Agent framework for Rails: bounded tool-calling loops, fast paths, structured extraction, and (v0.2) durable workflows on ActiveRecord. Built on ruby_llm.

Phronomy is a Ruby AI agent framework that provides composable building blocks — Agents, Workflows, Tools, Filters, and Tracing — for building AI agents in Ruby. Powered by RubyLLM for LLM abstraction.

pikuri-core is the lean, audit-friendly foundation of the pikuri family: Pikuri::Agent (a thin wrapper around ruby_llm's chat loop) with its Configurator + Extension protocol, the strict Pikuri::Tool framework, a listener surface for rendering / budgets / sub-agents, and four bundled stateless tools (calculator, web search, web scrape, fetch). Extensions (skills, MCP, workspace, coding stack, named-agent personas) live in sibling gems so a privacy-conscious user can install just this core and audit a minimal dependency tree. For the convenience bundle that pulls in everything, see the +pikuri+ metagem.

Probatio Diabolica runs custom *_spec.rb files with a DSL inspired by RSpec and supports text/image/PDF reporting.

Under development. Contributors welcome.

Ragents is a pure Ruby, Ractor-based AI agent framework designed for concurrent, isolated agent execution. It provides message-passing architecture, tool-based actions, and provider-agnostic LLM integration without Rails dependencies.

Ragnar is a high-performance RAG system that leverages Rust libraries through Ruby bindings for embeddings, vector search, and topic modeling. It provides a complete CLI for indexing documents and querying with LLMs.

Unified tool definition pipeline for Rails 8 applications using FastMCP and RubyLLM.

RainbowLLM provides intelligent routing and failover capabilities for multiple LLM providers. Automatically tries providers in sequence until one responds successfully, ensuring reliable access to LLM services.

RLM.rb is a Ruby runtime spine for Recursive Language Models. It runs bounded, typed, auditable AI jobs over files, records, and application context. RLM.rb includes RubyLLM provider access, a dspy.rb signature adapter, the recursive prompt loop, file/context mounting, recursive sub-LM calls, typed final output, budget controls, trace events, and a best-effort trace_store hook.

RobotLab is a Ruby framework for building and orchestrating multi-robot LLM workflows. Built on ruby_llm, it provides robots with template-based prompts, tools, and shared memory; networks for coordinating multiple robots with intelligent routing; MCP (Model Context Protocol) integration for external tool servers; and a memory system with Redis backend and semantic caching. Optional gems add Rails integration (robot_lab-rails), durable learning (robot_lab-durable), Ractor concurrency (robot_lab-ractor), document storage (robot_lab-document_store), and OS-level skill-script confinement (robot_lab-sandbox). Core itself imposes no execution limitations.

Standalone background daemon that watches a repo's git activity, Claude Code session transcripts, and terminal commands, then infers the current goal/direction via a one-shot RubyLLM call to a local model. Writes an intent artifact any tool can read — robot_lab-to uses it to seed takeover runs, but nothing here requires RobotLab.

A standalone, self-contained coding and automation agent built on ruby_llm. Provides an agent loop, persistent memory, SQLite sessions, context compaction, a job system, a tool registry, and an extensible UI layer.

Interactive rake tasks to generate context files for AI coding assistants (Claude, Cursor, Windsurf, Codex, llm.txt). Uses RubyLLM to analyze your codebase and create platform-specific context that helps AI understand your project.

RubyCoded is a terminal-based AI coding assistant built in Ruby. It provides a full TUI chat interface with support for multiple LLM providers (OpenAI, Anthropic, etc.), an agent mode with filesystem tools for reading, writing, and editing project files, a plan mode for structured task planning, and a plugin system for extensibility.

An unofficial RubyLLM ecosystem extension that turns one RubyLLM 2.x chat turn into one AG-UI 1.0 run. It parses and validates RunAgentInput, maps RubyLLM streaming, tool and lifecycle behavior onto schema-valid AG-UI events, turns requires_approval pauses into interrupt outcomes with host-authorized resume, reports cancellation as the cancelled outcome, and ships an SSE writer for plain Ruby plus an optional Rails adapter.

Emits the AI SDK UI message stream protocol from RubyLLM chats, so React front ends built with useChat talk to a Ruby backend: text, reasoning, tool calls and approvals.

Talks the native Anthropic Messages format to Bedrock InvokeModel / InvokeModelWithResponseStream, unlocking Anthropic-specific features that the Converse API cannot express — tool search with deferred tool loading, prompt caching via cache_control, and structured outputs — while keeping data inside the AWS boundary. Authenticates with the standard AWS credential chain (IRSA, Pod Identity, instance profiles, env vars) via aws-sdk-core.

An unofficial community extension for RubyLLM. It adds policy-bound, read-only browser Inspect and Extract tools with an offline FakeBackend. The Stagehand Ruby SDK adapter is present but unsupported for live operations until provider egress and cleanup are qualified. The first release also includes explicit URL/SSRF policy, bounded and sanitized results with provenance, durable operation and session lifecycle bookkeeping, budgets, offline Act simulation, and an optional Rails 8 / Active Job adapter. The full live Act protocol is deferred to a later version. The Gem core works in plain Ruby without Rails.

Use Claude Code (and its subscription login) as a RubyLLM provider: streaming, images, PDFs, Office files, structured output and emulated tool calls over `claude -p`.