Jido AI

Intelligent Agent Framework for Elixir

Jido AI is a comprehensive framework for building sophisticated AI agents and workflows in Elixir. It extends the Jido framework with powerful LLM capabilities, advanced reasoning techniques, and stateful conversation management.

Features

• Multi-Provider Support: Access 57+ LLM providers through ReqLLM (OpenAI, Anthropic, Google, Mistral, and more)
• Advanced Reasoning: Chain-of-Thought, ReAct, Tree-of-Thoughts, Self-Consistency, GEPA
• Structured Prompts: Template-based prompts with EEx and Liquid support
• Tool Integration: Function calling with automatic schema conversion
• Conversation Management: Stateful multi-turn conversations with ETS storage
• Context Window Management: Automatic token counting and truncation strategies

Installation

Add jido_ai to your mix.exs dependencies:

def deps do
  [
    {:jido_ai, "~> 0.5.3"}
  ]
end

Quick Start

alias Jido.AI.{Model, Prompt}
alias Jido.AI.Actions.ReqLlm.ChatCompletion

# Create a model
{:ok, model} = Model.from({:anthropic, [model: "claude-3-5-sonnet"]})

# Create a prompt
prompt = Prompt.new(:user, "What is the capital of France?")

# Get a response
{:ok, result} = ChatCompletion.run(%{model: model, prompt: prompt}, %{})
IO.puts(result.content)

Documentation

User Guides

Learn how to use Jido AI in your applications:

Guide	Description
Getting Started	Installation, configuration, and first steps
Models	Working with LLM models and providers
Prompts	Creating and templating prompts
Configuration	API keys and settings management
Chat Completion	Basic and advanced chat completions
Conversations	Multi-turn stateful conversations

Runners (Advanced Reasoning)

Implement sophisticated reasoning strategies:

Runner	Description	Accuracy Gain
Overview	Introduction to runners	-
Chain of Thought	Step-by-step reasoning	+8-15%
ReAct	Reasoning with tool use	+27%
Self-Consistency	Multiple paths with voting	+17.9%
Tree of Thoughts	Tree search exploration	+74%
GEPA	Evolutionary prompt optimization	+10-19%

Developer Guides

Understand the internals for extending Jido AI:

Guide	Description
Architecture	System architecture and components
Model System	Model creation, registry, and discovery
Prompt System	Prompt structs, templating, versioning
Actions System	Jido Actions and tool integration
Runners System	Runner implementations and patterns
Data Flow	Request lifecycle and data transformations

Supported Providers

Jido AI supports 57+ providers through ReqLLM:

Provider	Example Models
Anthropic	Claude 3.5 Sonnet, Claude 3 Opus
OpenAI	GPT-4o, GPT-4 Turbo
Google	Gemini 1.5 Pro, Gemini 1.5 Flash
Mistral	Mistral Large, Mixtral 8x7B
Groq	Llama 3.1 70B
Cohere	Command R+
Local	Ollama, LM Studio

See the Models Guide for complete provider documentation.

Configuration

Set API keys via environment variables:

export ANTHROPIC_API_KEY="sk-..."
export OPENAI_API_KEY="sk-..."
export GOOGLE_API_KEY="..."

Or configure in your application:

config :jido_ai, :keyring, %{
  anthropic_api_key: "sk-...",
  openai_api_key: "sk-..."
}

See the Configuration Guide for details.

API Documentation

Full API documentation is available at HexDocs.

Generate documentation locally:

mix docs
open doc/index.html

Resources

• GitHub Repository
• HexDocs
• Jido Framework
• ReqLLM

License

MIT License - see LICENSE for details.