- RoboRuby
- Posts
- Ruby AI News - December 19th, 2025
Ruby AI News - December 19th, 2025
2026 will be Ruby's best year yet
Matt Solt
December 19, 2025
Welcome to the 21st edition of Ruby AI News! This edition features a RubyLLM backed startup, an AI YouTube video repsitory, Videos on Ruby AI, and much more.
That’s all for Ruby AI News in 2025. Have a great holiday and a safe and happy new year. See you in 2026!
Contents
Top Stories
Community Collaboration
It’s been an incredible year. I’ve watched the number of AI gems double in size, the amount of Ruby AI content quadruple, and Ruby come back to life thanks to the dedicated community members and conference organizers around the globe.
The next year is what we make it. Embrace the changes that are coming. The capabilities that “vibe” coding allow are here. Collaboration with your fellow Rubyists is the biggest advantage we have. Reach out to someone you don’t know and ask if they want to work together. I promise you it is the way forward.
It’s already happening. Kieran Klaassen built an adapter connecting Vicente Reig's DSPy.rb with Carmine Paolino's RubyLLM gem, adding access to virtually all AI providers through a single integration.
Carmine’s RubyLLM gem has given the community a foundation to build upon, and now he’s launched Chat With Work. It is an AI workplace assistant in that searches across Google Drive, Slack, Notion, and GitHub using natural language queries. Carmine is building something special, test out the service, give him some feedback, and help him build something great!
Tuby
Alan Alves created Tuby.dev, a curated video platform that solves YouTube's algorithm problem for Rails learners. As Alan explained in his Reddit announcement, YouTube's algorithm is optimized for retention rather than learning, forcing developers to skip through disingenuous recommendation to find relevant, quality or advanced content. The platform uses a three-stage AI pipeline: RSS feed monitoring from major Rails channels, an initial AI filter that analyzes titles and descriptions to discard off-topic content, and a differentiating deep analysis stage that downloads videos and sends them to Gemini Vision AI. Unlike platforms that only summarize transcripts, Tuby's system performs OCR on the actual video frames to identify gems, versions, and patterns that authors use but forget to mention aloud. The Hacker News discussion highlighted the technical implementation using Gemini for code extraction. Built with Rails, PostgreSQL, Inertia.js, and Shadcn, the platform currently indexes 80+ videos.
RapidRuby AI
Pete Hawkins announced that Rapid Ruby courses and screencasts are now freely available, aiming to give as many developers as possible access to his Ruby on Rails, AI, and Hotwire educational content. In addition to his Rails Twitter Clone course, Pete has generously started releasing the videos in his AI on Rails series:
Intro to Building Dynamic Multi-Agent AI Systems in Rails Pete introduces his Rapid Ruby course on building multi-agent AI systems using LangChain.rb, covering intelligent agent creation, custom tools, and LLM integration.
Setting Up LangChain.rb Pete walks through configuring LangChain.rb in a Rails application to process LLM requests.
Persisting Chat History Pete Hawkins demonstrates storing AI chat history in a Rails database to enable continuous conversations with LLM agents across sessions.
Dynamic AI Prompts Pete shows how to back an AI agent class with ActiveRecord and store prompts in the database, enabling new agents to be created and tested without code deployment.
Be sure to subscribe to his YouTube channel to ensure you don’t miss the rest of the videos in the series!
Need to Know AI News
Relevant AI News for Ruby Developers
State of AI: An Empirical 100 Trillion Token Study OpenRouter analyzed over 100 trillion tokens of LLM usage across 300+ models, finding that programming surged to more than 50% of LLM usage with Claude dominating 60% of that share. The study revealed open-source models captured roughly 30% of usage, agentic inference became the default with reasoning models exceeding 50% of tokens, and Chinese models like DeepSeek and Qwen reshaped competitive dynamics.
Donating the Model Context Protocol and Establishing the Agentic AI Foundation Anthropic is donating MCP to the Agentic AI Foundation, a fund under the Linux Foundation. In its first year, MCP grew to over 10,000 active servers, with adoption across major AI coding platforms.
Introducing x402 V2: Evolving the Standard for Internet-Native Payments x402 released version 2 of the protocol that embeds payments into HTTP using the 402 status code. Since its May 2025 launch, x402 has processed over 100 million transactions, with the new version adding wallet-based sessions and extensible architecture to support AI agents purchasing compute and data on-demand.
Introducing the Agentic Commerce Suite Stripe announced a platform for businesses to sell through AI agents, featuring hosted ACP endpoints for product discovery, Shared Payment Tokens for secure transactions, and fraud protection via Stripe Radar.
Chrome in Claude Anthropic launched a beta browser extension enabling Claude to navigate, click, and fill forms directly in Chrome. The extension integrates with Claude Code via the /chrome command, allowing automated testing of code in the browser and visibility into client-side errors through console logs.
Agent Skills Anthropic released Agent Skills as an open standard for giving AI agents new capabilities through discoverable folders containing instructions, scripts, and resources. The format has been adopted by Cursor, Claude Code, GitHub, VS Code, and OpenCode, enabling skill authors to build once and deploy across multiple agent products.
Announcements
ActiveRabbit Alex Shapalov launched an open-source AI-powered performance monitoring and error tracking platform for Rails applications. The tool provides real-time monitoring across error types with AI root cause analysis and auto-fix suggestions under the Once O’Saasy license.
NanoGPT-rb Chris Hasiński ported Andrej Karpathy's nanoGPT to Ruby using torch.rb for training GPT-2 style language models from scratch. The implementation includes Flash attention optimization, GPU acceleration via MPS and CUDA, and achieves coherent text generation after roughly 2000 iterations on Apple Silicon.
Leva Kieran Klaassen added DSPy.rb integration to the Leva Rails prompt management engine, enabling automatic prompt optimization with Bootstrap, GEPA, and MIPROv2 strategies. The latest release includes a PromptOptimizer service, SignatureGenerator, and optimization tracking powered by the dspy.rb gem.
Google ADK for Ruby Landon Gray released an unofficial Ruby port of Google's Agent Development Kit. The gem supports LLM agents, sequential/parallel/loop workflows, tool integration, and session management using Gemini models.
HTM MCP Server Dewayne VanHoozer added an MCP server to the Hierarchical Temporal Memory gem enabling AI assistants to store, recall, and manage memories. The latest release includes Robot Groups for high-availability coordination with shared working memory, automatic failover, and synchronization across multiple robots.
dspy-rb-skill Vicente Reig released a Claude Skill that equips Claude Code with DSPy.rb knowledge for type-safe LLM development in Ruby, featuring Sorbet type annotations, ReAct/CodeAct agents, and MIPROv2 optimization.
Security in the Age of AI Workshop Greg Molnar announced a live online workshop (paid, $50) on December 19th at 5:30pm GMT covering LLM security risks and how AI-powered features can be exploited when integrated into applications.
Sentry Agents Istvan Hoka extracted a Ruby gem that brings Sentry's Generative AI monitoring to Ruby applications. The gem provides observability for agent execution flows, token tracking, and tool execution with auto-instrumentation for RubyLLM and LangChain.rb.
AI Guardrails Faisal Raza built a Ruby gem that validates, repairs, and secures LLM outputs. The gem automatically fixes broken JSON, enforces schema validation, blocks unsafe content, and includes retry logic with auto-correction for production use.
Rails AI Agents Thibaut Baissac released a collection of 20+ specialized GitHub Copilot agents for Rails 8.1 development that follow TDD workflows. The agents cover feature planning, red-green-refactor testing cycles, and implementation specialists for models, services, controllers, policies, ViewComponents, and Solid Queue jobs.
LLM Classifier Dmitry Sychev created a Ruby gem for building LLM-powered classifiers with a clean DSL. The gem supports multi-label classification, domain knowledge injection, Rails integration with ActiveRecord concerns.
Claude Ruby Plugins Stephen Ierodiaconou released a Claude Code plugin marketplace providing skills, agents, and commands for writing Ruby RBS type signatures. The plugin includes agents for creating, maintaining, reviewing, and converting RBS annotations, plus a runtime type tracer that helps discover actual method return types during execution.
Unofficial 37signals Coding Style Guide Marc Köhlbrugge extracted transferable Rails patterns from analyzing pull requests in 37signals's Fizzy codebase, documenting conventions for CRUD controllers, Stimulus/Hotwire, multi-tenancy, and AI/LLM integration with emphasis on rich domain models and vanilla Rails.
Fizzy Skill Rob Zolkos made a Claude Code skill that integrates with Fizzy for managing boards, cards, and tasks through natural language commands in the terminal.
fizzy-cli Rob also released a Ruby CLI for 37signals' Fizzy project management tool, designed to be LLM-friendly for AI assistant integration. The tool was built using the newly released Fizzy API.
Custom GitHub Copilot Agent for Fixing Rails Errors with Honeybadger Joshua Wood shared an agent that integrates Honeybadger's MCP server to debug Rails production errors. The agent fetches real-time error data, stack traces, and affected user patterns, then follows a structured workflow to diagnose root causes.
DeepRails launched a guardrails platform that detects and fixes LLM hallucinations in real-time. The service offers SDKs for Ruby, Python, Go, and TypeScript with metrics for correctness, completeness, adherence, and safety evaluation.
Ruby Users Forum Is Launching Soon Javier Cervantes announced a new discussion platform for Ruby featuring topic-based organization, advanced search, and discussions designed to preserve knowledge that gets lost in chat-based communities.
Articles
AI for Ruby Devs Part I: From the Basics to Building a Neural Network David Sanchez walked through machine learning fundamentals and built a working neural network from scratch in pure Ruby. The tutorial implemented forward propagation, backpropagation, and gradient descent using a student test score prediction example.
AnyCable, Rails, and the Pitfalls of LLM Streaming Vladimir Dementyev of Evil Martians exposed Action Cable's threading issues causing out-of-order messages and connection loss problems when streaming LLM responses, then demonstrated how AnyCable provides message ordering and at-least-once delivery guarantees with a simple gem swap. He also revealed AnyCable has begun adopting Durable Streams!
Announcing Raix 2.0 Obie Fernandez released a major update to the Ruby AI eXtensions library featuring RubyLLM as a unified backend for multiple providers and a new before_completion hook system. The hooks enable dynamic model selection, PII redaction, request logging, system prompt injection and A/B testing.
Design Previews for Ruby on Rails Avi Flombaum introduced ActionVersionPreview, a Rails gem that preserves multiple LLM-generated UI iterations for side-by-side comparison. The gem uses Rails view variants with a +variant suffix pattern and requires zero configuration. Learn more at the ActionVersionPreview documentation site.
Rails MCP Server: Context-Efficient Tool Architecture Mario Alberto Chávez redesigned the Rails MCP Server to reduce context consumption by cutting registered tools from 12 to 4 with progressive discovery via search_tools. The update replaced regex parsing with Rails reflection APIs and Prism AST analysis, added a sandboxed execute_ruby tool, and introduced an interactive configuration tool with Quick Start guides.
Introducing the Rails Superhero Card Generator Fiona Lapham of OmbuLabs walked through building a Rails-themed holiday card generator (in Python) using LlamaIndex orchestration. Ernesto Tagwerker announced the open-source release as one of several experiments exploring practical AI applications, while Gelsey Torres wrote about Hacktoberfest Open Source Contributions featuring updates to Active Agent, DSPy.rb, and langchain.rb.
Same Revolution, Different Tools Robert Beene synthesized insights from three conference talks into a framework for developers directing AI systems: control the objective through clear intent, control the context via mandatory guidelines, and control access with proper governance.
Why We Built Claude OS and What It Actually Is Bob Roberts released Claude OS, a persistent memory system for Claude Code that reduces the time spent rebuilding context across sessions. The local-first tool combines SQLite storage, vector embeddings, and tree-sitter AST parsing to index files, and works well with Ruby on Rails projects.
Building LLM-Powered Applications in Ruby: A Practical Introduction Germán Giménez Silva summarized Koichi Ito’s “Ruby × LLM Ecosystem” Ruby World Conference presentation covering RAG implementation, LangChain.rb, and the Model Context Protocol's new Ruby SDK.
Exploring the World of Electronic Engineering with PicoRuby and A Definitive Guide to Running PicoRuby on ESP32 Germán also published companion guides covering PicoRuby's hardware capabilities on RP2040 and ESP32 microcontrollers, including GPIO, PWM, and sensor control with practical examples from Hayao Kimura's Ruby World Conference presentation.
Git Worktrees with Parallel Agents in Practice Rich Steinmetz offered an assessment that parallel agent workflows remain limited by both engineer skill and agent behavior (tasks complete too quickly, agents exit early when in doubt), but shared his Ruby script with Ultrathink A/B mode for evaluating multiple Claude Code solutions simultaneously.
A Trap with GlobalIDs Julik Tarkhanov warned that Rails GlobalIDs are unsafe for LLM tool calls because models can fabricate GIDs that bypass authorization and access other users' records. Julik recommended using signed IDs with purpose parameters or a custom locator that validates numeric primary keys.
What Happens When the Coding Becomes the Least Interesting Part of the Work Obie Fernandez argued that "senior thinking" (blast radius awareness, sequencing instincts, reversibility preferences, social awareness) is what coding agents can't replicate, but pair programming with Claude Code forces engineers to articulate this tacit knowledge aloud, making them better.
Vibe Coding in style.md Irina Nazarova and Vladimir Dementyev of Evil Martians documented how they created an AGENTS.md file by comparing vibe-coded Rails apps against Vladimir's refactored versions, capturing patterns like domain naming, enums for state, and namespaced classes to make AI-generated code maintainable.
Cooking with Gas, Saving an Afternoon in the Library, and On Thoughts About Language The Weekly Vivarium by Brian Shirai detailed progress reviving the Rubinius language platform with a new rbx compiler that unifies Ruby and Python via LLVM that preserves dynamic features through deoptimization, with plans for unified testing specifications across Ruby implementations.
Compound Engineering: How Every Codes with Agents Dan Shipper and Kieran Klaassen shared details on compound engineering, a methodology where each feature makes the next easier through a learning loop that documents bugs and solutions for future agents. Their four-step process (Plan, Work, Assess, Compound) uses Claude Code with 12 parallel review subagents, enabling single developers to run entire products.
Is Traditional Machine Learning Dead? Comparing LLMs and Gradient Boosting Wouter Massa of Gusto compared LLMs against XGBoost for mapping Gusto categories to bookkeeping accounts, finding gradient boosting achieved 20% higher user acceptance and reduced inference time. The article concluded that LLMs enable rapid prototyping, but specialized machine learning algorithms excel at specific ranking tasks.
Git Worktree: Scaling Your AI Workflow Harisankar P S explained how git worktree enables parallel AI development by creating sibling directories where multiple Claude Code or Cursor instances can work simultaneously on different branches.
The AI Wildfire Is Coming Justin Searls shared Dion Lim's analysis on the coming AI market correction, noting that he and Scott Werner used the same wildfire analogy in their Hot Fix episode. Justin argued that rather than retreating during downturns, companies should invest aggressively when competitors are cashing out.
Rails 6 to Rails 8 in 130 Minutes Roger Heykoop documented upgrading an open-source API across two major Rails versions in just over two hours, combining AI assistance with 19 years of Rails expertise to handle the compatibility debugging while the agent performed repetitive fixes.
How the Junior to Senior Dev Problem Will Be Solved Daniel Tenner argued that the traditional conveyor belt from junior to senior developer is economically dead since tokens are cheaper than training humans, but predicted passionate developers will bootstrap themselves using AI tutors while the rest exit the field.
Rich Prompting, Agentic Viewports, and More José Valim announced Tidewave updates including agentic viewports that allow AI coding agents to automatically resize browser dimensions when debugging responsive design issues. Depth-Based Inspector, Sunsetting /tidewave, and More introduced a depth-based inspector that displays all DOM elements at a clicked position via contextual menu, eliminating full DOM traversal. The /tidewave route is also being deprecated in favor of a dedicated app to support long-running processes required by coding agents.
RubyMine 2025.3: Multi-Agent AI Chat, Rails-Aware MCP Server, Faster Multi-Module Projects Startup, and More Alexey Varfolomeev of JetBrains announced RubyMine 2025.3 featuring a unified multi-agent AI chat combining Claude Agent with Junie and a Rails-aware MCP server for AI-assisted code exploration.
Videos
Technology for Humans: The Survey Results Errol Schmidt from reinteractive presented findings from a survey of Ruby and Rails leaders and executives on current trends, AI, and the future of Rails.
DSPy Interview Series: Vicente on DSPy for Ruby, BAML, and the New Era of AI Engineering Information Shelf interviewed Vicente Reig, author of dspy.rb, about porting Stanford's DSPy framework to Ruby, integrating with RubyLLM, the importance of observability in AI projects, and the evolving landscape of AI agents and workflows.
Building a Podcast Directory in Rails with an AI Assistant Chad Pytel and Moses Amama from Thoughtbot kicked off development of a creator-friendly podcast directory in Rails, demonstrating how AI can accelerate both design and development workflows during their livestream.
AI (on Rails) 101 Episode II: Uncluttered AI-Driven Experiences in Rails João Malheiros of Codeminer42 explored and compared Ruby gems that integrate LLM providers into Rails projects as part of their ongoing AI on Rails series.
Building Insignia Live: Learning and Building (Rails + Hotwire) Ft ChatGPT x Codex Abdulrasaq Mustapha livestreamed building a digital ID card SaaS platform with Rails 8, Hotwire, and Stimulus, working through user roles, organization onboarding, and card creation workflows in public.
Podcasts
The Ruby AI Podcast: Running Self-Hosted Models with Ruby Joe Leo and Valentino Stoll interviewed Chris Hasiński about self-hosting AI models for control over updates and fine-tuning, LLM limitations when generating Ruby code, and the need for investment in Ruby's ML tooling ecosystem.
Indie Rails: Ernesto Tagwerker: OmbuLabs on AI Jeremy Smith and Jess Brown talked with Ernesto Tagwerker about how OmbuLabs uses LLMs to accelerate Rails application upgrades, their Bonsai fixed-cost maintenance service, and new AI-powered offerings for code assessments and development.
Shawn Ryan Show: Tobi Lütke: How Shopify Became a Cheat Code for Entrepreneurs Shawn Ryan interviewed Shopify CEO and Rails core contributor Tobi Lütke about building Shopify on Ruby on Rails from a snowboard shop, creating Active Merchant, and AI's growing role in business and the future of work.
Discussions
LLM Request Logging for Ruby Avi Flombaum asked the community for recommendations on logging LLM requests in Ruby, seeking tools to track tokens, response time, and costs with RubyLLM integration, ideally as a Rails engine.
Ruby Is the Perfect Language for This Style of Vibe Coding A thread on Reddit introduced "Inside-Out Vibe Coding," an experimental approach where method_missing intercepts undefined calls and triggers small LLMs to generate missing code iteratively, leveraging Ruby's metaprogramming and Prism parser.
What Does Your AI Dev Setup Look Like? A Reddit user considering Tidewave asked Rails developers about their AI coding setups, with responses ranging from Claude Code with claude-on-rails and superpowers in devcontainers to RubyMine with Junie, Zed with Copilot, Gemini CLI, and Cursor with Opus plan mode.
Events
There are currently no Ruby AI events scheduled within the next month.
Previously: Victoria Melnikova and Irina Nazarova recapped Why Evil Martians Hosted a Ruby Conference in San Francisco, explaining how monthly San Francisco Ruby meetups starting in March 2024 grew into a three-day conference at Fort Mason in November of this year. The event brought 400+ Ruby engineers together with dozens of startup demos, with technical talks covering YJIT performance, MCP tooling, and Rails scaling. The organizers positioned SF Ruby as a statement that Ruby remains the framework powering the next generation of startups, with talk recordings coming soon.
Open Source Updates
Code Spotlight
Paulo Arruda's Swarm Ruby framework for orchestrating multi-agent AI systems has constantly improved and innovated. Built on RubyLLM , and recently decoupled from Claude Code, it enables agents to collaborate with specialized roles, tools, and delegation capabilities across agentic coding providers. The latest improvements introduced include context preservation controls for delegation, citation extraction with streaming support, lazy tool activation, asynchronous execution timeouts, and intelligent LLM retry logic differentiating client from server errors.
New Gems
Links to the RubyGems page, newest releases are first:
exaonruby - Client for the Exa.ai API with CLI, middleware, and Rails integration
jules-ruby - Ruby CLI for the Jules API
hotwire_club-mcp - MCP server for Hotwire Club knowledge base - provides tools for searching, browsing, and discovering documentation
action_version_preview - Preview multiple view variants side-by-side using Rails built-in view variants.
rails_prompts - Manage AI prompts in Rails applications with ERB templates
sentry-agents - Sentry Gen AI instrumentation for AI/LLM agents in Ruby
google-adk - Unofficial Ruby implementation of Google's Agent Development Kit
ferrum-mcp - Browser automation server implementing the Model Context Protocol
personality - MCP server for personality communication rules
labnocturne - Ruby client for Lab Nocturne Images API
ruby_llm-semantic_cache - Semantic caching for RubyLLM applications
ai_screen_analyzer - Adds a floating button on any page to capture the screen, send for analysis, and display the result on a sidebar
ai_guardrails - Schema validation and safety layer for AI-generated output in Ruby
tensor - Fast tensor/matrix operations for Ruby using OpenMP and OpenBLAS
shai-cli - CLI tool for managing shared AI agent configurations
ruby_llm-red_candle - Red Candle provider for RubyLLM - local LLM execution using quantized GGUF models
helicone-rb - Ruby client for Helicone AI Gateway with agentic tool support
dspy-ruby_llm - RubyLLM adapter for DSPy.rb - unified access to 12+ LLM providers
naiso - Auto-split long detailed images based on solid/gradient background areas
portable_mruby - Build portable Ruby executables using mruby and Cosmopolitan Libc
cov-loupe - MCP server + CLI for SimpleCov coverage data
fal-ai - Ruby client for fal.ai generative AI platform
torch-ddp - Distributed data parallel support for torch-rb
magika - Determines file content types using AI
New Open Source
Links to the Github repository:
Hotwire Club MCP - MCP server providing searchable access to Hotwire Club documentation with category filtering and related content discovery
Decision Agent - Deterministic decision-making engine with explainable outputs, audit replay, conflict resolution, and optional AI enhancement for regulated domains
MCP on Rails - Rails server implementing the Model Context Protocol to connect AI assistants with Azure DevOps and PostgreSQL databases
Trait AI - Rails chatbot with persistent user memory, adaptive personality traits, and content moderation that evolves through conversation patterns
AI Shopping Assistant - Conversational e-commerce platform using OpenAI with multi-agent architecture for natural language product search and basket management
Redmine MCP - Redmine plugin that exposes an MCP server with tools for AI assistants to manage issues, projects, time tracking, and wikis
AI Screen Analyzer - Rails engine that captures page screenshots and sends them to an LLM for visual analysis with results displayed in a sidebar
AI Voice Bot - Embeddable voice-enabled chat widget using Web Speech API with OpenAI integration and an admin dashboard for managing FAQs and products
One Last Thing
Tnkr is a collaboration platform for robotics builders that unifies hardware documentation, software integration, operational data, and model deployment. The platform introduces Leonardo, an AI assistant that analyzes POV build videos, CAD files, and code to automatically generate step-by-step assembly documentation. Key features include integration with development tools like Onshape, SolidWorks, and GitHub, along with a data contribution loop where community members can deploy VLA models to hardware and contribute real-world operational data back to improve training. The platform targets open-source robotics projects ranging from humanoids to quadrupeds with 3D visualization and documentation capabilities.
That’s all for this edition! Be sure to reach out if you have any stories, content, jobs, or events you want featured in the newsletter.