Drag & Drop to Build AI Agents

Today’s top AI Highlights:

1. Build AI agents without writing code using a visual workflow editor

2. Open-source physics engine runs 430,000x faster than real-time simulation

3. Google releases new Gemini 2.0 “thinking” model that shows its thought process

4. ChatGPT can now directly access your app’s content for more context

5. Visual IDE for React, powered by AI

& so much more!

Read time: 3 mins

AI Tutorials

Building powerful RAG applications has often meant trading off between model performance, cost, and speed. Today, we're changing that by using Cohere's newly released Command R7B model - their most efficient model that delivers top-tier performance in RAG, tool use, and agentic behavior while keeping API costs low and response times fast.

In this tutorial, we'll build a production-ready RAG agent that combines Command R7B's capabilities with Qdrant for vector storage, Langchain for RAG pipeline management, and LangGraph for orchestration. You'll create a system that not only answers questions from your documents but intelligently falls back to web search when needed.

Command R7B brings an impressive 128k context window and leads the HuggingFace Open LLM Leaderboard in its size class. What makes it particularly exciting for our RAG application is its native in-line citation capabilities and strong performance on enterprise RAG use-cases, all with just 7B parameters.

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

Drag-and-Drop to Build AI Agents 👨‍💻

MindStudio is making it easier to build and deploy custom AI agents. The platform offers a visual, no-code workflow editor, so you can create complex, multi-step processes involving language, image, and voice without having to code the core logic.

Access 50+ models from vendors like OpenAI and Anthropic that you can mix and match, fine-tune agents using your own data via RAG and other analysis techniques, plus create apps or serverless functions that deploy these agents across your team. What's compelling is how the system makes the agent development lifecycle so modular— with individual blocks for every step which you can test with generated test cases before publishing.

Key Highlights:

1. Visual Workflow Editor - Construct AI agents using a visual, drag-and-drop interface. The workflow editor allows chaining individual blocks together to execute complex workflows involving different AI models and your custom data, creating agentic apps from start to finish without writing custom code.

2. Testing and Optimization - Built-in tools help validate AI agents’ performance, including a profiler for comparing different models' outputs, costs, and latency. The debugger gives detailed execution logs and billing events, while the evaluations feature lets you generate test cases to ensure consistent results across different scenarios.

3. Production-Ready Architecture - AI agents can be deployed as standalone apps, background automations, or serverless cloud functions. The platform handles scaling, monitoring, and failover between model providers automatically. Every agent gets detailed logging, cost tracking, and usage analytics out of the box.

4. Integrations - Agents can be triggered via REST API, npm package, or integrated with tools like Zapier and Make. The platform provides unified billing across all model providers and includes enterprise features like SSO, audit logs, and compliance controls.

Writer RAG tool: build production-ready RAG apps in minutes

RAG in just a few lines of code? We’ve launched a predefined RAG tool on our developer platform, making it easy to bring your data into a Knowledge Graph and interact with it with AI. With a single API call, writer LLMs will intelligently call the RAG tool to chat with your data.

Integrated into Writer’s full-stack platform, it eliminates the need for complex vendor RAG setups, making it quick to build scalable, highly accurate AI workflows just by passing a graph ID of your data as a parameter to your RAG tool.

Learn more about our production ready RAG tooling here.

Simulate Robots, Fluids, and Soft Bodies in One Python Environment 🤖🌊

Genesis emerged from an ambitious two-year collaboration across 20 research labs, introducing a groundbreaking approach to physics simulation. It is an open-source physics engine that can generate dynamic 4D worlds for robotics and physical AI applications.

Built from scratch in Python, Genesis simulates the complete spectrum of physical interactions - from rigid robotics to fluid dynamics and soft-body deformation. What sets it apart is its remarkable speed of processing millions of frames per second, and its ability to generate rich, interactive environments. It is a universal data engine, autonomously creating everything from robot task scenarios and reward functions to character motions and interactive 3D scenes, paving the way for automated data generation in robotics and physical AI.

Key Highlights:

1. Fast Pythonic Development - Genesis offers a clean Python API that lets you set up complex physics simulations with minimal code. It is 10-80x faster than existing GPU-accelerated stacks like Isaac Gym and MJX. It delivers a simulation speed ~430,000 faster than real-time and takes only 26 seconds to train a robotic locomotion policy transferrable to the real world on a single RTX4090.

2. Unified Physics Environment - Genesis breaks away from the limitations of specialized simulation tools by bringing together rigid body dynamics, fluid simulation, and soft material deformation in one cohesive platform. This allows for testing complex scenarios that better reflect real-world conditions.

3. Versatile Robotics Platform - From industrial arms to legged robots and soft robotics, Genesis supports a wide range of robotic systems and their interactions with different materials and environments.

4. Production Performance - Scale your simulations across thousands of parallel environments for faster training and testing. The engine supports various hardware backends (CPU, NVIDIA, AMD, Apple Metal) and includes photo-realistic rendering capabilities for visualization and synthetic data generation.

Quick Bites

OpenAI has enhanced the ChatGPT desktop app with deeper app integration capabilities. ChatGPT can now directly access content from supported apps to better understand your work context. The app offers two key shortcuts: Option + Space (released previously) to quickly open ChatGPT, and Option + Shift + 1 to give ChatGPT access to your current app's content. You can seamlessly switch between AI models, toggle web search functionality, and even use Advanced Voice Mode, all while staying in your workflow. You can check the supported apps here.

IBM released Granite 3.1, upgraded version of their open-source Granite series of models, including dense (2B, 8B), MoE (1B, 3B), and guardrail (2B, 8B) models, all boasting significant enhancements in performance, 128K context, and functionality. These new models are designed to improve enterprise use cases like tool use, RAG, and agentic AI.

• Flagship Performance: The Granite 3.1 8B Instruct model achieves top scores among open models in its weight class, demonstrating significant improvement on benchmark evaluations.

• Multilingual Embeddings: A new family of retrieval-optimized embedding models are available in four sizes (30M-278M), with multilingual support across 12 languages.

• Availability: Granite 3.1 models are now available on IBM watsonx.ai and through popular platforms including Docker, Hugging Face, LM Studio, Ollama, and Replicate.

ElevenLabs has launched Flash, a new text-to-speech model that generates speech in just 75ms plus network latency, making it ideal for real-time conversational voice agents. Available through their API with model IDs "eleven_flash_v2" (English-only) and "eleven_flash_v2_5" (32 languages), Flash trades some emotional depth for significantly faster performance. Both versions are priced at 1 credit per 2 characters.

Google has not stopped cooking and shipping yet! They have released a new model Gemini 2.0 Flash Thinking in the experimental phase. This model can reason over very complex problems by planning and thinking step-by-step, while showing the thinking process. With a 32K context window and multimodal capabilities, it is currently available for free in the Google AI Studio and via Gemini API.

Tools of the Trade

1. PandasAI: Open-source Python library for conversational data analysis using LLMs. It allows you to query data from various sources, such as dataframes and databases, using natural language, and generate insights, visualizations, and reports.

2. Tempo: An IDE for React that combines AI-powered UI generation with visual and code-based editing, letting teams rapidly prototype while maintaining design control in their production codebase. It works with existing React projects and supports real-time collaboration, Tailwind CSS, and standard dev workflows.

3. Bodo: A compute engine that automatically parallelizes Python data processing code (especially Pandas and NumPy) into optimized binaries without requiring code changes. It achieves 20-240x performance improvements over alternatives while maintaining native Python API compatibility

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-powered coding assistants at the end of 2024:
   • Useful to professional developers
   • Useful for normies who want to learn
   • Useful to generate quick prototypes
   • Useful to keep Dunning-Kruger patients happy
   • Useless in most complex domains
   • Useless to replace humans in any meaningful way ~
   Santiago
   
   

2. Sam Altman is one of the best thing that happened in the last years. because he is so clearly lying at every step that he became a great trust index.
   Now you can just look at whoever are the people that associated or even worse continue to associate with him to know who to avoid and who is dangerous. ~
   Rafa Schwinger

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 😉