Serverless AI Agents with Memory

Today’s top AI Highlights:

1. The first web AI framework for building serverless AI agents with memory

2. Opensource graph-based RAG tool with Perplexity-style chat UI

3. AI agent that learns your coding patterns from your usage

4. Prototype your own multi-agent real-time voice app in under 20 minutes

5. Build a Video RAG under 30 lines of code

& so much more!

Read time: 3 mins

AI Tutorials

Ever wondered what would happen if you let two AI agents play chess against each other? Let's build something fun - a chess game where two AI agents powered by GPT-4o battle it out on the board, while a third agent makes sure they play by the rules.

We'll create a Streamlit app where two AI agents play chess against each other. One controls the white pieces, the other black, and a board proxy agent acts as the referee to keep everything fair and square.

Three main AI agents:

1. Player White: GPT-4o-powered strategic decision-maker

2. Player Black: GPT-4o-powered tactical opponent

3. Game Master: Validation agent for move legality and game state

You'll be able to watch the agents' strategic thinking process in real-time through the terminal output while the game is in progress. The number of turns is configurable - perfect for quick test games or longer battles. Once all turns are complete, you'll see a visual history of the entire game in the UI, with the chess board showing each move (complete with helpful arrows showing piece movement).

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

Local-First AI Agent Framework for Web Developers 🖥️🤖

BaseAI is the easiest way to build serverless AI agents with TypeScript and Node.js, emphasizing local development and clean, composable architecture. The framework enables developers to create AI agents with integrated memory and tools locally, then deploy them as scalable APIs with a single command.

What makes BaseAI particularly interesting is its zero-bloatware philosophy – there's no boilerplate code to wade through, just clean primitives for building AI pipes, tools, and memory systems. Built specifically for web developers, it seamlessly integrates with popular frameworks like Next.js and Vue while providing complete observability for debugging AI like regular JavaScript.

Key Highlights:

1. Web-First Development - BaseAI feels natural for web developers with its TypeScript support and RESTful API design. Local development comes with zero cloud costs, Git integration works out of the box, and you get Chrome DevTools-style debugging for AI components – letting you trace decisions, monitor data flows, and debug outputs right on your machine.

2. Self-Healing Agents - The framework reduces hallucinations by 21% through its self-healing agentic tool-calling system. Your AI agents can detect and correct their own errors by re-querying or fetching additional data when needed. With support for over 100 LLMs, you can implement deep reasoning while maintaining reliability.

3. Full-Stack Memory - Goes beyond basic vector stores to provide a complete memory system with intelligent chunking, context understanding, and advanced similarity search. The built-in attribution system keeps track of information sources, while retrieval testing ensures your AI finds not just similar information, but the most relevant content for each query.

4. Composable AI Building Blocks - Think of AI components like React components for your AI logic – pieces you can snap together to build larger systems. Create a sentiment analysis pipe, connect it with a support ticket handler, add memory modules, and chain multiple tools together. The framework lets you context switch between different memories and handle 200+ tool calls in a single workflow.

This Is the Last Day You’ll Ever Pay for Docusign.

Let’s face it: e-signature hasn’t changed in 25 years, and nobody loves Docusign. That’s where Agree comes in. We’ve combined e-signature and payments into one sleek platform designed to get you paid on time, every time. Use our templates, upload your own, and edit everything directly in our intuitive interface. No clunky workflows, no hidden fees—just faster signatures and fewer headaches. Finally.

Upgrade to Agree Now

Vector Search meets Graph RAG in Open Source 🕸️📚

Autoflow is an open-source RAG system that combines vector search and graph-based RAG capabilities. Built on TiDB Vector Storage and powered by LlamaIndex and DSPy, it streamlines document processing and search implementation.

The platform provides a Perplexity-style chat interface with a built-in website crawler that processes documentation sites, elevating the overall browsing experience. You can also embed Autoflow's search capabilities directly into your websites using a simple JavaScript widget, making it easy to add AI-powered documentation search.

Key Highlights:

1. Document Processing - Native support for multiple document formats including PDF, Word, and Markdown, with an automated crawler that extracts content from documentation sites via sitemap URLs. The system handles document chunking and maintains source references automatically, letting you focus on building your application logic.

2. Deployment Options - Runs smoothly with TiDB Serverless's free tier (25GB storage) for the vector database. Works with both cloud-based LLMs like OpenAI and local models through providers like Ollama. The entire stack can be spun up locally using Docker Compose or deployed to any cloud platform.

3. Integrations - Embeddable chat widget that drops into any website with a simple JavaScript snippet. The widget appears as a floating window, similar to Intercom, and comes with configurable UI elements including example questions and branding options.

4. Built-in Evaluation - Includes evaluation metrics for factual correctness and semantic similarity out of the box. You can upload your own test datasets via CSV and run systematic evaluations to measure and improve your RAG system's performance over time.

Quick Bites

Codeium's Windsurf IDE just dropped "Wave 2," adding powerful web search to its AI agent, Cascade. Cascade can now pull context from the web and specific URLs. Another new feature is automated Memories where Cascade learns your coding patterns to become more personalized. Plus, improved code execution and a problems tab integration now allow you to fix issues in the editor seamlessly.

Google Research released "Titans," a new family of neural network architectures designed for better long-term memory. Titans use a novel module that learns to memorize at test time, and they outperform both Transformers and recent recurrent models on multiple tasks, even scaling to 2M+ context windows. The research team also provides code and training materials.

OpenAI has released a reference implementation for building real-time voice apps with multi-agent flows using their Realtime API. This allows you to prototype a voice app in under 20 minutes, demonstrating sequential agent handoffs, background escalation for high-stakes decisions, and state machine prompting. The demo app includes examples for authentication, customer service, and more

Tools of the Trade

1. tl;dw: API for building AI applications that understand video content. You can easily implement video-based RAG systems and create AI agents that can analyze video data with minimal code. The API supports semantic video search, moment-level RAG, intelligent scene segmentation, and various video formats.

2. Nimbus: AI agent that monitors your AWS, GCP, and Azure cloud resources and sends daily recommendations via email to reduce costs. It identifies inefficiencies and allows you to implement changes with one-click approval.

3. DataChain: Python-based AI-data warehouse for transforming and analyzing unstructured data like images, audio, videos, text and PDFs. It integrates with external storage (e.g. S3) to process data without data duplication and manages metadata in an internal database.

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. Agents are going to have their own scaling laws
   Multi agent scaling, revenue generation, task completions, etc. ~
   Logan Kilpatrick
   
   

2. AI coding replacing programmers is the modern version of calculators replacing mathematicians. ~
   Santiago

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 😉