Fast and Open Source Cursor Alternative

This week, we’re spotlighting three powerful updates reshaping the developer experience.

First up: Zed, the blazing-fast, open-source code editor built in Rust, now comes with a fully integrated AI Agent Panel. No setup, no indexing—just instant AI-powered code editing, exploration, and customization. You can run local models via Ollama or plug in your own API keys, all with strong privacy controls and flexible permissions.

Then there’s ChatGPT’s GitHub Deep Research beta, letting you connect your repos for real-time code understanding. Ask questions about your codebase, generate docs, and get context-rich answers—straight from your own files, with GitHub permissioning and data privacy built in.

And finally: ZeroSearch, a reinforcement learning breakthrough that lets LLMs learn search and retrieval—without using real search engines. It’s cheaper, more stable, and surprisingly effective at training models to reason through noisy data and surface the right answers.

The bottom line? AI isn’t just a layer on top of your dev tools anymore—it’s becoming the foundation.

Let’s explore the details 👇

Zed: The Fast, Open-Source AI Code Editor with Agent Panel

Zed is an open-source, Rust-based code editor designed for speed and enhanced by powerful AI features, now available through an integrated Agent Panel that allows users to interact with AI agents for tasks like code editing, codebase exploration, and direct code modifications-all without prior setup or indexing. The editor prioritizes privacy, ensuring conversations with the AI remain local unless users opt to share feedback, and includes built-in security prompts for sensitive actions. Users can choose from various AI models, including popular cloud options and custom models running locally via Ollama, and can fine-tune the agent’s permissions and capabilities through customizable profiles and tool access. Zed is free to use as a standard code editor, with AI features available via a limited free plan (50 prompts/month), a Pro plan ($20/month for 500 prompts), or by connecting your own API keys at no extra cost. Currently, Zed offers stable releases for macOS and Linux, with Windows support in progress, and continues to expand with upcoming features like advanced debugging and improved collaboration between programmers and AI agents.

ChatGPT Deep Research: Connect GitHub for Real-Time Code Insights

ChatGPT’s new deep research GitHub connector, now in beta for Plus, Pro, and Team users (with Enterprise access coming soon), allows you to link your GitHub repositories directly to ChatGPT so it can analyze your actual code, documentation, and README files in real time, answering technical questions and citing relevant code snippets from your projects158. After connecting via the Deep research tab or ChatGPT settings, you can select which repositories to grant access, and ChatGPT will respect existing GitHub permissions-only accessing what you authorize and never using your business data for model training by default58. If some repositories don’t appear immediately, it may be due to indexing delays, privacy settings, or admin restrictions, but these can usually be resolved through GitHub settings or manual indexing8. This integration streamlines developer workflows by letting you prompt ChatGPT to break down complex codebases, generate documentation, and surface critical implementation details directly from your own repositories, all while keeping you in control of your data and access.

ZeroSearch: Reinforcement Learning for Enhanced LLM Search

ZeroSearch is a novel reinforcement learning framework designed to enhance the search capabilities of large language models (LLMs) without relying on real search engines, thereby eliminating issues like unpredictable document quality and high API costs. Instead, ZeroSearch uses a supervised fine-tuning process to turn an LLM into a retrieval module that can generate both relevant and noisy documents in response to queries. During training, a curriculum-based mechanism gradually degrades the quality of the generated documents, challenging the model to improve its reasoning and retrieval skills in increasingly difficult scenarios. Extensive experiments show that ZeroSearch not only matches but often surpasses the performance of models trained with real search engines, generalizes well across different model sizes and types, and scales effectively, especially as larger LLMs are used for simulation. This approach offers a stable, cost-effective, and highly adaptable alternative for training LLMs to perform complex search and reasoning tasks.

Hand Picked Video

In this video, we'll explore n8n, a powerful platform for building automated agent workflows.

Top AI Products from this week

• DeckSpeed - DeckSpeed is a revolutionary AI presentation tool that creates professional, personalized slides based on your conversations, without templates. In other words, it's "Cursor for Slides", "McKinsey for everyone".

• Tate-A-Tate - Tate-A-Tate is a no-code platform to rapidly turn ideas into full-stack AI agents — complete with user systems, subscription payments, and complex business logic powered by AI. Launch a business in minutes, using just natural language.

• AI Knowledge Search by Amurex - Our AI-powered search connects Gmail, Google Docs, Notion, Obsidian, and more - so you can find anything across all your knowledge sources instantly, from one simple search bar. No more silos, no more wasted time. Just instant answers.

• Znote - Znote combines ChatGPT, Notion, and a terminal — in one private, local app. Continue your AI chats inside smart notes, run code, build reports, automate workflows. No subscription. Full control. One workspace to think, write, and build.

• RockPic - Identify stones, gems and jewelry in seconds with RockPic — our AI-powered scanner app. Just upload a picture to get accurate insights about any kind of rock.

• SpeakingAI - SpeakingAI is a mobile app that helps you practice and improve your spoken language skills through realistic, interactive conversations with an adaptive AI tutor. Get instant feedback, personalized grammar tips, and learn at your own pace.

This week in AI

• Fidji Simo Joins OpenAI - OpenAI appoints Fidji Simo as CEO of Applications to lead product growth, while Sam Altman focuses on research, compute, and safety.

• Amazon Unveils Vulcan - Amazon introduces Vulcan, its first robot with a sense of touch, to boost safety and efficiency in fulfillment centers by handling items with human-like dexterity.

• Gemini 2.5 Adds Implicit Caching - Gemini 2.5 models now auto-apply up to 75% cost savings for repeated prompt prefixes, no setup needed; just structure static content first for best results.

• Baidu Patents AI to Decode Animal Sounds - Baidu has filed a patent for AI that analyzes animal sounds, behavior, and signals to translate pet emotions into human language; the tech is still in research.

• LTX-Video Real-Time AI Video Gen - LTX-Video is a DiT-based AI model that generates 30 FPS, 1216×704 videos in real time from text or images, supporting keyframes, upscaling, and fast inference.