AI Agent that Thinks in Code

Today’s top AI Highlights:

1. Ultimate toolkit for RAG pipelines with small, specialized models

2. Build AI agents that think in code

3. Microsoft’s upgraded AutoGen framework for better AI agent interactions

4. ChatGPT learns to set reminders for you

5. RAG on codebases is fine, but you need more

& so much more!

Read time: 3 mins

AI Tutorials

LLMs are great at generating educational content and learning roadmaps, but they struggle with complex, multi-step workflows. While you could ask an LLM to create a curriculum, then separately ask it to design exercises, then manually compile resources – this process is tedious and requires constant human coordination.

In this tutorial, we'll solve this by building an AI Teaching Agent Team. Instead of isolated tasks, our AI agents work together like a real teaching faculty: one creates comprehensive knowledge bases, another designs learning paths, a third curates resources, and a fourth develops practice materials.

The user just needs to provide a topic. Everything is automatically saved and organized in Google Docs, creating a seamless learning experience without manual overhead. We are using Phidata and Composio to build our AI agents.

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

End-to-End RAG Framework With Small Specialized Models 🧰

llmware is an integrated framework for building RAG applications with small, specialized models, tailored for enterprise environments. It provides end-to-end capabilities for entire RAG lifecycle, from document parsing and embedding to model prompting and agentic workflows, all designed with a focus on secure, local processing and scalability.

It supports over 50 pre-trained models specifically for RAG and seamless integrations with major vector databases. llmware streamlines everything from quick prototypes to production deployments by combining parsing, embeddings, model management, and agent orchestration in one cohesive package.

Key Highlights:

1. Specialized Model Toolkit - LLMWare ships with 150+ models including over 50 RAG-optimized models. SLIMs (Structured Language Instruction Models) generate Python dictionaries and lists for extraction, classification, and summarization. Models range from 1-7B parameters and are designed to run locally.

2. Data Pipeline - The framework comes with built-in support for 15+ file types including PDFs, Office docs, and audio files. Text chunking, metadata extraction, and vector storage are automatically managed through a unified Library interface. You can use SQLite locally and transition to PostgreSQL/MongoDB for production without changing application code.

3. Integrated Agent Framework - Build multi-step workflows using the LLMfx class. The framework includes 18 SLIM models fine-tuned for function calling. You can easily chain models together while maintaining state and managing prompt history.

4. Production Features - Enterprise-ready features include comprehensive logging, state management for long-running processes, built-in evaluation tools through Ragas, and deployment options from single machine to distributed clusters. The framework supports 9 vector databases including Milvus, PGVector, and Redis, with built-in connection management and query optimization.

Hire Ava, the Industry-Leading AI BDR

Ava automates your entire outbound demand generation so you can get leads delivered to your inbox on autopilot. She operates within the Artisan platform, which consolidates every tool you need for outbound:

• 300M+ High-Quality B2B Prospects

• Automated Lead Enrichment With 10+ Data Sources Included

• Full Email Deliverability Management

• Personalization Waterfall using LinkedIn, Twitter, Web Scraping & More

Book a demo to see what Ava can do.

Build AI Agents That Express Plans as Code 🤖📝

AgentScript introduces a new paradigm for building AI agents by shifting the planning process from iterative LLM calls to generating executable code. This open-source framework prompts an LLM to create an execution plan in JavaScript (a subset) which is then converted into an Abstract Syntax Tree (AST) that runs in a dedicated runtime. Instead of step-by-step interactions, the agent logic is encapsulated as code, giving a fresh approach for creating agents with complex workflows.

It helps you build more sophisticated agent behaviors while keeping humans in the loop through built-in pause/resume capabilities and tool-level state management. The framework eliminates the need for multiple LLM calls and sandboxing.

Key Highlights:

1. Code-First Planning - Let your AI agents generate JavaScript code that outlines their execution plan upfront. The framework handles AST transformation and provides a safe runtime environment, so you don't need sandboxing. You can implement loops, conditionals, and local variables just like regular code while maintaining control over execution.

2. Built-in State Control - Pause execution at any point, serialize the state to your database, and resume later - perfect for handling human approvals or long-running tasks. Each tool maintains its own state and the entire execution state is accessible for inspection and modification. The framework automatically handles durability and failure recovery.

3. Tool Integration - Mix deterministic functions with LLM-powered tools seamlessly in your agent workflows. The framework works with tools built using LangChain or other libraries. Tols can store state, wait for user input, or handle time-based operations.

4. Developer Experience - Monitor your agents through execution timeline visualization, state inspection panels, and comprehensive tracing capabilities. The framework runs in both Node.js and serverless environments, making it practical for production deployments.

Quick Bites

Microsoft has released AutoGen v0.4, completely redesigning their multi-agent AI framework to enhance code quality, robustness, and scalability in agentic workflows. The update addresses previous architectural constraints and API inefficiencies, while introducing new capabilities.

• The new asynchronous messaging enables agents to communicate through both event-driven and request/response patterns.

• You can now customize their systems with pluggable components including agents, tools, memory, and models.

• Built-in metric tracking, message tracing, and debugging tools for monitoring and control over agent workflows, with support for OpenTelemetry for industry-standard observability. 

• The new AutoGen Studio introduces a low-code interface with real-time agent updates, mid-execution control, and an intuitive drag-and-drop multi-agent team builder for quick prototyping of AI agents.

Google’s Gemini now has a Saved Info feature where you can save information about your preferences to get more context-aware and personalized responses from Gemini. Add new info here or ask Gemini to remember something during a chat. This feature is currently available to Gemini Advanced subscribers in English language only.

OpenAI has released a new feature “Tasks” in beta that lets you schedule future or recurring actions like workout plans, classes, and reminders. Either ask ChatGPT to remind you in the chat or schedule a task on the Tasks page. Tasks has been rolled out to Plus, Pro, and Teams users and will be available soon to all users with a ChatGPT account.

Tools of the Trade

1. Greptile: AI that understands your codebase and can do things like review PRs, diagnose failing tests, and Q&A. Unlike “RAG on codebases”, it parses abstract syntax trees and uses "agentic search" to trace code relationships. Use it via the web app, VS Code plugin, CLI, or integrate with tools like Slack, Linear, and GitHub.

2. Featherless: Serverless AI inference platform providing API access to a growing library of Llama-based Hugging Face models, starting at $10/month. It uses FP8 quantization and a custom stack for fast model switching. It offers an OpenAI-compatible API.

3. Freeact: Lightweight Python library that enables LLMs to operate as autonomous agents by executing actual Python code rather than using restricted formats like JSON. It allows agents to dynamically install Python packages at runtime, learn from feedback to build reusable skills, and execute code securely in a Docker-based sandbox.

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. AI will run your life whether you like it or not. That’s where things are clearly headed. ~
   Aravind Srinivas
   
   

2. I still don’t know what an ai agent is. Is it that thing called Kubernetes? ~
   Suhail

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 😉