See the AI spend hiding inside your client's n8n stack.

The problem nobody catches in time

An n8n workflow with three LangChain Agent nodes, running every fifteen minutes against a vector store with 200 retrieved documents per call, can cost more in a month than the client's entire automation retainer. n8n itself doesn't surface this. The executions view shows duration, not dollars. The billing email arrives from OpenAI or Anthropic, addressed to the client's accountant.

By the time you find out, you've spent two weeks of attention you didn't budget — explaining the spend, redesigning the workflow, and rebuilding the trust.

What WorkflowRadar tracks

We extract token usage from every LangChain node execution that n8n already records. For each execution we compute:

• Input tokens × the model's input price
• Output tokens × the model's output price
• Per-model breakdown when a workflow uses more than one

We ship explicit pricing maps for six providers — OpenAI (4o, 4o-mini, o1, o3, o4), Anthropic (Claude 3.5, 3.7, 4.x), Google (Gemini 1.5, 2.x), Azure OpenAI, OpenRouter (proxied OpenAI / Anthropic / Google / Meta / Mistral models), and AWS Bedrock (Anthropic Claude on Bedrock, Llama 3 on Bedrock, Mistral on Bedrock). Token tracking itself works for any LangChain node that emits a standardusage shape — the pricing map is the part that turns tokens into dollars.

Azure deployment names: Azure customers configure custom deployment names (e.g. my-gpt4-prod) that don't match canonical model IDs. The workspace's default model setting is the price proxy for unrecognized deployment names — pick the underlying model (e.g. gpt-4o) in the workspace AI cost settings and your custom deployments inherit its rates.

The result rolls up two ways:

• Per workflow: 30-day tokens in, tokens out, and cents — visible in the workflow detail drawer and on the per-workflow Scorecard PDF.
• Per organization: total 30-day spend across every workspace, with the top-spending workflows surfaced first.

Threshold alerts

In each workspace's alert settings you set a monthly AI spend ceiling — say $200. Every sync, we project the month based on the current run-rate. When the projection passes 80% of your threshold, you get an email with:

• Current month-to-date spend
• Projected month-end total
• The five workflows driving the most cost, with their individual month-to-date totals

The alert dedupes per workspace per 24 hours, so a flaky cost spike doesn't page you six times.

How it shows up in client reports

The Value PDF (one of four report types — Status, Audit, Value, per-workflow Scorecard) has a dedicated AI cost section. It lists total tokens in/out, dollar spend, month-over-month delta, and the top three workflows by cost. White-labeled to your branding, so it lands on your client's desk looking like your own deliverable — not a WorkflowRadar one.

Privacy of the inputs

We only read token counts and timing — never the prompt content or the model output. The Anthropic / OpenAI / Google requests we analyze were already made by your workflow; we just read the usage metadata n8n stores. The product never sends your client's AI input data anywhere.

Available on

AI cost monitoring is part of the AI add-on. Free for the first 60 days of any paid tier — see pricing for details.