Package Information
Available Nodes
Documentation generating…
Documentation
Valen for n8n
Route support tickets to the right team, flag images for review, or triage alerts with one Valen node. Ask a concrete question and connect the named outputs to your next steps. Yes/No routing includes an unsure output for human review.
Your original fields, binary data and item links stay with each item. Answers and probabilities appear under valen (replacing an existing field of that name).
Install
In self-hosted n8n, open Settings → Community nodes → Install and enter:
@valen-team/n8n-nodes-valen
Install the package, then add a node and search for Valen. Requires Node.js 22.16+; compatible with the tested n8n 2.45.6. This community package is not verified for n8n Cloud.
For a manual installation, run as the user that runs n8n:
mkdir -p ~/.n8n/nodes
cd ~/.n8n/nodes
# Run npm init -y first if this directory has no package.json.
npm install @valen-team/n8n-nodes-valen@0.2.0 --omit=dev --legacy-peer-deps
Restart n8n after manual installation. In Docker, use the persistent /home/node/.n8n/nodes directory as the node user. npm package · n8n installation guide
Connection
Create a Valen API credential. The default free endpoint needs no API key. Saving/testing the credential checks /v1/models. You can also use https://llm-proxy.openxxlab.com with your key, or a self-hosted server. A trailing /v1 is optional. Keep the key in the credential, not in workflow parameters.
Choose Model Name or ID from the live model list. For a self-hosted model whose ID doesn't start with valen, switch this field to Expression and enter a literal expression such as {{ "my-model" }}.
Try the included workflow
Import examples/support-routing.json through n8n's Import from File menu. Select your Valen credential on Route tickets, then execute the workflow. It creates three synthetic tickets and routes them to Billing, Technical or Human review. Request failures go to the separate Request failure output. No external actions run. Inspect valen.branch, valen.probabilities and valen.model on each output. Predictions can vary by model; the example is not an assertion of accuracy.
The review option describes out-of-scope requests explicitly: Choice always selects one of the supplied options. Use a separate Yes/No or workflow condition when a probability gate is needed; Choice and Score Route have no automatic uncertainty gate.
Operations
- Evaluate: supply a Decisions API JSON array of 1–8 questions. Each needs a unique
name,type, andinstructions. Choice addschoices: [{value, description}]; Score addslevels: [{label, description}]. Abboxquestion describes what to find in the image from Image Binary Field and returnsbbox([x1, y1, x2, y2]),pointandconfidence, as fractions of the image's width and height. Results appear undervalen.answers.<name>. - Route / Yes/No: write one statement/question. Outputs are
yes(probability ≥ 0.7),no(≤ 0.3), andunsure(between), with editable thresholds. Require0 ≤ No at < Yes at ≤ 1. - Route / Choice: enter 2–16 distinct options, one per line,
valueorvalue: description. Each creates a named port. Outputs use the selected choice. - Route / Score: enter ordered levels in the same format. Route uses the most probable level, with the earlier level winning a tie. It does not round the expected score.
Route options and question type are fixed configuration, so ports stay the same for every item. Questions and input can use per-item expressions. Text is sent unchanged; JSON and Input Item use two-space JSON with Unicode preserved.
For Image Binary Field, enter the binary property from an earlier node, for example data. File bytes are validated as PNG, JPEG or WebP and sent inline with the input. Fetch remote files with an earlier n8n HTTP Request node. This node accepts one image, not a URL or video.
Errors and data
HTTP 429/502/503/504 and supported network failures retry twice, after 0.25 and 0.5 seconds. Each attempt times out after 120 seconds. Other errors stop immediately, including malformed or incomplete answers. API errors include HTTP status and request ID when supplied.
Use On Error → Continue (using error output) to send failures to n8n's separate Error port, as the example does. Failed items carry error and details.input (the original input), without a normal answer or branch. Stop Workflow stops the run instead. The legacy Continue (using regular output) option sends failures to unsure for Yes/No or the first port otherwise; use the separate Error port for action routing.
Inputs, images and questions are sent to the configured endpoint. The node adds no logs or storage; n8n may retain execution data. See Valen data handling.
Build from source
Use Node.js 22.16+ and run npm ci --ignore-scripts, npm test, then npm pack in this directory. See RELEASING.md for maintenance and SPEC.md for shared behavior.
