About

Misconfigured automation workflows cost development teams hours of debugging. A missing connection between nodes, an incorrect typeVersion, or a malformed parameters object can silently break an entire pipeline. This tool parses natural language descriptions and outputs structurally valid n8n workflow JSON conforming to the n8n schema specification. It maps keywords to node types from a dictionary of 30+ supported n8n nodes, resolves execution order via topological dependency analysis, and auto-generates node positions using a layered graph layout algorithm with 250px horizontal spacing. The validator checks root structure, per-node integrity, connection graph consistency, and duplicate ID collisions. Note: this tool performs client-side rule-based generation. It does not call an external AI model. Complex multi-branch workflows with conditional logic may require manual refinement of the generated output.

Formulas

The converter applies a weighted keyword matching algorithm to determine node selection confidence. For each candidate node type N, the match score is computed as:

S_N = k∑i=1 w_i ⋅ m_ik∑i=1 w_i

Where w_i = weight of keyword i in the node's dictionary (range 0.1 to 1.0), m_i = 1 if keyword i is found in the prompt, 0 otherwise, and k = total keywords for node type N. A node is selected when S_N ≥ 0.15 (threshold). Node positioning follows a layered grid formula:

x_j = 250 + layer_j × 250 , y_j = 300 + offset_j × 150

Where layer_j is the topological depth of node j (trigger = 0, first action = 1, etc.) and offset_j is the vertical index within nodes sharing the same layer. Validation checks 4 structural invariants: root schema conformance, node schema conformance, connection graph integrity (no dangling references), and unique ID constraint.

Reference Data

Node Type	n8n Type Key	Trigger Keywords	Category	Default Version
Webhook	n8n-nodes-base.webhook	webhook, http trigger, api endpoint	Trigger	1.1
Cron / Schedule	n8n-nodes-base.scheduleTrigger	cron, schedule, every, daily, hourly	Trigger	1.1
Manual Trigger	n8n-nodes-base.manualTrigger	manual, start, begin	Trigger	1
HTTP Request	n8n-nodes-base.httpRequest	http, fetch, request, api call, get, post	Action	4.1
Slack	n8n-nodes-base.slack	slack, message slack, notify slack	Action	2.1
Email (Send)	n8n-nodes-base.emailSend	email, send email, mail, notify email	Action	2.1
Gmail	n8n-nodes-base.gmail	gmail, google mail	Action	2.1
Telegram	n8n-nodes-base.telegram	telegram, send telegram	Action	1.1
Discord	n8n-nodes-base.discord	discord, send discord	Action	2
PostgreSQL	n8n-nodes-base.postgres	postgres, postgresql, database, db insert, sql	Action	2.3
MySQL	n8n-nodes-base.mySql	mysql, mariadb	Action	2.3
MongoDB	n8n-nodes-base.mongoDb	mongodb, mongo	Action	1.1
Google Sheets	n8n-nodes-base.googleSheets	google sheets, spreadsheet, gsheet	Action	4.2
Airtable	n8n-nodes-base.airtable	airtable	Action	2.1
IF	n8n-nodes-base.if	if, condition, check, filter, branch	Logic	2
Switch	n8n-nodes-base.switch	switch, route, case	Logic	3
Merge	n8n-nodes-base.merge	merge, combine, join	Logic	2.1
Code	n8n-nodes-base.code	code, javascript, transform, process, parse, script	Action	2
Set	n8n-nodes-base.set	set, assign, map, rename field	Action	3.3
OpenAI	@n8n/n8n-nodes-langchain.openAi	openai, gpt, ai, chatgpt, analyze, sentiment	AI	1.3
RSS Feed Read	n8n-nodes-base.rssFeedRead	rss, feed, blog monitor	Trigger	1
Wait	n8n-nodes-base.wait	wait, delay, pause, sleep	Logic	1.1
Split In Batches	n8n-nodes-base.splitInBatches	batch, split, chunk, loop	Logic	3
No Operation	n8n-nodes-base.noOp	noop, nothing, placeholder	Logic	1
Respond to Webhook	n8n-nodes-base.respondToWebhook	respond, return, reply webhook	Action	1.1
Date & Time	n8n-nodes-base.dateTime	date, time, format date, timestamp	Action	2
Crypto	n8n-nodes-base.crypto	hash, encrypt, hmac, sha256	Action	1
XML	n8n-nodes-base.xml	xml, parse xml, convert xml	Action	1
HTML	n8n-nodes-base.html	html, scrape, extract html	Action	1.1
FTP	n8n-nodes-base.ftp	ftp, upload file, sftp	Action	1
Error Trigger	n8n-nodes-base.errorTrigger	error, on error, error trigger, catch error	Trigger	1

Frequently Asked Questions

The converter supports over 30 node types including triggers (Webhook, Schedule, Manual, RSS, Error), actions (HTTP Request, Slack, Email, Gmail, Telegram, Discord, PostgreSQL, MySQL, MongoDB, Google Sheets, Code, Set, OpenAI, FTP), and logic nodes (IF, Switch, Merge, Wait, Split In Batches). Selection uses weighted keyword matching against each node type's dictionary. For example, the words "slack" and "notify slack" map to the Slack node with weight 1.0, while "message" alone maps with weight 0.3. A node is included when its aggregate match score exceeds the threshold of 0.15.

The converter builds a dependency graph based on the semantic order keywords appear in the prompt. Trigger nodes are always placed at layer 0 (leftmost). Subsequent nodes are layered by their dependency depth. Connections use the n8n format with source node name, output index (default 0), destination node name, and input index (default 0). For IF/Switch nodes, the converter creates branched connections: output 0 for the "true" path and output 1 for the "false" path, each linking to appropriate downstream nodes.

Unrecognized keywords are ignored during scoring. If no action nodes score above the 0.15 threshold, the converter generates a minimal workflow with only the detected trigger and a Code node placeholder, flagging it with a warning. If no trigger is detected, a Manual Trigger is inserted by default. The validation step will report warnings for workflows with fewer than 2 nodes or disconnected subgraphs. All ambiguities are surfaced in the validation report rather than silently resolved.

Yes. The output follows the n8n workflow export schema with all required fields: name (string), nodes (array of objects with id, name, type, typeVersion, position, parameters, credentials placeholder), connections (object keyed by source node name, containing main output arrays), and settings (executionOrder set to 'v1'). The JSON can be copied and pasted directly into n8n's "Import from JSON" dialog. Credential placeholders will need to be configured in n8n after import.

The layout algorithm assigns each node a layer (horizontal position) based on topological depth from the trigger. Within each layer, nodes are stacked vertically with 150px spacing. IF and Switch nodes cause branch splits: downstream nodes on the "true" branch get a negative vertical offset, "false" branch nodes get a positive offset. This prevents overlap for workflows up to approximately 15 nodes. Beyond that, manual repositioning in n8n's editor may be needed. The algorithm processes nodes in BFS order to maintain left-to-right flow consistency.

Four validation passes run sequentially. Pass 1: Root schema check verifies presence of "name" (string), "nodes" (non-empty array), and "connections" (object). Pass 2: Node schema check ensures each node has "id" (string, UUID format), "name" (string), "type" (string matching known n8n type patterns), "typeVersion" (number), and "position" (2-element number array). Pass 3: Connection integrity verifies all source and target node names referenced in connections exist in the nodes array. Pass 4: Uniqueness check detects duplicate node IDs and duplicate node names, which would cause import errors in n8n.