GoModel & n8n - GoModel

Overview

n8n has no built-in trace of what its AI nodes send and receive. Pointing n8n at GoModel fixes that without any community node: one OpenAI credential with GoModel’s base URL, and every workflow’s model call is logged, priced, and exported as metrics. The same credential also lets n8n’s OpenAI nodes use Anthropic, Gemini, and any other provider configured in GoModel. Flow: n8n workflow -> OpenAI Chat Model / OpenAI node -> GoModel -> OpenAI/Anthropic/Gemini/... What you get:

Full request and response bodies for every AI call in the GoModel audit log.
Token usage and cost per model, per API key, and per label.
Prometheus metrics and OpenTelemetry traces for AI traffic.
Failover, caching, rate limits, and budgets applied to n8n traffic.
Non-OpenAI models inside n8n’s OpenAI nodes, without extra credentials.

1. Run GoModel next to n8n

Run both services on one Docker network so n8n can reach GoModel by container name:

docker-compose.yml

services:
  gomodel:
    image: enterpilot/gomodel
    environment:
      GOMODEL_MASTER_KEY: change-me
      OPENAI_API_KEY: sk-...
      ANTHROPIC_API_KEY: sk-ant-...
    ports:
      - "8080:8080"

  n8n:
    image: n8nio/n8n
    environment:
      N8N_COMMUNITY_PACKAGES_ENABLED: "false"
    ports:
      - "5678:5678"
    volumes:
      - n8n_data:/home/node/.n8n

volumes:
  n8n_data:

Confirm GoModel lists your models:

curl -s http://localhost:8080/v1/models \
  -H "Authorization: Bearer change-me"

Create a dedicated API key for n8n in the GoModel dashboard (API Keys -> Create API Key) instead of using the master key. Keys carry labels, budgets, and rate limits, so n8n’s spend shows up separately in the dashboard.

2. Create the OpenAI credential in n8n

In n8n open Settings -> Credentials -> Add credential -> OpenAI and fill in:

Field	Value
API Key	The GoModel API key (or `GOMODEL_MASTER_KEY`)
Organization ID	Leave empty
Base URL	`http://gomodel:8080/v1`
Add Custom Header	Optional, see Attribute workflows with labels

n8n tests the credential against GET /v1/models when you save it. GoModel serves that endpoint, so the test should pass and the model picker in the OpenAI Chat Model node lists exactly the models GoModel exposes, including Anthropic and Gemini models and any virtual models.

Use the Docker service name (gomodel) when both run in containers. http://localhost:8080/v1 only works when n8n runs on the host machine.

3. Use it in a workflow

Any node that accepts the OpenAI credential now goes through GoModel:

AI Agent and Basic LLM Chain with an OpenAI Chat Model sub-node.
Embeddings OpenAI for vector stores.
The plain OpenAI node (chat, image, audio operations).

Attach the credential to the OpenAI Chat Model sub-node and pick a model. Because the list comes from GoModel, you can select anthropic/claude-haiku-4-5-20251001 or gemini/gemini-2.5-flash in the same node that used to be OpenAI-only. Both the provider-prefixed ids from the picker and plain ids such as gpt-5-mini work. Run the workflow, then open the GoModel dashboard: the call appears in the audit log with the full prompt, the tool definitions n8n attached, the response, tokens, and cost.

Verified with n8n 2.36 (AI Agent 3.1 and 2.2, OpenAI Chat Model 1.3) and the Calculator tool against OpenAI, Anthropic, and Gemini models routed through GoModel.

Chat Completions or Responses API

The OpenAI Chat Model node has a Use Responses API switch. In n8n 2.x it is on by default for new nodes. Both paths work through GoModel:

Setting	GoModel endpoint	Notes
On (n8n default)	`/v1/responses`	Native for OpenAI. Translated for Anthropic and Gemini; `conversation_id` and OpenAI-hosted tools are not available there.
Off	`/v1/chat/completions`	Works with every provider and every GoModel version.

Tool calling works on both paths for OpenAI, Anthropic, and Gemini models. Turn the switch off if you need a Responses-only feature to be rejected loudly rather than translated, or if you run a GoModel release older than the annotations fix described in Troubleshooting. See Responses compatibility for the exact limits on translated providers.

Attribute workflows with labels

To tell workflows apart in usage reports, send a tagging header from n8n and declare it in GoModel:

In GoModel, set TAGGING_HEADER_1=X-N8N-Workflow (or add it under Settings -> Tagging based on headers in the dashboard).
In the n8n credential, enable Add Custom Header with name X-N8N-Workflow and a value such as invoice-triage.

The header value becomes a label on every usage entry made with that credential; the dashboard’s usage-by-label view then shows the workflow’s requests, tokens, and cost. n8n normalises the header name’s case (X-N8n-Workflow); GoModel matches it case-insensitively. Create one credential per workflow or team if you want separate labels; they can share the same GoModel API key. See Labelling for prefixes, delimiters, and keeping headers from reaching the upstream provider.

Other n8n model nodes

The n8n Anthropic Chat Model credential also has a Base URL field. Point it at http://gomodel:8080 (no /v1) with a GoModel API key to route native /v1/messages traffic through GoModel; the credential test and AI Agent tool calls work (see Anthropic Messages API). In most workflows the single OpenAI credential above is simpler, since it already reaches every provider.

Troubleshooting

Symptom	Fix
Credential test fails with a connection error	n8n cannot resolve the base URL. Check the Docker network and use the service name, not `localhost`.
`401 Unauthorized` in GoModel logs	The n8n credential key does not match a GoModel API key or `GOMODEL_MASTER_KEY`.
Model picker is empty	GoModel has no provider configured for that model. Check `GET /v1/models` and provider environment variables.
`Cannot read properties of undefined (reading 'map')` with Use Responses API on	GoModel releases before the `annotations` fix omitted `annotations: []` from `output_text` parts, which n8n’s OpenAI client requires. Upgrade GoModel, or turn Use Responses API off.
Same error persists right after upgrading	GoModel’s response cache is replaying a body stored before the upgrade. Change the prompt or clear the cache.
Responses API errors on Anthropic or Gemini	Drop `conversation_id` and hosted tools; those are not translatable.
Model answers look “off”	Check GoModel guardrails; a `system_prompt` guardrail rewrites the prompt n8n sends.

Notes

GoModel failover and caching apply to n8n traffic automatically; no change in n8n is needed. With the response cache on, repeated identical prompts return cached answers, which is usually what you want for retried workflow runs.
Streaming to the n8n chat trigger was not part of the verification run; non-streaming agent runs were.
Community nodes such as n8n-nodes-openai-litellm exist to add gateway metadata to n8n requests. They are not required with GoModel; the built-in credential and tagging headers cover the same need.
For dashboards and alerts, see Prometheus metrics and OpenTelemetry.

​Overview

​1. Run GoModel next to n8n

​2. Create the OpenAI credential in n8n

​3. Use it in a workflow

​Chat Completions or Responses API

​Attribute workflows with labels

​Other n8n model nodes

​Troubleshooting

​Notes

Overview

1. Run GoModel next to n8n

2. Create the OpenAI credential in n8n

3. Use it in a workflow

Chat Completions or Responses API

Attribute workflows with labels

Other n8n model nodes

Troubleshooting

Notes