What n8n is, in its own words
n8n describes itself as “AI agents and workflows you can see and control” — “Build visually, go deep with code, connect to anything. Every step of your agents’ reasoning, traceable on the canvas. Deploy on your infrastructure or ours.” Its own summary of the product is “Automate without limits”, backed by 500+ integrations plus custom API calls for anything not already a node. (Source: n8n.io .)
That is worth stating carefully, because the lazy version of this comparison is out of date. n8n today is not “an automation tool with an AI node bolted on”. It ships multi-agent architectures, RAG, model swapping between cloud and offline, MCP connectivity, human-in-the-loop approvals, guardrails, structured inputs and outputs, native evaluations, and per-execution prompt/response inspection.
So the difference that decides this is not on a feature list at all. It is what you author and where the record lives.
The difference in one table
| n8n | Forge | |
|---|---|---|
| What you author | A workflow graph on a canvas — nodes, edges, branches | A brief in plain English; Forge proposes the roles, transitions and validator for your approval |
| Unit of work | A node execution | A persona’s handoff — a named role’s complete piece of work |
| Who checks a step | You, by placing approval nodes and guardrails where you want them | A management agent, on every handoff by default — pass, send back with enrichment instructions, or escalate |
| A rejected step | Re-run the workflow, or branch on the error | A new commit; the previous attempt stays in history and git diff shows what changed |
| Where the record lives | Execution traces on the canvas, inside n8n | A branch in your repository — forge/workflow/{workflowId}/run/{runId} — with spec, brief and every handoff as commits |
| Model choice | Connect any model; swap cloud and offline without rebuilding | A routing gateway across providers with priority weights and automatic failover, including self-hosted and any OpenAI-compatible endpoint |
| Where it runs | Cloud, Docker, on-prem, air-gapped | Managed SaaS, or self-hosted runners on your own hardware |
| The hard part it solves | Moving data between systems on a trigger | Judging whether the previous step was good enough to build on |
Where Forge is different
You describe the team, not the graph. The input is a brief — “a researcher who gathers sources, an analyst who re-opens each one and verifies the quote, and a writer who turns it into something I can send a client”. Forge proposes the roles, the transitions and the management agent, and waits for approval before it spends anything. See Workflow Engine .
Validation is the default edge. In n8n, human-in-the-loop is a node you place. In Forge, a management agent reads every handoff and decides: pass it on, send it back with specific enrichment instructions, or escalate to a person. That is the loop, not a feature you remember to add. See Multi-Agent Pipeline .
The audit trail is git, and it is yours. Every run gets a branch in a repository you own. The spec, the original brief, and each persona’s structured and freeform handoff are commits, with the persona recorded as Co-authored-by — so the history reads like a team’s git log. Send a persona back for enrichment and the retry is another commit; git diff shows exactly what changed between attempts. If we vanished tomorrow, your workflow history is still a git repo on your disk. See Git-Backed Audit
.
Provider routing, not a model dropdown. Requests route across a provider pool with priority weights and automatic failover — Anthropic, Google, Z.AI, Ollama, MiMo, self-hosted models, and any OpenAI-compatible endpoint. See Multi-Provider Routing .
The one number a single agent cannot report
A rejection rate is a number only a system with something able to reject can have. n8n reports that an execution succeeded; it has no place to record that the output was weak, because no node in a graph is allowed to hold that opinion. Here is ours.
| Twelve weeks of Forge building Forge | |
|---|---|
| Runs dispatched | 322 |
| Handoffs judged by a management agent | 1216 |
| Sent back to their author for rework | 56 |
| Escalated to a human | 24 |
| Runs where the supervisor stopped at least one handoff | 51 |
| Runs that completed | 215 |
| Runs that failed, escalated or were cancelled | 103 |
| Median wall-clock, dispatch to finished run | 39 minutes |
| Median handoffs per completed run | 5 |
| Pull requests merged, across 7 repositories | 202 |
Where these came from. One workflow — Main Forge SDLC, the one that builds Forge — every run it made in twelve consecutive weeks, counted from the control plane’s own records. Not a sample, not a pilot, not a customer case study we cannot show you.
Forge stopped its own work 80 times. 56 handoffs went back to their author with specific rework instructions; 24 went to a person. One run in six was interrupted by its own supervisor before any human looked at it. That is precisely the review a single-agent tool leaves on your desk — here it is priced at one model call per edge, and counted.
It merged 202 pull requests across 7 repositories in the same period — roughly 17 a week, into the platform you are reading about, with a human performing every merge. Organisation-wide, agent-authored work merged at 773 of 875 opened. Forge is not a demo running beside the product; it is how the product gets built, and these are its commits.
A finished run takes about 39 minutes and 5 handoffs. Median dispatch-to-done — and inside those 39 minutes sit the reviews that stopped 80 pieces of work from reaching the next role in the state they were first written in.
Every one of those 56 returns is a step that, in a node graph, would have succeeded and moved on. The workflow would have been green. The work would have been wrong.
Ask us to walk you through the runs themselves — the branch, the handoff commits, the verdicts that sent work back, and the ones that went to a person. Every other number on this site belongs to a named run and says so: 18 sources gathered, 11 verified by direct retrieval, 4 corroborated and 2 shipped marked unverifiable, on the research use-case page .
Sources
- n8n.io — hero copy, AI capabilities, deployment options.
- Forge claims on this page link to the corresponding feature pages , which describe the mechanism in detail.