Skip to content

agentjido/jido

BranchesTags

Repository files navigation

Jido

Hex.pm Hex Docs CI License Website Ecosystem Discord

Jido is an autonomous agent framework for Elixir, built for workflows and multi-agent systems.

Define agents, connect them to actions, signals, and directives, and run them with supervision and fault tolerance built in.

The name "Jido" (自動) comes from the Japanese word meaning "automatic" or "automated", where 自 (ji) means "self" and 動 (dō) means "movement".

Learn more about Jido at jido.run.

Overview

Jido helps you build agent systems as ordinary Elixir and OTP software.

  • Agents hold state and implement cmd/2
  • Actions do work and transform that state
  • Signals route events into the system
  • Directives describe effects for the runtime to execute

The purity boundary is the agent's decision logic: Jido keeps agent decisions and state transitions explicit, actions may be pure or effectful, and directives are for effects you want the runtime to own.

Use Jido when software needs to inspect context, choose among multiple steps, coordinate with other agents, and keep running reliably over time.

AI is optional. The core package gives you the agent architecture and runtime; companion packages such as jido_ai add model integration when you need it.

At the core, Jido agents are immutable data structures with a single command function:

defmodule MyAgent do
  use Jido.Agent,
    name: "my_agent",
    description: "My custom agent",
    schema: [
      count: [type: :integer, default: 0]
    ]
end

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

State changes are explicit data transformations. If an action needs a result back immediately to continue reasoning or update state, it may perform that work itself. If the workflow has already decided on an outbound effect and wants the runtime or integration layer to own delivery, return a directive.

The Jido Ecosystem

Jido is the core package of the Jido ecosystem. The ecosystem is built around the core Jido Agent behavior and offer several opt-in packages to extend the core behavior.

Package Description
req_llm HTTP client for LLM APIs
jido_action Composable, validated actions with AI tool integration
jido_signal CloudEvents-based message envelope and supporting utilities for routing and pub/sub messaging
jido Core agent framework with state management, directives, and runtime
jido_ai AI/LLM integration for agents

For demos and examples of what you can build with the Jido Ecosystem, see https://jido.run. For the package map and support levels, see https://jido.run/ecosystem.

Why Jido?

OTP primitives are excellent. You can build agent systems with raw GenServer. But when building multiple cooperating agents, you'll reinvent:

Raw OTP Jido Formalizes
Ad-hoc message shapes per GenServer Signals as standard envelope
Business logic mixed in callbacks Actions as reusable command pattern
Implicit effects scattered in code Directives as typed effect descriptions
Custom child tracking per server Built-in parent/child hierarchy
Process exit = completion State-based completion semantics

Jido isn't "better GenServer" - it's a formalized agent pattern built on GenServer.

Key Features

Immutable Agent Architecture

  • Functional agent state model inspired by Elm/Redux
  • cmd/2 as the core operation: actions in, updated agent + directives out
  • Schema-validated state with NimbleOptions or Zoi

Directive-Based Effects

  • Actions transform state and may perform required work
  • Directives describe runtime-owned external effects
  • Built-in directives: Emit, Spawn, SpawnAgent, StopChild, StartSensor, StopSensor, Schedule, Stop
  • Protocol-based extensibility for custom directives

OTP Runtime Integration

  • GenServer-based AgentServer for production deployment
  • Parent-child agent hierarchies with lifecycle management
  • Signal routing with configurable strategies
  • Instance-scoped supervision plus logical partitions for multi-tenant deployments

Composable Plugins

  • Reusable capability modules that extend agents
  • State isolation per plugin with automatic schema merging
  • Lifecycle hooks for initialization and signal handling

Execution Strategies

  • Direct execution for simple workflows
  • FSM (Finite State Machine) strategy for state-driven workflows
  • Extensible strategy protocol for custom execution patterns

Multi-Agent Orchestration

  • Multi-agent workflows with configurable strategies
  • Plan-based orchestration for complex workflows
  • Durable groups of agents with named topology, hierarchical runtime ownership, nested pod nodes, and partition-safe tenancy boundaries

Installation

Using Igniter (Recommended)

The fastest way to get started is with Igniter:

mix igniter.install jido

This automatically:

  • Adds Jido to your dependencies
  • Creates a MyApp.Jido instance module (use Jido, otp_app: :my_app)
  • Creates configuration in config/config.exs
  • Adds MyApp.Jido to your supervision tree

Generate an example agent to get started:

mix igniter.install jido --example

Manual Installation

Add jido to your list of dependencies in mix.exs:

def deps do
  [
    {:jido, "~> 2.0"}
  ]
end

Then define a Jido instance module and add it to your supervision tree:

# In lib/my_app/jido.ex
defmodule MyApp.Jido do
  use Jido, otp_app: :my_app
end
# In config/config.exs
config :my_app, MyApp.Jido,
  max_tasks: 1000,
  agent_pools: []
# In your application.ex
children = [
  MyApp.Jido
]

Supervisor.start_link(children, strategy: :one_for_one)

Quick Start

1. Define an Agent

defmodule MyApp.CounterAgent do
  use Jido.Agent,
    name: "counter",
    description: "A simple counter agent",
    schema: [
      count: [type: :integer, default: 0]
    ],
    signal_routes: [
      {"increment", MyApp.Actions.Increment}
    ]
end

2. Define an Action

defmodule MyApp.Actions.Increment do
  use Jido.Action,
    name: "increment",
    description: "Increments the counter by a given amount",
    schema: [
      amount: [type: :integer, default: 1]
    ]

  def run(params, context) do
    current = context.state[:count] || 0
    {:ok, %{count: current + params.amount}}
  end
end

3. Execute Commands

# Create an agent
agent = MyApp.CounterAgent.new()

# Execute an action - returns updated agent + directives
{agent, directives} = MyApp.CounterAgent.cmd(agent, {MyApp.Actions.Increment, %{amount: 5}})

# Check the state
agent.state.count
# => 5

4. Run with AgentServer

# Start the agent server
{:ok, pid} = MyApp.Jido.start_agent(MyApp.CounterAgent, id: "counter-1")

# Send signals to the running agent (synchronous)
# Signal types must be declared in signal_routes
{:ok, agent} = Jido.AgentServer.call(pid, Jido.Signal.new!("increment", %{amount: 10}, source: "/user"))

# Look up the agent by ID
pid = MyApp.Jido.whereis("counter-1")

# List all running agents
agents = MyApp.Jido.list_agents()

Core Concepts

The cmd/2 Contract

The fundamental operation in Jido:

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

Key invariants:

  • The returned agent is always complete - no "apply directives" step needed
  • directives describe runtime-owned external effects only - they never modify agent state
  • Agent decision logic stays explicit and testable; deterministic actions produce deterministic cmd/2 results

Use this rule of thumb for side effects:

  • If the step needs a result back now to continue reasoning or update state, an effectful action is acceptable.
  • If the workflow has already decided on an outbound effect and wants the runtime or integration layer to own delivery, return a directive.

Actions vs Directives vs State Operations

Actions Directives State Operations
Transform state, may perform side effects Describe runtime-owned effects Describe internal state changes
Executed by cmd/2, update agent.state Bare structs emitted by agents Applied by strategy layer
Can call APIs, read files, query databases Runtime (AgentServer) interprets them Never leave the strategy

State Operations (Jido.Agent.StateOp)

State operations are internal state transitions handled by the strategy layer during cmd/2. Unlike directives, they never reach the runtime.

StateOp Purpose
SetState Deep merge attributes into state
ReplaceState Replace state wholesale
DeleteKeys Remove top-level keys
SetPath Set value at nested path
DeletePath Delete value at nested path

Directive Types

Directive Purpose
Emit Dispatch a signal via configured adapters
Error Signal an error from cmd/2
Spawn Spawn a generic BEAM child process
SpawnAgent Spawn a tracked child Jido agent (restart: :transient by default)
StopChild Gracefully stop and remove a tracked child agent
StartSensor Start or replace a tagged sensor runtime
StopSensor Stop a tagged sensor runtime
Schedule Schedule a delayed message
Stop Stop the agent process

Sensor runtimes started by StartSensor are tracked under {:sensor, tag} and are owner-monitored by default. Unexpected sensor exits emit jido.agent.sensor.exit back to the owning agent; controlled StopSensor shutdowns do not. Use link?: true only when a sensor should fail fast with its owning AgentServer.

Documentation

Start here:

Guides:

Advanced:

API Reference: hexdocs.pm/jido

FAQ

General Questions

Q: What is Jido and how does it differ from other agent frameworks?

A: Jido is an autonomous agent framework for Elixir, built for workflows and multi-agent systems. Key differentiators:

  • OTP-native architecture: Built on GenServer with supervision and fault tolerance built in
  • Immutable agents: Functional state model inspired by Elm/Redux
  • AI optional: Core package provides agent architecture without requiring AI/LLM
  • Explicit effect boundaries: actions may own immediate work; directives describe runtime-owned effects

Compared to LangChain/CrewAI:

  • LangChain: Chain-based orchestration, Python-first
  • CrewAI: Role-playing autonomous agents, team-based collaboration
  • Jido: OTP-native with supervision trees, Elixir/BEAM ecosystem

Q: What is the Agent/Action/Signal/Directive architecture?

A:

  • Agents: Immutable structs plus modules that hold state and implement cmd/2
  • Actions: Receive validated params and context, then return state updates and optional directives
  • Signals: CloudEvents-compliant messages routed into the system
  • Directives: Bare structs that describe effects for the runtime to execute

cmd/2 returns the updated agent plus directives. Directives never mutate agent state; the OTP runtime interprets them for runtime-owned external effects. Actions may still perform work such as API calls, file I/O, or database queries when that result is needed immediately by the action or state transition.

Q: Is AI required for Jido?

A: No! The core package (jido) provides agent architecture and runtime without AI. AI/LLM integration is optional via companion packages:

  • jido_ai: AI/LLM integration for agents
  • req_llm: HTTP client for LLM APIs

Use Jido when software needs to inspect context, choose steps, coordinate agents, and run reliably - AI is optional.

Q: What are the Jido ecosystem packages?

A:

Package Description
req_llm HTTP client for LLM APIs
jido_action Composable, validated actions with AI tool integration
jido_signal CloudEvents-based message envelope and routing utilities
jido Core agent framework with state management, directives, runtime
jido_ai AI/LLM integration for agents

Getting Started

Q: How do I install Jido?

A: Jido is available on Hex.pm:

def deps do
  [
    {:jido, "~> 2.0"}
  ]
end

See jido.run for demos and examples.

Q: How do I create an agent?

A:

defmodule MyAgent do
  use Jido.Agent,
    name: "my_agent",
    description: "My custom agent",
    schema: [
      count: [type: :integer, default: 0]
    ]
end

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

See hexdocs.pm/jido for API reference.

Q: How do I run agents in production?

A: Use AgentServer (GenServer-based) for production deployment:

  • Parent-child agent hierarchies with lifecycle management
  • Signal routing with configurable strategies
  • Instance-scoped supervision plus logical partitions for multi-tenant deployments

Features

Q: What directives are available?

A: Built-in directives:

  • Emit: Emit signals to the system
  • Error: Signal an error from cmd/2
  • Spawn: Spawn child processes
  • SpawnAgent: Spawn child agents
  • AdoptChild: Attach an orphaned or unattached child to the current parent
  • StopChild: Stop child processes/agents
  • StartSensor: Start or replace owner-monitored tagged sensor runtimes
  • StopSensor: Stop tagged sensor runtimes
  • Schedule: Schedule future actions
  • RunInstruction: Execute an instruction at runtime and route the result back to cmd/2
  • Stop: Stop the agent
  • Cron / CronCancel: Register and cancel cron-based schedules

External packages can also define custom directive structs. See Agent Directives for details.

Q: How do plugins work?

A: Composable plugins extend agents with:

  • Reusable capability modules
  • State isolation per plugin with automatic schema merging
  • Easy capability sharing across agents

See Plugins Guide for details.

Q: What execution strategies are available?

A:

  • Direct Strategy: Immediate execution
  • FSM Strategy: State machine workflows

See Strategies Guide and FSM Strategy Deep Dive.

Troubleshooting

Q: I'm getting GenServer timeout errors. What should I check?

A:

  1. Check action execution time
  2. Consider using worker pools for throughput (guides/worker-pools.md)
  3. Review supervision tree configuration

Q: My agents aren't receiving signals. What's wrong?

A:

  1. Check signal routing configuration
  2. Verify signal envelope format (CloudEvents)
  3. Review signal dispatch strategy

Help Resources

Development

Prerequisites

  • Elixir 1.17+
  • Erlang/OTP 26+

Running Tests

mix test

Quality Checks

mix quality  # Runs formatter, dialyzer, and credo

Contributing

We welcome contributions! Please see our Contributing Guide for details on:

  • Setting up your development environment
  • Running tests and quality checks
  • Submitting pull requests
  • Code style guidelines

License

Copyright 2024-2025 Mike Hostetler

Licensed under the Apache License, Version 2.0. See LICENSE for details.

Links

About

🤖 Autonomous agent framework for Elixir. Built for distributed, autonomous behavior and dynamic workflows.

jido.run

Topics

agent workflow elixir ai functional-programming orchestration artificial-intelligence event-driven-architecture

Resources

Readme

License

Apache-2.0 license

Contributing

Contributing
Activity
Custom properties

Stars

1.7k stars

Watchers

23 watching

Forks

104 forks
Report repository

Releases 2

Release v2.3.0 Latest
May 22, 2026
+ 1 release

Packages

 
 
 

Contributors

Languages