- unwind ai
- Posts
- AI Agents Across Discord, X, and Telegram
AI Agents Across Discord, X, and Telegram
PLUS: DeepSeek R1 + Web search, Manipulate code at scale
Today’s top AI Highlights:
Build AI Agents that work across Discord, Twitter, and Telegram
New Python library to programmatically refactor codebases
DeepSeek R1 on Azure AI, GitHub, and Copilot + PCs
vLLM’s new version delivers 1.7x higher LLM throughput
Free and opensource DeepSeek R1-powered web researcher
& so much more!
Read time: 3 mins
AI Tutorials
Sales teams spend countless hours manually searching for and qualifying potential leads. This repetitive task not only consumes time but also results in inconsistent lead quality. Let’s automate this process to help sales teams focus on what matters most - building relationships and closing deals.
In this tutorial, we'll build an AI Lead Generation Agent that automatically discovers and qualifies potential leads from Quora. Using Firecrawl for intelligent web scraping, Phidata for agent orchestration, and Composio for Google Sheets integration, you'll create a system that can continuously generate and organize qualified leads with minimal human intervention.
Our lead generation agent will help sales teams identify potential customers who are actively discussing or seeking solutions in their target market.
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.
Latest Developments

Eliza is a simple, fast, and lightweight framework for building and deploying AI agents across Discord, Twitter, and Telegram platforms. Built with TypeScript, it provides an extensible platform for developing agents that can interact across platforms while maintaining consistent personalities and knowledge.
What sets Eliza apart is its advanced RAG-based memory system, empowering agents with genuine contextual awareness and long-term memory capabilities. Also, with its modular architecture, you can create custom actions, evaluators, providers, and even your own clients, making it highly adaptable to various use cases.
Key Highlights:
Platform Integration - Deploy agents to Discord (with voice channel support), Twitter, and Telegram using pre-built clients. Each platform integration handles the complexities of message formats and communication protocols. The framework also provides direct API access for custom integrations.
Built-in Memory Management - It comes with a RAG system, combining vector embeddings with relational database storage for long-term memory. The system automatically handles fact extraction, memory retrieval, and context management - no need to build a custom memory architecture from scratch. Compatible with both PostgreSQL and SQLite for flexible deployment options.
Modular Plugin Architecture - Extend agent capabilities through a plugin system that lets you add custom actions, evaluators, and providers without touching core code. Create reusable modules for specific tasks like market analysis, document processing, or platform integration, and share them across different agent implementations.
Local Development to Production Pipeline - Start development with local models like Llama using llama-cpp-python, then seamlessly switch to cloud providers for production. The framework maintains consistent APIs regardless of the model backend, and includes built-in support for hardware acceleration to optimize performance across different environments.

Codegen is a new Python library designed to make large-scale code manipulation easier and more reliable. Built on top of Tree-sitter, Codegen lets you programmatically refactor, analyze, and visualize your Python and TypeScript/JSX multi-million lines codebases with surgical precision.
It builds a detailed graph of your code, understanding dependencies and relationships, then automatically updates imports and references when you make changes. This means you can confidently automate complex tasks like API migrations or feature flag removal, freeing you to focus on higher-level work, and even sets the stage for AI agents to directly manipulate your codebase in a structured way.
Key Highlights:
Automated Code Transformations - Codegen allows you to move, rename, and modify functions, classes, and other symbols, while automatically handling updates to all related imports, references, and dependencies across your project. It ensures that changes maintain correctness and don't break the build.
Deep Codebase Understanding - The library constructs a comprehensive graph representation of your codebase, enabling quick analysis of dependencies, call relationships, and inheritance hierarchies. You can easily find dead code, analyze import structures, and traverse the call graph with built-in APIs.
AI-Ready Code Manipulation - Codegen's structured approach to code modification is designed to be usable by both humans and AI. It provides a consistent interface for AI agents to interact with and modify codebases, opening up possibilities for automated code refactoring and optimization by AI.
IDE and AI Assistant Integration - The tool comes with a system prompt you can use to teach AI assistants like ChatGPT or Claude about its APIs, making them instantly useful for code generation and refactoring suggestions. Also, there is seamless integration with popular IDEs like VSCode, Cursor, and PyCharm, including the ability to configure your IDE to use Codegen's custom Python environment.
Quick Bites
Microsoft has made DeepSeek R1 with 671B parameters available in Azure AI Foundry's model catalog, offering enterprise-ready features with built-in content safety and security evaluations. For developers looking to build with the model, R1 is now also available on GitHub. For local deployment, Microsoft is bringing NPU-optimized versions of DeepSeek R1 to Copilot+ PCs, starting with a distilled 1.5B parameter model, and larger 7B and 14B variants planned for release soon.
JetBrains has launched Junie, an AI coding agent that integrates directly into IDEs. The agent, which achieved a 53.6% success rate on SWEBench's 500-task benchmark, combines JetBrains IDE capabilities with LLMs to handle code generation, testing, and quality verification while adapting to project-specific contexts and coding guidelines.
Simple IDE integration requiring no workflow changes - developers can start with basic tasks and gradually scale to more complex assignments
Built-in code quality features including automated inspections, test writing, and verification processes
Project context awareness enables adaptation to specific coding styles and guidelines while maintaining developer control
LLM serving engine vLLM has released V1 alpha, a major architectural overhaul of its high-performance inference engine. The update delivers up to 1.7x higher throughput compared to the previous version, through significant CPU overhead reductions, enhanced vision-language model support, and seamless integration of features like prefix caching and FlashAttention 3—all while maintaining full backward compatibility with existing APIs.
Here’s an open-source minimalist PyTorch implementation for training small language models from scratch, SmolGPT, featuring flash attention and modern sampling techniques. The lightweight codebase comes with ready-to-use features like mixed precision training and SentencePiece tokenizer integration. Ships with a pre-trained model checkpoint with 8 layers and 512 embedding dimensions for quick testing.
Tools of the Trade
Exa & Deepseek Chat App: Free and open-source chat app that uses Exa's API for web search and Deepseek R1 LLM for reasoning. This app provides a cool and simple chat experience that you can clone and build upon.
Pgchat: Interact with Postgres databases using natural language queries, powered by Inferable and Claude 3.5. The tool runs locally and includes features like query approval workflows, privacy modes, and dynamic schema learning, while keeping database connections secure by proxying queries through your local machine.
Kestra: Open-source orchestration platform to define, schedule, and automate workflows using YAML, with built-in support for running code in any programming language and integrating with 500+ plugins for databases, APIs, and cloud services.
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
OpenAI scrapes the Internet and trains a model with everyone’s data with impunity and without asking for permission — All good.
DeepSeek distills OpenAI models to train their own — outrageous!
You gotta have balls to consider this “proprietary data.” ~
SantiagoPath to AGI
> rag chatbots
> simple tasks
> complex autonomous tasks
> long-running workflows
> remove human-in-the-loop
> human supervision of complex long-running agents
MISSION ACCOMPLISHED ~
Bindu Reddy
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!
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 😉
Reply