Jido Ecosystem

A learning repository for exploring the Jido Ecosystem - a collection of Elixir libraries for building autonomous multi-agent workflows.

The name "Jido" (自動) comes from Japanese meaning "automatic" or "self-moving".

Purpose

This repository serves as a local workspace for:

• Understanding the Jido Ecosystem codebase
• Providing context to AI coding agents (Claude, Cursor, Copilot, etc.)
• Experimenting with the libraries in a unified environment

The actual Jido packages are cloned into packages/ and excluded from version control, keeping this repo lightweight while providing full access to the source code.

Quick Start

# Clone this repository
git clone [email protected]:dbhowmick/jido_ecosystem.git
cd jido_ecosystem

# Clone all Jido packages
./pull_all.sh

# Navigate to any package
cd packages/jido
mix deps.get
mix test

Repository Structure

jido_ecosystem/
├── CLAUDE.md         # Detailed guidance for AI coding agents
├── README.md         # This file
├── pull_all.sh       # Script to clone/pull all packages
└── packages/         # Cloned Jido repositories (git-ignored)
    ├── jido/             # Core agent framework
    ├── jido_action/      # Composable, validated actions
    ├── jido_signal/      # CloudEvents-based messaging
    ├── jido_ai/          # AI/LLM integration
    ├── req_llm/          # HTTP client for 57+ LLM providers
    └── jido_workbench/   # Phoenix demo application

Package Overview

Package	Description	Key Concepts
jido	Core agent framework	Immutable agents, cmd/2 pattern, directives
jido_action	Action system	Validated actions, workflows, DAG execution
jido_signal	Messaging layer	CloudEvents, pub/sub, signal routing
jido_ai	AI integration	LLM models, prompts, reasoning strategies
req_llm	LLM HTTP client	57+ providers, streaming, tool calling
jido_workbench	Demo app	Phoenix LiveView showcase

Dependency Graph

req_llm (standalone)
jido_action (standalone)
jido_signal (standalone)
       │
       ▼
jido (core) ─────► depends on jido_action, jido_signal
       │
       ▼
jido_ai ─────────► depends on jido, req_llm
       │
       ▼
jido_workbench ──► depends on jido, jido_ai, req_llm

For AI Coding Agents

When working with this codebase, refer to CLAUDE.md for:

• Core architecture patterns (the cmd/2 pattern)
• Actions vs Directives vs State Operations
• Code style conventions
• Common commands for each package
• Key modules and their purposes

Key Architectural Concepts

The cmd/2 Pattern (Elm/Redux-inspired):

{agent, directives} = MyAgent.cmd(agent, action)

• Agents are immutable data structures
• cmd/2 is pure: same inputs produce same outputs
• Side effects are described as directives (executed by runtime)

Three-Layer Separation:

1. Actions - Transform state, may perform side effects
2. Directives - Describe external effects (Emit, Spawn, Schedule, etc.)
3. State Operations - Describe internal state changes (SetState, DeleteKeys, etc.)

Common Commands

All packages use Mix with consistent patterns:

mix deps.get          # Fetch dependencies
mix test              # Run tests
mix quality           # Format + compile + dialyzer + credo (alias: mix q)
mix format            # Format code
mix dialyzer          # Type checking
mix credo             # Linting

Prerequisites

• Elixir ~> 1.17
• Erlang/OTP 26+

Using Local Dependencies

To develop across packages with local changes:

LOCAL_JIDO_DEPS=true mix deps.get

This switches jido_ai and jido_workbench to use local path dependencies instead of Hex packages.

Links

• agentjido GitHub Organization
• jido on Hex.pm
• jido_ai on Hex.pm
• req_llm on Hex.pm