Process isolation
Each agent runs in its own lightweight process with isolated memory. One agent failing never takes down another.
Jido is an agent framework for Elixir. Define agents, give them tools, and let them work together, with fault tolerance and supervision built in.
New to Elixir? Start here. • Already an Elixir developer? Jump to the expert guide.
From single-purpose assistants to teams of agents that coordinate autonomously.
Agents that read, analyze, and refactor code across repositories.
Multi-step research agents that find sources, verify facts, and produce reports.
Extract, classify, and route documents: invoices, contracts, support tickets.
Agents that resolve issues using your knowledge base and escalate when needed.
Agents that watch systems, diagnose problems, and run remediation playbooks.
Agents that collect, transform, and load data from multiple sources on schedule.
Define an agent, start it supervised, ask it questions.
defmodule MyApp.SupportAgent do
use Jido.AI.Agent,
name: "support_agent",
description: "Customer support agent",
tools: [MyApp.Tools.KnowledgeBase,
MyApp.Tools.TicketSystem],
system_prompt: "You help customers resolve product issues."
end
Agent frameworks are everywhere. Here's what makes this one different.
When an agent crashes, its supervisor restarts it automatically with clean state. No orchestrator, no manual recovery, no downtime.
Learn more →Agents coordinate through typed Actions and Signals, not prompt chains. Debug and test each step independently, just like regular code.
Learn more →Built-in telemetry and tracing across every agent. See what's happening, trace workflows across processes, catch problems before users do.
Learn more →Add one agent to your existing Elixir app. No rewrite, no platform migration. Add more agents, tools, and packages only when you need them.
Learn more →The same runtime that powers WhatsApp and Discord turns out to be ideal for agent workloads.
Each agent runs in its own lightweight process with isolated memory. One agent failing never takes down another.
OTP supervisors detect crashes and restart agents in milliseconds. Failure recovery is built into the runtime, not bolted on.
The BEAM scheduler handles thousands of concurrent agent processes with true parallelism. No thread pools, no async/await gymnastics.
Start with the core. Add AI, tools, and integrations as you need them.
jido jido_action jido_signal
jido_ai req_llm llm_db
ash_jido jido_messaging jido_otel
Go from zero to a running, supervised agent in under ten minutes.