Agentic Memory for AI Agents

Today’s top AI Highlights:

1. AI agents organize and update their own memory systems

2. Forget function-calling, this framework uses agents as tools

3. Google’s Data Science Agent in Google Colab now available

4. Open-source LLM observability tool for lazy devs

5. Low-memory, fast Firecrawl alternative for LLM markdown

& so much more!

Read time: 3 mins

AI Tutorials

LLM App

Finding the perfect property involves sifting through countless listings across multiple websites, analyzing location trends, and making informed investment decisions. For developers and real estate professionals, automating this process can save hours of manual work while providing deeper market insights.

In this tutorial, we'll build an AI Real Estate Agent that automates property search and market analysis. It helps users find properties matching their criteria while providing detailed location trends and investment recommendations. This agent streamlines the property search process by combining data from multiple real estate websites and offering intelligent analysis.

Tech Stack:

• Firecrawl's Extract Endpoint to collect structured data from websites

• Agno (formerly Phidata) for building the AI agent

• OpenAI GPT-4o as the LLM

• Streamlit for a clean, interactive web interface.

AI Workflow

This workflow combines Grok-3's image generation capabilities with Pika AI's video animation features to create stunning transformation videos that show the evolution from vintage to modern aesthetics. Perfect for photo restorations, concept visualizations, or creative storytelling.

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

Agentic Memory for AI Agents 🧠✨

Agentic applications can be truly personalized and helpful when they can maintain context during long-term interactions. Memory solutions for AI agents, even those incorporating graph databases like Mem0, fall short due to their reliance on pre-defined schemas. This limits their ability to adapt to new information.

A-MEM is a novel agentic memory system for AI agents that goes beyond simple storage and retrieval. It allows agents to dynamically organize, link, and—most importantly—update their memories over time, mimicking aspects of human learning. A-MEM draws inspiration from the Zettelkasten note-taking method, creating interconnected networks of knowledge that continuously evolve as new information arrives. This improves performance in long-term interactions, especially in complex, open-ended tasks.

Key Highlights:

1. Agentic Memory Management - A-MEM isn't just passive storage. The AI agent actively participates in memory organization, making it highly adaptable to different tasks and new information. This is controlled via an agentic memory update mechanism.

2. Automatic Linking and Evolution - When new information arrives, the AI agent automatically generates notes with contextual descriptions and tags, and decides on its own how to link this new information to existing memories. This continuous evolution is a crucial differentiator.

3. Zettelkasten-Inspired Structure - The system uses principles from the Zettelkasten method, creating "atomic" memory notes with rich attributes (descriptions, keywords, embeddings). This facilitates efficient searching and relationship discovery.

4. Open Source - The complete source code is available on GitHub, with straightforward setup instructions for running experiments. You can test it directly on the LoCoMo dataset.

Build AI Agents That Call Other Agents Directly 🤖📞🤖

Agentic is an opinionated Python framework for building autonomous AI agents that can understand natural language and use tools on your behalf. Unlike other frameworks developed before having experience, Agentic incorporates lessons learned from building and running hundreds of agents over the past year.

It's designed to address common pain points with existing agent frameworks, like managing long-running processes, handling human-in-the-loop scenarios, and scaling beyond simple prototypes. Agentic features an event-driven architecture, distinguishing it from typical synchronous, function-calling-based approaches.

Key Highlights:

1. Event-Driven, Not Just Function Calling - Agentic uses an asynchronous, event-driven model. This means your agents can handle long-running tasks, easily wait for external events (like a human response or an API call to complete), and are not limited by the synchronous nature of typical function calling.

2. Simplified Architecture - Agentic provides clean abstractions with just a few core concepts (Agent, Tool, Thread, Run) and a thin layer over LLM APIs. The entire agent loop is around 200 lines of code, making it easy to understand what's happening under the hood.

3. "Tools are Agents" - Agentic blurs the line between tools and agents. You can seamlessly use other agents as tools within your main agent, enabling complex, hierarchical agent systems without complicated orchestration.

4. Production-Ready Features - Agentic includes built-in logging, context management to handle token limits, human-in-the-loop capabilities, and support for agent teams. Every agent automatically gets an API interface, and the system tracks context length, token usage, and timing data in a standard format.

Quick Bites

DeepSeek AI has open-sourced 3FS (Fire-Flyer File System), a high-performance distributed file system specifically designed for AI workloads, achieving 6.6 TiB/s read throughput on a 180-node cluster. Alongside 3FS comes Smallpond, a lightweight data processing framework built on DuckDB and 3FS, capable of handling petabyte-scale datasets while maintaining simplicity without the complexity of long-running services or infrastructure overhead.

Google’s Data Science Agent in Google Colab is now available. This Gemini-powered agent can create notebooks, removing tedious setup tasks like importing libraries, loading data, and writing boilerplate code fully autonomously. Open a blank Colab notebook > Upload your data file> Describe what kind of analysis or prototype you want to build > Just relax and watch your agent generate the necessary code, import libraries, and analysis.

Hugging Face now integrates with Arize Phoenix - a comprehensive platform to trace, evaluate, and debug AI agents in real-time. The new tooling allows you to visualize agent workflows, monitor performance metrics, and implement various evaluation methods—including LLM-as-judge techniques—to assess everything from response relevance to factual accuracy.

Tools of the Trade

1. Qodo Gen: AI coding assistant that helps you write, understand, test, and review code across all programming languages. It integrates directly into IDEs like VSCode and JetBrains. It can help with code explanations, quality improvements, bug detection, test generation, and docstring creation.

2. Sublingual: Open-source LLM observability tool that automatically intercepts and logs all LLM calls from your application without code changes. It captures prompts, responses, parameters, and even attempts to extract prompt templates, providing a local dashboard for analysis and running evaluations.

3. Pathik: A high-performance, open-source web crawler, built with Golang and offering Python bindings, consuming 134x less memory than Playwright. It extracts website content into clean, LLM-ready MD, optimizing web data for use in RAG systems and LLM fine-tuning, with minimal memory usage.

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. bro wtf anthropic launched their devin/open hands-style SWE agent and barely anyone noticed because it doesn’t have a UI and lives in the terminal
   also it’s really really good, will one-shot diffs well over 1000 lines at a time ~
   James Campbell

   
   

2. I’ve never used a better model than GPT-4.5 fair to say it's "Baby AGI." ~
   Ashutosh Shrivastava

   
   

3. Pretty incredible to watch Apple not only completely lose the AI race, but barely even compete in it ~
   Alex Cohen

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 😉