Open Workflow Library is an open workflow intelligence project for collecting, indexing, validating, repairing, and eventually generating automation workflows across frameworks. It starts with a large n8n workflow collection and is being expanded into a universal workflow knowledge base and tooling layer.
Workflow automation is fragmented across many frameworks — n8n, Dify, LangGraph, Node-RED, Make, Zapier, Pipedream, Airflow, Temporal, and others — each with its own file format, node model, and ecosystem. Most public workflow collections are stuck in two failure modes:
- Single-framework, no validation. Large dumps of workflow JSON files that may or may not work, with no signal about which ones do.
- Generator-only, no library. Tools that promise prompt-to-workflow generation without a grounded knowledge base, so they hallucinate node types, integrations, and patterns that don't exist.
This project's bet is that the right shape is library + catalog + knowledge base + tooling, with prompt-to-workflow generation built on top of all three — not floating above them.
- A large collection of n8n workflows (the existing dataset that ships with this repository).
- A set of JSON schemas defining a Universal Workflow IR, catalog metadata, repair proposals, and learning events.
- A read-only audit tool that catalogues the existing workflows, detects risks, and produces structured reports.
- The beginnings of an LLM-usable wiki of patterns, integrations, repair rules, and failure cases.
- A documented architecture for what comes next.
A universal workflow intelligence layer with five capabilities working together:
- Library — workflows from multiple frameworks, stored in their native format.
- Catalog — a structured index of those workflows, including detected integrations, triggers, risks, and validation status.
- Universal Workflow IR — a framework-agnostic intermediate representation that lets the rest of the system reason about workflows without caring which framework they came from.
- Validator and Repair Engine — schema and behavioural validation, plus human-reviewed repair proposals for stale or broken workflows.
- Prompt-to-Workflow Generator — given a goal in natural language, retrieve the relevant catalog entries and wiki content, assemble a Universal Workflow IR, validate it, and export to the target framework.
All five capabilities feed a human-reviewed self-improvement loop: failed validations and successful repairs produce learning events, which can be promoted into wiki entries or repair rules — only after human review.
n8n is a good starting dataset because the workflow file format is well understood, the ecosystem is large, and the workflows are readable JSON. But the long-term value is in the IR and the knowledge base, not the file format. A workflow library that only understands n8n is a workflow library that has to be rewritten every time someone adopts a different tool. We'd rather pay the IR tax up front.
- Not a vendor. No hosted service, no managed runtime.
- Not autonomous. The self-improvement loop is human-gated. Workflows are never mutated without a reviewed proposal.
- Not a replacement for the workflow engines themselves. Workflows still run on n8n, Dify, LangGraph, and so on. This project is the layer around them.
- We are not trying to be the largest workflow library by file count. We're trying to be the most useful one to retrieve from, validate against, and generate from.
- We are not trying to support every framework in V1. The supported-frameworks list is honest about what exists today versus what is planned.
- We are not building a workflow IDE.
- architecture.md — system design.
- supported-frameworks.md — honest status per framework.
- prompt-to-workflow.md — planned generator pipeline.
- self-improvement.md — learning loop and human review gate.
- security.md — security and contribution constraints.