Live Demo

Watch a workflow survive a crash.

Hit Run, then crash it mid-execution. The engine recovers from the last persisted checkpoint. External effects with unknown outcomes still require reconciliation before retry.

This browser simulation uses the same checkpoint model as the engine. It does not call a backend; run the linked sequence locally to test real process termination and external-effect reconciliation.

Source-available (BUSL-1.1)/Written in Rust/Single binary/PostgreSQL or SQLite

AI Content Pipeline

Generate content with an LLM, quality-check it, route through human approval, then publish. See crash recovery resume a 4-step AI workflow.

Execution Timeline · 4 steps · 0s

Click Run to start the sequence. Try Simulate Crash mid-execution to see durable recovery.

🧠

Generate contentllm_call

pending

🧠

Quality checkllm_call

pending

👤

Human reviewhuman_review

pending

🌐

Publish to CMShttp_request

pending

Before you run

Run takes about 8 seconds. Crash after the first completed step to see the snapshot restore without repeating saved work.

Instance Stateidle

Context

{
  "topic": "How to build durable workflows",
  "audience": "Backend engineers",
  "tone": "Technical but approachable",
  "format": "blog_post"
}

Saved step outputs (0)

No outputs yet

Event Log

No events yet

What would happen without durable execution?

A conventional in-memory runner restarts from the beginning, repeating already completed external calls. Orch8 persists each completed output and resumes at the first unfinished step.

View the engine source ↗

What you just saw

Each completed step wrote a durable snapshot. Recovery restored those outputs, skipped finished work, and resumed at the first unfinished step. Human approval paused the workflow without keeping a worker occupied.

Understand the model Study crash recovery Run it locally

Ready to build durable workflows?

Self-host free with no feature gates, or use Orch8 Cloud for managed infrastructure.

Start free Self-Host Quickstart