Refact.ai

Shipping ideas is way faster when you don’t have to code every line yourself. Just delegate it to an AI Agent! Today’s Weekly Showcase proves it with two solid projects, built with Refact.ai: • A smart dictionary that finds words based on descriptions and provides a clear pronunciation. • A GitHub scanner that detects and tests exposed API keys. 🎬 Live on YouTube. Link in comments. Tune in and see what you could be building with AI!

💥 It's Friday, and our new Changelog is live! • Team-level Memory in Refact.ai: Join the waitlist to bring shared knowledge to every Agent on your team! • Webinar recording: Context & prompt engineering that works. • Built with Refact: Real projects from our Weekly Showcase. • OSS/Enterprise updates. Explore the full edition👇

Ever feel like you’re stuck in Groundhog Day when using AI? Yesterday you explained your code and project, and it helped. Today all the context is gone, and the same mistakes repeat. 🧠 The solution is Memory. It lets AI Agents remember and learn over time: • Keeps context across sessions • Saves what worked and reuses it when similar tasks come up • Builds a team-level knowledge base, so proven solutions discovered by any Agent on the project help the entire team. Developers learn from experience. Their AI Agents should too. We just published a deep dive on how long-term memory can change the way you build software with AI. Link in comments — and there’s something useful for your team at the end 👀

Join us tomorrow for the Refact.ai Weekly Showcase! (Spoiler: 🫁 , 🎬 , ☸️ ) Every Thursday, users demo real projects built with Refact.ai Agent + explain how autonomous programming shaped each workflow. 🗓 July 3, 5:00 PM CEST This week’s lineup: • A breathing app to help relax and focus • A content tool that autogenerates Reels & Shorts • A full Kubernetes disaster recovery workflow — with backup, restore, testing, and monitoring. 🎁 Bonus: stay till the end for a chance to WIN a branded Refact.ai T-shirt. 📍 Live on YouTube, link in the comments. Hit “Notify me” so you don’t miss it!

Something’s getting built live today 😎 And you're invited to join in! 🔥 We're hosting a Hack Night on Discord: a live coding session where we develop a project with Refact.ai Agent, adding features every two weeks. Share ideas and watch them turn into real features on the spot. Or just watch and explore autonomous programming with Refact.ai in action. Everything is open, transparent, and collaborative, with a public repo so you can contribute between sessions!   📅 Today, July 1 🕟 16:30 CEST 🎙️ Led by our Ambassador, Utkarsh Umre 📍 Live on Discord. Follow the link in the comments and click “Interested” to get notified when we start.

Real value in an AI Agent for programming? It gets things right on the 1st attempt. 🧠 That’s what pass@1 means in SWE-bench — and how we ran Refact.ai Agent on the benchmarks: one chance to solve each task, no retries. It’s not required, but we believe it’s the most honest way to evaluate how useful AI Agent is in everyday workflows. Developers shouldn’t have to double-check AI-written code or babysit its fixes — they need results they can trust. 🏆 SWE Verified (Python): 74.40% — best under pass@1 🏆 SWE Multimodal (screenshot-based issues): 35.59% — #1 overall Refact.ai now leads across SWE-bench Verified, Multimodal, and Lite. Reliable AI programming starts here📍

The first rule of the open-source club is: you do talk about open-source — and you build with the community 🫂 Refact.ai is an open-source AI Agent. If you’re excited to shape how developers code with AI, we’d love your contributions! Here are a few open issues you can jump into right now: • Add support for new models or providers • Expand the MCP catalog • Improve documentation • … Or request new features. Plus, we’ve got a short contribution guide to help you get started. GitHub link’s in the comments. PRs welcome 🫶

🔥Refact.ai just achieved 1st place on SWE-bench Multimodal! This new top score confirms that Agent is effective across various programming languages. What makes this benchmark more advanced: • Issues include visuals — screenshots, mockups, UI errors, diagrams, etc. • JavaScript-focused — covers challenges missing in the Python-only SWE-bench Verified and requires true cross-domain action. SWE-bench M tests whether an AI Agent can interpret both text & images and ship a solution. Open-source Refact.ai, powered by Claude 4, nailed it. #1 result on a key industry benchmark. The number that speaks for itself. Excited to celebrate this milestone with you!

Always inspired by what devs build with Refact.ai Agent — real projects where autonomous AI does the programming, so people can focus on ideas and creativity. At yesterday’s Weekly Showcase, users shared how our Agent helped create: 🏋️ A fitness app with AI-powered movement tracking + Web3 leaderboards 🧠 An explainer tool that turns dev topics into animations and audio Big thanks to everyone who joined! The recording’s now on YouTube — as always, check the comments for the link.

Join us today for Refact.ai Weekly Showcase: users share projects made with our Agent. On the agenda: • Live demos: a study app & a fitness app (yep, squats included 🏋️) • New ideas & AI programming tips • Giveaway for attendees. 🖥️ Today [June 19], 5:00 PM CEST 🖇️ Watch live — link in comments.