Comparison

OpenRig vs n8n

Published April 2026

n8n and OpenRig started from different questions about where agents fit in your work. n8n asks what if AI is one capability among hundreds of automation integrations, slotted into a visual workflow. OpenRig asks what if Claude Code and Codex are extraordinarily capable on their own, and the interesting design work is in running several of them together as a team. Same general space, different paradigms — and the comparison is useful because the paradigms produce different shapes for different users.

Why I publish comparisons. This category is new. The vocabulary isn't settled. Nobody has a complete mental model of what's in the space, including me. The comparisons here are an attempt to find the shape of the category — what approaches are being taken, what each approach looks like, what we might learn from each other. Not a bake-off.

Two different paradigms about where agents fit

n8n started from: what if business and developer workflows could be visual node-based pipelines that anyone could compose, with hundreds of integrations available and AI as one more capability you slot in? The answer is a workflow platform. You drag nodes onto a canvas, connect them, configure a trigger, and the workflow runs when the trigger fires. In 2026 the platform added genuine multi-agent capabilities — AI Agent nodes, Supervisor-Worker patterns, an @n8n/agents TypeScript SDK, MCP Server Triggers that let workflows be called as tools by external AI systems. The unit of work is the workflow execution.

OpenRig started from a different question: what if Claude Code and Codex are already extraordinarily capable, and what I actually want is to run several of them together as a coordinated team I can attach to and talk to? The answer is a topology-first system. OpenRig doesn't treat agents as nodes in a fired-on-trigger workflow — it treats them as persistent participants in a topology you operate. Each agent is a real Claude Code or Codex session, alive in a tmux pane, ready to take work. The unit of work is the rig.

Neither paradigm is obviously right. They're different bets about what agents are for and how you work with them. n8n bets that agents belong inside automation pipelines that fire on events. OpenRig bets that agents belong as continuously-running participants in a topology you operate at the terminal.

Side-by-side

Axis	OpenRig	n8n
What the primitive is	The rig — a topology of running shipped harnesses.	The workflow — a graph of nodes that fires on a trigger and runs to completion.
What runs as a node	Claude Code and Codex as they ship. Terminal processes too.	Workflow nodes — hundreds of integrations (HTTP, Slack, databases, etc.) plus AI Agent nodes that wrap LLM calls.
How you start	rig up demo. A shipped spec gives you 8 agents in one command.	Open the visual editor. Drag nodes onto the canvas. Connect them. Configure a trigger. Save and activate.
How agents coordinate	Directly through tmux (rig send / broadcast / chatroom). Rig queue for durable cross-pod signaling, shipping soon.	Workflow execution flow. Multi-agent via Supervisor-Worker or Router-Branch patterns. Workflows can call other workflows as tools via MCP Server Trigger.
When agents are alive	Persistent. The agents are running. You attach to them and talk to them.	On execution. A workflow fires, runs, ends. Agents exist for the duration of an execution.
Who operates it	A developer at the terminal.	A no-code or low-code operator wiring up workflows in a visual editor.
License	Apache 2.0.	Sustainable Use License (fair-code, source-available with commercial-use restrictions).

What it looks like when agents are persistent

Because OpenRig agents are continuously running rather than firing on a trigger, the way you work with them is different:

You boot 8 agents with one command. rig up demo gives you orchestration, development, and review pods, all running Claude Code and Codex, ready to take work. You don't configure a workflow before you can use them; the spec library ships proven shapes to start from.
You attach and talk. Each agent is a real Claude Code or Codex session in a tmux pane. You read its screen, type into it, watch it work. The back-and-forth is live; there's no execution log to read after a workflow finishes.
Agents don't end when a task ends. A workflow fires, runs, and finishes. A rig keeps going. You can hand the same agent ten different things over a day, or a week. The agent accumulates context and stays ready.
Mixing runtimes is the default. Claude Code and Codex coordinate as peers in the same rig. Different models catch different things; a rig treats cross-runtime as the normal case.
You save the whole rig and come back to it. rig down snapshots the topology; rig up <name> restores it. Sessions resume where they were able to. The rig is operator infrastructure that survives your machine restarting.

The tradeoffs are real. n8n is the right shape for trigger-driven automation across many integrations, especially when AI is one capability among many you're composing. If your work is wiring webhooks, scheduling jobs, or building cross-system pipelines that include some LLM steps, n8n is well-designed for that and OpenRig is the wrong tool. OpenRig is the right shape for running coding agents together as a persistent team you operate from the terminal.

Other Comparisons

Compare other approaches

Quickly build a mental model for multi-agent harnesses and topologies by comparing the leading approaches, including more traditional multi-agent systems.

Compare