OdinLabsTalk to Odin
Menu
AI Gateway

Check your AI
Gateway connection.

Connect an existing tool to the Gateway configured for your team. This first check requests the model catalogue; it does not ask a model to generate an answer.

Before you start

You need a terminal with curl installed. Get these connection details from the person who manages your Odin deployment:

  • Your configured Gateway API base URL, including /v1 and any deployment path prefix.
  • A personal Gateway key enabled for that deployment.

This is a Gateway key, rather than an upstream provider’s API key. Access, routing policies and budgets are configured for your environment.

Keep your personal key private. Use your tool’s secret or credential settings, and keep the key out of browser client code and shared logs. For this terminal check, supply it in your private environment; the example below shows a placeholder.

Request the model catalogue

Replace both placeholders with your supplied connection details. Keep /v1 in the API base URL; the command appends only /models.

export ODIN_GATEWAY_API_BASE='https://gateway.example.invalid/your-prefix/v1'
export ODIN_GATEWAY_KEY='YOUR_PERSONAL_GATEWAY_KEY'

curl --include --silent --show-error --max-time 20 \
  "${ODIN_GATEWAY_API_BASE%/}/models" \
  --header "Authorization: Bearer ${ODIN_GATEWAY_KEY}"

The placeholder host above is not a service endpoint. This command is for your configured deployment; it does not create access. It waits at most 20 seconds and prints response headers and body. Check the HTTP status in the output, rather than treating command completion as success.

A successful response is a JSON list with object: "list" and a data array of model entries. The catalogue can be empty. Receiving a list confirms this request reached the catalogue endpoint; it does not prove that every listed model is permitted for your key, that generation will succeed, or where a later request will be processed. Model requests still depend on the configured routing, access, budget and residency controls.

If the check is refused

  • 401: the key is missing, malformed or invalid. Check your supplied Gateway key.
  • 403: the key lacks required permission or is restricted by policy. Ask your deployment administrator to check its configuration.
  • 429: the request-rate limit has been reached. Wait before retrying and check the configured limit.
  • 503: a required service could not complete the check. Preserve the response and ask your deployment administrator to investigate.

These are common Gateway responses, not an exhaustive list of proxy or network failures. Use the response body when diagnosing the result; do not bypass a refusal.

Connect your tool

In an OpenAI-compatible tool, use the same supplied API base URL and Gateway key. Confirm the tool’s expected URL format before appending paths. Choose a model or routing profile permitted for your environment, then review the applicable usage and budget settings before generating work.

Read the spend-control details, or discuss your Gateway setup with Odin.