Why?
It pisses me off when an AI agent ignores instructions, churns out slop, and then lies about it.
It says, “I did everything.” You look at the result, and it did nothing. Well, almost nothing: maybe 5% right, another 40% of made-up slop, and 95% of what you actually asked for lost, forgotten, or ignored.
And then: “You’re absolutely right. I fucked up.”
I know the feeling, so I went looking for a fix. I couldn’t find one, and gradually cobbled together my own harness out of shit and sticks for deterministic end-to-end execution — so an agent finishes a workflow and checks the result instead of ignoring instructions and calling unfinished work “done.” I also added an SDD pipeline and other automations for routine processes where AI is actually useful.
I let a few friends who were just as fed up try it. They liked it, so I decided to turn it into a product.
QFactory.io is that product: an open-source engine called Q-Core for running workflows, plus a community registry of workflow templates and components. It runs locally and it’s free. To work with an LLM, use your Claude or ChatGPT subscription, or the OpenRouter API.
To get started, find a workflow template you like on the site, copy its prompt, and give it to your AI agent. It will install and set everything up.
To create your own workflow, just describe what you need and ask your agent to build it. On your first run, mention QFactory.io in the request. The agent will find our MCP instructions and generate the config.
Why not
| QFactory | n8n | |
|---|---|---|
| Agent-native | Lives inside the AI agent you already work with — Claude Code, Codex or Cursor. No new app, no canvas, no second place for your work. | Runs on n8n Cloud or on a self-hosted n8n instance. |
| Easy to start | One prompt is the whole setup: paste it into your agent, and it installs, configures and runs the workflow for you. | Start from an n8n template, then configure it for your instance and credentials. |
| Workflow as a component | Every workflow is a building block: drop a finished workflow into another one and reuse what already works. | One workflow can call another as a sub-workflow. |
| Done = criteria + accept/reject | “Done” is not the agent’s opinion: the workflow checks explicit criteria, and you accept or reject the result. | — |
Agent-native
- QFactory
- Lives inside the AI agent you already work with — Claude Code, Codex or Cursor. No new app, no canvas, no second place for your work.
- n8n
- Runs on n8n Cloud or on a self-hosted n8n instance.
Easy to start
- QFactory
- One prompt is the whole setup: paste it into your agent, and it installs, configures and runs the workflow for you.
- n8n
- Start from an n8n template, then configure it for your instance and credentials.
Workflow as a component
- QFactory
- Every workflow is a building block: drop a finished workflow into another one and reuse what already works.
- n8n
- One workflow can call another as a sub-workflow.
Done = criteria + accept/reject
- QFactory
- “Done” is not the agent’s opinion: the workflow checks explicit criteria, and you accept or reject the result.
- n8n
- —