Opensource Fully-Managed Agentic RAG Platform

Today’s top AI Highlights:

1. Run AlphaEvolve-style evolutionary AI agents on your own problems

2. Build Agentic RAG with deep reasoning and citations in seconds

3. Anthropic’s interactive guide on the “Timeless Art of Vibe Coding”

4. Vibe code AI agent crews without writing a single line of code

5. This AI reviews your PRs like a senior engineer

& so much more!

Read time: 3 mins

AI Tutorial

Picture this: you're deep in a coding session when you need to update your project documentation in Notion. Instead of context-switching to a browser, navigating through pages, and manually editing content, you simply type "Add deployment notes to the API docs" in your terminal. The magic happens instantly—your Notion page updates without you ever leaving your development environment.

In this tutorial, we'll build a Terminal-based Notion Agent using MCP and Agno framework. This agent will allow you to interact with your Notion pages through natural language commands directly from your terminal, enabling operations like content updates, searches, block creation, and comment addition.

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

Opensource Implementation of DeepMind's AlphaEvolve 🔓♾️

Last week, Google DeepMind dropped AlphaEvolve, an AI agent that uses an evolutionary loop to solve a problem continuously. It doesn't just tweak a few lines of code, it evolves entire codebases using LLMs, automatically discovering new algorithms and optimizing existing ones.

It has already improved Google's data centers by 0.7% efficiency, accelerated Gemini training by 1%, and even found the first improvement to matrix multiplication algorithms since 1969. But here's the thing: DeepMind kept the implementation under wraps.

Now someone has made an opensource version of it. OpenEvolve replicates the same setup: a loop of program mutation, evaluation, and selection, using multiple LLMs. It supports OpenAI-compatible APIs and works on full code files, not just single functions. You can run it locally, define your own problems, and track progress through saved checkpoints.

It’s wild to think DeepMind ran AlphaEvolve a year ago, and now anyone can try the same ideas with better LLMs, out in the open.

Key Highlights:

1. Complete Evolutionary Pipeline - OpenEvolve includes all four core componentsr: a MAP-Elites inspired program database, prompt sampling, LLM ensemble, and evaluation pools. The system runs asynchronous pipelines to maximize throughput and can evolve entire codebases rather than just single functions. 

2. LLM Support - Works with any OpenAI-compatible API, so you can use everything from local models to cloud providers. Gemini Flash 2.0 + Claude Sonnet 3.7 has shown the best results for complex problems.

3. Replication Results - OpenEvolve has successfully replicated key AlphaEvolve benchmarks, including the exact circle packing configuration and transforming basic optimization into advanced algorithms. The two-phase approach (exploration then exploitation) proved most effective for complex problems.

4. Checkpointing and Evolution Tracking - OpenEvolve saves checkpoints every 10 iterations by default. Each checkpoint preserves the best program found at that point, complete with metrics and the full program database state. This makes it easy to compare solutions across different stages of evolution and analyze how algorithms improve.

Opensource Fully-Managed Agentic RAG Platform 🔎📚⚡️

Agentset is an open-source, fully-managed RAG platform, meaning you just upload your documents and get an instant API endpoint to query them. It handles the full RAG pipeline for you. No need to stitch together LangChain or manage vector DBs manually.

The platform handles document parsing across 22+ file types, chunking, embedding, and retrieval using hybrid search and reranking techniques to deliver high-accuracy results out of the box. What sets Agentset apart is its built-in Deep Research with planning, reasoning, and answer validation, taking longer to return a result but surpassing the capabilities and accuracy of traditional RAG systems. You can also self-host the system for full control.

Key Highlights:

1. Agentic RAG - Agentset incorporates planning and reasoning capabilities that go deeper than traditional RAG systems. When you need comprehensive answers, it takes longer to process but delivers superior accuracy by validating responses and expanding search results through multi-step reasoning processes.

2. Production-Ready Stack - The platform uses hybrid search combined with reranking algorithms to maximize retrieval accuracy before you make any customizations. Built-in support for 22+ file types, metadata filtering, and partitioning means you can work with complex document sets and filter results based on specific data subsets.

3. Implementation - Skip the weeks of RAG pipeline development with ready-to-use API endpoints that include automatic citations and source tracking. You get full visibility into which document chunks were used for each answer, with configurable metadata display and links back to original documents.

4. Deployment Options - Choose between their managed service for quick deployment or self-host on your infrastructure for complete control. The platform runs on a modern stack including Next.js, TypeScript, Supabase, and Prisma, making it easy to integrate with existing development workflows.

Quick Bites

CrewAI has released several new updates that make it super easy to build and deploy production-grade multi-agent apps with tools like reasoning and MCP servers.

• You can now vibe code multi-agent applications in the Crew Studio by simply describing what you want to build; the platform then generates the underlying code with agents, tasks, and tools, which you can easily download and modify.

• Add reasoning capabilities to your AI agents by simply setting reasoning=true, allowing them to plan actions before execution, even when the LLM being used is not inherently designed for reasoning.

• You can now connect your CrewAI agents to external tools and services via MCP servers. It supports different transport mechanisms, primarily Stdio (for local servers) and SSE (Server-Sent Events).

Most payment APIs still expect humans, demanding signups, secret keys, and credit cards. AI agents can’t handle any of that - they can’t solve CAPTCHAs, enter payment info, or go through onboarding flows, which makes it hard for them to use paid web services. This setup also hurts developers and small services who miss out on micro-use cases and long-tail agent traffic.
To fix this, Coinbase has launched x402, an open payments standard that lets any HTTP API charge per request using USDC or other tokens, no signup or secrets needed. AI agents can find these endpoints, get a “402 Payment Required” response, make the payment, and try again, all on their own. One line of middleware sets this up, and it’s chain-agnostic, opensource, and fast; payments settle on-chain in ~2 seconds.

India’s Sarvam AI has released Sarvam-M, a 24B parameter opensource LLM. Built on Mistral Small and fine-tuned for Indian languages, math, and programming, it features dual “think” and “non-think” modes. The model shows solid benchmark gains of over 20% on Indic language tasks, math, and coding. Though the adoption has been sluggish, their blog offers a detailed look at their fine-tuning and RLVR pipeline, which might interest those working on post-training workflows.

This is probably the most interesting resource on Vibe Coding to date. Anthropic collaborated with Rick Rubin to release this 81-chapter project that fuses ancient wisdom with the “art” of vibe coding. It’s an experimental, interactive digital book that explores creativity, technology, and coding, filled with amazing art that you can modify further with Claude 4. If you have a quiet moment this week, give it a read. You might walk away thinking differently about how you build, tweak, or even feel your code.

Tools of the Trade

1. Infinitcode.ai: Offload your basic review tasks to this AI PR reviewer that summarizes code changes, flags bugs, security issues, and performance problems, and suggests fixes in real-time. It integrates directly with GitHub, works out of the box without onboarding, and doesn’t store your code.

2. Memoram: A persistent memory layer that stores your conversations and personal data across different AI tools through permissioned access. It uses end-to-end encryption and MemoryKeys to let you decide what information gets shared and with which tools.

3. Davia: Generate web UIs for Python functions by simply decorating them and prompting a UI, without touching frontend code. It runs on FastAPI, supports real-time updates, and works well with AI tools and agent workflows.

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. AGI timelines are very bimodal. It's either by 2030 or bust.

   AI progress over the last decade has been driven by scaling training compute of frontier systems (3.55x a year, 160x over 4 years).

   This simply cannot continue beyond this decade, whether you look at chips, power, even fraction of raw GDP used on training.

   After 2030, AI progress has to mostly come from algorithmic progress. But even there the low hanging fruit will be plucked (at least under the deep learning paradigm).

   So the yearly probability of AGI craters. And we're plausibly looking at 2040+ timelines. ~
   Dwarkesh Patel
   
   

2. Not enough people care about Indic languages and transitively Indic LLMs.

   Tell me the last time we wrote and typed in our regional language. Most people who hold any kind of purchasing capability are not even comfortable reading an Indic script.

   I don't think there is enough incentive to do Indic (anything) except hopium - gambling, astrology, mandir apps, etc. and brainrot content.

   I'd be glad to be proven wrong. But I don't see that happening. ~
   Arpit Bhayani

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 😉