Custom AI Agents for Your Codebase

Today’s top AI Highlights:

1. Build custom AI engineering agents that specialize in your codebase

2. All-in-one LLM toolkit with chat, RAG, and agents in your command line

3. Create entire iOS and Android apps without writing a single line of code

4. 671B DeepSeek-R1 running on a single machine at 286 tokens/s prefill

5. Open source Cursor-like IDE for designers

& so much more!

Read time: 3 mins

AI Tutorials

For businesses looking to stay competitive, understanding the competition is crucial. But manually gathering and analyzing competitor data is time-consuming and often yields incomplete insights. What if we could automate this process using AI agents that work together to deliver comprehensive competitive intelligence?

In this tutorial, we'll build a multi-agent competitor analysis team that automatically discovers competitors, extracts structured data from their websites, and generates actionable insights. You'll create a team of specialized AI agents that work together to deliver detailed competitor analysis reports with market opportunities and strategic recommendations.

This system combines web crawling, data extraction, and AI analysis to transform raw competitor website data into structured insights. Using a team of coordinated AI agents, each specializing in different aspects of competitive analysis

We share hands-on tutorials like this 2-3 times a week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.

Don’t forget to share this newsletter on your social channels and tag Unwind AI (X, LinkedIn, Threads, Facebook) to support us!

Latest Developments

Custom AI Engineering Agents for Your Codebase 🔩🤖

Imagine having AI agents built specifically for your codebases, that understand the nuances of your project, ready to debug tricky errors, answer deep-dive questions about your architecture, or even draft integration tests.

Potpie creates AI agents that specialize in understanding and working with your codebase. The open-source platform builds a comprehensive knowledge graph of your code, enabling agents to grasp complex relationships between components and assist with tasks ranging from debugging to feature development. Each agent is equipped with specific tools to query the knowledge graph and analyze code, making them effective for both small projects and large codebases.

Key Highlights:

1. Ready-to-Use Agent Suite - Jump in with pre-built agents for essential tasks like debugging, code Q&A, and test generation. The debugging agent analyzes stacktraces and provides targeted solutions, while the test agents create contextually aware unit and integration tests that understand your code structure.

2. Knowledge Graph Foundation - Rather than just scanning files, Potpie creates a detailed knowledge graph of your codebase that captures relationships between components. This enables agents to understand how different parts of your code interact and provide more accurate assistance with tasks like debugging and feature development.

3. Flexible Custom Agent Creation - Build your own agents using a straightforward configuration system. Define their goals, tasks, and which tools they can access - like code retrieval, graph queries, and change detection. Each agent can be fine-tuned to handle specific workflows in your development process.

4. Open Source and Fully Customizable - Potpie is completely open-source, giving you full control and the ability to self-host. Deploy it locally, integrate it into your existing development environment, and deeply customize agents and tools to perfectly match your team's needs and coding style.

Build AI Agents and Tools Directly in Your Terminal 🖥️💡

AIChat combines the functionality of a shell assistant, a chat interface, RAG capabilities, and even AI agent creation, all in one package. It's designed to be an all-in-one solution, supporting over 20 LLM providers including OpenAI, Claude, Gemini, Ollama, and Groq, so you can easily switch between models.

Beyond standard chat capabilities, AIChat offers powerful RAG functionality through its built-in vector database, eliminating the need for external services. The tool's flexibility extends to running a local server that exposes chat completion APIs and provides a web playground for direct LLM interactions.

Key Highlights:

1. Function Calling & Agent System - Build custom tools and agents using familiar languages like Bash, JavaScript, and Python. Create complex workflows by combining function calls, document context, and specialized prompts, all manageable through a straightforward YAML configuration system.

2. Enhanced Shell Integration - Convert natural language to shell commands with OS-aware suggestions, complete with alt+e shortcut for intelligent completions right in your terminal. Shell integration scripts available for bash, zsh, PowerShell, fish, and nushell, making it seamless to incorporate into your existing workflow.

3. Unified Provider Interface - Access over 20 LLM providers through a consistent interface. Switch between models by changing a single string parameter (e.g., from 'openai:gpt-4' to 'anthropic:claude-3-5-sonnet'), making it easy to benchmark different models or implement fallbacks.

4. Built-in RAG Capabilities - Process documents from local files, directories, and URLs with the built-in vector database and full-text search engine. Configure custom document loaders for PDFs and DOCXs, with support for reranking to improve result relevance without external dependencies.

Quick Bites

Replit now comes with native mobile app support in early access. You can build iOS and Android apps that you can publish to the App Store without writing any code, powered by Replit Assistant. Go to Replit, create app, pick the “Expo” template, click run, scan the QR code, and get found!

Andrew Ng’s Landing AI has unveiled Agentic Object Detection, a breakthrough approach that lets AI agents detect objects in images using natural language prompts without any training data labeling. Unlike traditional object detection or LLMs that give instant results, this system takes a more deliberate approach - spending 20-30 seconds to reason about the image using reflection, planning, and multi-agent collaboration patterns, similar to how GPT-4V and DeepSeek R1 process complex tasks.

DeepMind has released Alpha Geometry 2 which has surpassed International Mathematical Olympiad (IMO) gold medalist performance, solving an impressive 84% of geometry problems from the past 25 years of IMO competitions—a significant leap from its predecessor's 54% success rate. The system combines Google's Gemini model with a symbolic reasoning engine, showcasing how hybrid approaches might outperform pure neural networks in complex mathematical reasoning tasks.

Tools of the Trade

1. Jam: Browser extension that allows you to create the perfect bug report in just one click. It captures an instant replay of the bug happening on your web app and includes all the developer logs required to cut your bug reporting time by 20x.

2. KTransformers: Python framework that optimizes LLM inference, offering a Transformers-compatible interface, REST APIs, and a web UI with simple integration. It runs DeepSeek-Coder-V3/R1 671B model's Q4_K_M version using only 14GB VRAM and 382GB DRAM, achieving significant speedups over llama.cpp.

3. Onlook: Open source desktop app that lets you visually edit your locally running React app, then write your changes back to code in real-time. It lets you visually develop UI while fully owning your code. Onlook runs locally, deterministically, and only needs a single line config change (npx onlook) to work.

4. Awesome LLM Apps: Build awesome LLM apps with RAG, AI agents, and more to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos, and automate complex work.

Hot Takes

1. what a time to be alive
   you can just prompt for apps
   you can just prompt for deep research
   you can just prompt for music
   you can just prompt for websites
   you can just prompt for podcasts
   this is just the beginning
   next step is to do all of this at the speed of thought ~
   elvis

   
   

2. no thank you but we will buy twitter for $9.74 billion if you want ~
   Sam Altman
   
   

3. There is legitimately no reason to use Figma anymore. ~
   Pietro Schirano

That’s all for today! See you tomorrow with more such AI-filled content.

Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!

Unwind AI - X | LinkedIn | Threads | Facebook

Awesome LLM Apps | Sponsor Us

PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉