nodecode-ngrok-ai-gateway

The ngrok AI Gateway: one access key for every model provider you wired up at ngrok, with failover. A Nodecode rewrite of ngrok/hermes-ngrok by ngrok.

Get it

Tell the agent "add the ngrok-ai-gateway cell", or run nodecode add nc://ngrok-ai-gateway.

Set it up

Make an access key (ng-…) at https://app.ngrok.ai/keys, then save it with /connect, or set NGROK_AI_GATEWAY_KEY. A saved key wins over the variable. Nothing else is needed: the models are whatever your gateway routes, so the picker asks the gateway itself.

Use it

/connect offers ngrok AI Gateway; pick a model from /models and the turn goes to https://gateway.ngrok.ai/v1/chat/completions, or a turn names it as ngrok/<model>.

Model ids are the gateway's own: openai:gpt-4o, anthropic:claude-sonnet-4-5 — the provider, a colon, the model. They go out exactly as written, and the gateway routes each one to the account credential configured behind it.
The model list is live: the gateway's /models answer, with your key, is what /models lists. Nothing is bundled, because the list is per-account gateway configuration — a key that routes nothing lists nothing.
A failure says what the gateway said: an unrouted model reads as the model's fault, an access policy or a rejected upstream key as forbidden rather than "sign in again", and an account out of credit as billing.

Set a gateway of your own with the section, or with NGROK_AI_GATEWAY_BASE_URL (the section wins):

{
  "ngrok-ai-gateway": {
    "base_url": "https://gateway.ngrok.ai/v1"
  }
}

Test

From this repository's root:

.github/test-cell.sh providers/nodecode-ngrok-ai-gateway

MIT licensed. The original's licence and what was left out are in NOTICE.