires a simple plugin-style addition.<p>Would love feedback, PRs, or even test reports from APIs you’re using. Especially interested in how some lesser-known services compare.
Show HN: Query your Rust codebase and generate types for anything
Show HN: Query your Rust codebase and generate types for anything Hello HN!<p>As a long-time professional Rust developer. I've always been frustrated by the difficulty and "hackiness" of producing bindings to other languages, whether a frontend, an FFI library, etc.<p>Not just in Rust but in any language.<p>After many years of trying existing solutions and trying to make my own, I've finally developed a solution I'm very happy with.<p>RTK (aka Rust Type Kit) allows you to write Lua scripts that perform queries on your code, such as method calls to Axum's `.route`, function definitions, and more, and then receive rich type information including all argument types, function paths, proc macro attributes, and more.<p>Your Lua script can then read this information and emit an output file in any language of your choosing. Or, you can emit compiler errors and use it as a linter of sorts. You can even directly re-emit Rust code itself and use this as a richer proc macro solution!<p>The code example is a bit verbose, so I encourage you to take a look at the repo's README. I look forward to hearing your thoughts, or any usecases you may come up with!
No other tools from this source yet.