n8n

The trait that decides it is everything around the agent. Real automations are mostly plumbing between systems, and here the model node sits beside the connectors, triggers and error handling you were going to need anyway. That is why this outlasts the pure agent builders for anything that runs on a schedule and touches five systems.

It is not a coding tool and it will not read your repository. Pick it when the agent is one interesting node in a workflow full of boring ones. Pick Langflow when the agent is the point and you want a smaller thing to reason about.

The architectural tension is inherited. This engine was designed for steps that either succeed or fail and retry the same way every time. An agent node breaks that contract: it can succeed differently on each run, and the documented mitigation is inserting a human approval step, which is a control that does not scale to a workflow firing hourly.

Instrument the agent node separately and alert on output shape, not just on errors. What it does right: workflows publish over MCP, so the platform becomes a tool provider for outside clients instead of a place things go in and never come out.

Two distinct mechanisms are worth separating. 1. The workflow builder takes a description and emits a graph, so the model output is a reviewable artefact a human edits before anything runs; the risk of a bad generation is bounded by that review. 2. A chat hub fronts multiple models, which is routing, not orchestration, and should not be confused with the first.

Guardrails are named as a capability but not specified, so the reader cannot tell whether they are output filters, schema validation or approval gates. No benchmark accompanies any of it.

This is the strongest distribution story on the board and it was not bought with agent hype. Two hundred thousand stars, a first release in 2019, and years of integration surface accumulated before anyone needed an agent node. Enterprise terms are negotiated privately, which for a company with this much bottom-up adoption means genuine pricing leverage rather than an empty page.

Moat: switching cost measured in the number of workflows a customer has already built, which is the most durable kind. Likely acquirer: an automation incumbent, and the price would be uncomfortable for them. Position: long, and it does not need the agent narrative to work.

Finally, pricing that does not multiply by headcount. The published cloud ladder is EUR 20, 50 and 667 a month, charged for the workspace rather than per person, so sixty engineers cost the same as six. The business tier is billed annually, which is a commitment I have to defend in a budget cycle but a predictable number once defended.

It runs unattended, so it belongs in our pipeline properly, and the self-hosted route answers the residency question outright. Directory integration sits in the negotiated tier. Approved with conditions: self-hosted first, and a named owner for the workflows.

Let us be precise about the licence, because the marketing is not. This is a Sustainable Use Licence with a separate commercial one on top, which means I can read it, run it and change it for myself, and I cannot offer it to anyone else as a service. Fair-code is a real category and it is not open source, whatever the readme implies.

Inside those limits the deal is good: one docker run and a volume gets me the whole platform on my own box, local model endpoints are supported, and nothing phones home that I cannot see. Grudging respect, with the licence noted.