mruby-llm

About

mruby-llm brings the full llm.rb AI runtime to mruby, making it possible to embed AI capabilities into small standalone applications. It supports the same execution model as llm.rb — providers, agents, tools, MCP, A2A, streaming, and persisted state — but adapted for mruby's constraints.

The project began as a port of llm.rb to mruby, and a large number of features proved portable. Code continues to flow between both projects, with improvements in each benefiting the other.

mruby-llm is the runtime that powers robert, a native 2MB FreeBSD AI binary.

Supported Providers

• OpenAI and OpenAI-compatible endpoints
• Anthropic
• Google Gemini
• DeepSeek
• xAI
• Z.ai
• Ollama
• llama.cpp

Key Features

• LLM::Context — Stateful execution environment with message history, tool state, and streaming
• LLM::Agent — Automatic tool-loop execution with loop guards
• Tools — Subclass LLM::Tool to extend model capabilities
• MCP — Model Context Protocol client over stdio and HTTP
• A2A — Agent-to-Agent protocol client for remote skills
• Skills — Reusable instructions from SKILL.md directories
• Streaming — Content, reasoning, and tool event callbacks
• Serialization — Context persistence to JSON
• Context Compaction — Automatic history summarization

Build Integration

Add to your mruby build config:

MRuby::Build.new("app") do |conf|
  conf.gem github: "llmrb/mruby-llm", branch: "main"
end

Requires libcurl for HTTP support.

Resources

• GitHub Repository
• API Documentation