Multi-Agent Team Simulates Real World Trading Firm

Today’s top AI Highlights:

1. Drop a TypeScript file and get a production-ready MCP tool

2. Opensource multi-agent team that simulates a real-world trading firm

3. Columbia students built Truely to catch Cluely cheaters during interviews

4. Claude Code Pro hitting rate limit? Hack it while you sleep

5. This Excel agent can complete most knowledge work faster than humans

& so much more!

Read time: 3 mins

AI Tutorial

Building tools that truly understand your documents is hard. Most RAG implementations just retrieve similar text chunks without actually reasoning about them, leading to shallow responses. The real solution lies in creating a system that can process documents, search the web when needed, and deliver thoughtful analysis. Moreover, running the pipeline locally would reduce latency and ensure privacy and control over sensitive data.

In this tutorial, we'll build a powerful Local RAG Reasoning Agent that runs entirely on your own machine. You'll be able to choose between multiple state-of-the-art opensource models like Qwen 3, Gemma 3, and DeepSeek R1 to power your system.

This hybrid setup combines document processing, vector search, and web search capabilities to deliver thoughtful, context-aware responses without cloud dependencies.

We share hands-on tutorials like this every 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) to support us!

Latest Developments

Easiest Way to Build and Ship MCP Applications 🛠️🖇️🚀

Building MCP servers shouldn't feel like wrestling with protocol specifics and complex configurations. You write a simple TypeScript function, drop it in a folder, and suddenly you have a production-ready MCP tool that works across any AI client.

xmcp framework treats MCP development like modern web development - with file-system routing, hot reloading, and zero-config deployment options that actually work.

The framework automatically discovers tools in your directory and handles all the MCP protocol complexity behind the scenes. Whether you're building for local development with STDIO transport or deploying to production with HTTP, xmcp provides the same developer experience with built-in authentication, middleware support, and one-command deployment to platforms like Vercel.

Key Highlights:

1. File-system magic - Drop TypeScript files in /src/tools/ and they automatically become MCP tools with schema validation, metadata, and proper typing through Zod integration.

2. Transport flexibility - Switch between STDIO for local development and HTTP for production deployments without changing your tool code, plus built-in support for authentication and custom middlewares.

3. Developer experience - Hot reloading during development, one-command scaffolding with npx create-xmcp-app, and seamless integration with existing Next.js or Express applications.

4. Production ready - Deploy anywhere with built-in Vercel support, custom webpack configuration options, and experimental OAuth provider implementation for enterprise authentication needs.

AI Agents Simulating a Real Trading Firm 🤑📝🌐

A team of AI agents that simulates an actual trading firm.

Researchers from UCLA, MIT, and Tauric Research have built TradingAgents, a framework that mirrors how actual trading firms operate, complete with specialized analysts, researchers who debate bullish vs bearish positions, traders making decisions, and risk managers keeping everything in check.

The framework proved its worth during backtesting from January to March 2024, consistently outperforming traditional strategies like MACD and buy-and-hold approaches.

The implementation is interesting: they strategically mix quick-thinking models (GPT-4o-mini) for data retrieval with deep-thinking models (o1-preview) for analysis and decision-making. Communication happens through structured JSON reports rather than endless message histories, eliminating the "telephone effect" that kills most multi-agent systems.

Key Highlights:

1. Trading team architecture - Seven specialized agent types with clear interfaces: analysts use tools like FinnHub API, YFinance, Reddit, etc., while traders synthesize structured reports into trading decisions using configurable LLM backends.

2. Hybrid communication - Combines structured JSON outputs for control flow with natural language debates for complex reasoning, preventing context corruption while maintaining decision quality.

3. Model selection - Quick models handle API calls and summarization while reasoning-heavy models (o1-preview) tackle decision-making, optimizing both cost and performance across the pipeline.

4. Deploy - Fully opensourced with CLI interface, supports model swapping, includes comprehensive backtesting simulation. Just give it a ticker and watch these agents research, reason, and report.

Quick Bites

1. What happens when you give up on waiting for Tesla's Optimus? A YC startup is selling fully opensource humanoid robots for under $9,000, complete with hardware designs, software stack, and a promise of free upgrades until the bot achieves full autonomy. K-Scale Labs built their first walking prototype in just two months using 3D printers and parts from Amazon and Alibaba, targeting indie developers who want to hack on humanoids without the typical $50k+ price tag. The startup's bet is that a distributed army of tinkerers will crack household autonomy faster than billion-dollar labs.
   
   

2. While Cluely raised $15M to help people "cheat on everything," a pair of Columbia students built the antidote, Truely. Truely is a free, opensource detection tool that automatically joins meetings as a monitoring bot, scanning for AI assistance tools and alerting participants when suspicious processes are detected. When Truely spots flagged applications, it immediately alerts everyone in the meeting chat with timestamped warnings.
   
   

3. Most models force you to choose between reasoning capability and efficiency. Hugging Face just hit the sweet spot with SmolLM3, a 3B model that outperforms Llama-3.2-3B and Qwen2.5-3B while staying competitive with larger 4B alternatives (Qwen3 & Gemma3). SmolLM3 brings dual-mode reasoning (think / no_think), 128k context length, and multilingual support across six languages. The release includes architecture details, exact data mixtures, and a complete methodology for building competitive small-scale reasoning models.

4. While everyone focuses on building smarter agents, the real bottleneck often lies in how they manage their context windows during task execution. LangChain's latest video on Context Engineering dissects four proven strategies (write, select, compress, isolate) used by popular agents to handle the constant flow of instructions, knowledge, and tool feedback. The breakdown shows how deliberate context curation can turn unreliable agents into consistent performers.

Tools of the Trade

1. Claude Auto Resume: A shell script that automatically resumes Claude CLI tasks when usage limits are lifted. It detects restriction messages, waits with a countdown timer, then continues execution WITHOUT asking for permission.

2. Shortcut: An autonomous Excel operator that can complete professional-level spreadsheet work, including financial modeling championship cases, in minutes in a single shot. It provides a complete Excel environment with native functionality plus AI automation for data analysis, model building, and formula generation.

3. Merit: Payment interface that allows open-source project maintainers to identify and compensate high-impact contributors based on their GitHub activity and PR contributions. The platform handles tax administration and uses stablecoins for global payouts.

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

Hot Takes

1. a ton of ppl are overengineering ai products right now & worse many are showcasing tech instead of solving anything real.

   you gotta start simple. stupidly simple. then layer in narrative & features like you’re sending love letters. each iteration should feel handcrafted, like a letter pressed card with your name on it. less “look what ai can do.” more “this was made just for you.”

   the best ai experiences aren’t maximal, they’re specific. this is what we are doing. ~
   signüll
   
   

2. Quitting programming as a career right now because of LLMs would be like quitting carpentry as a career thanks to the invention of the table saw. ~
   Simon Willison

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

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 😉