nc://otari

Otari: one key for every model an LLM gateway serves, with budgets, routing and usage tracking on each call. A Nodecode rewrite of mozilla-ai/hermes-provider-otari by mozilla-ai.

With it, /connect offers Otari, /models lists the models your key may use — read live from the gateway's GET /api/v1/models — and a turn on any of them goes to https://api.otari.ai/api/v1 over the OpenAI chat wire.

Get it

Tell the agent "add the otari cell", or run nodecode add nc://otari.

Set it up

Make a key at https://otari.ai. Save it with /connect, or set OTARI_API_KEY. A saved key wins over the variable.

The model id is whatever the gateway's catalog lists: a routed id such as openai/gpt-4o, a pinned provider:model selector such as anthropic:claude-sonnet-4-6, or the name of a routing policy.

A gateway other than the hosted one is the section's base_url, /api/v1 included:

An EU key sent to the default host is answered with 421 Misdirected Request: give the EU host in base_url instead.

Use it

{
  "otari": {
    // where the gateway is; the EU host or a self-hosted Otari goes here
    "base_url": "https://api.otari.ai/api/v1"
  }
}

Then pick a model the gateway serves (/models lists them, as your key sees them) and turn. The section is on unless it says "enabled": false.

Reasoning effort goes out as reasoning_effort. Otari takes minimal, low, medium, high and xhigh, so the strongest rung Nodecode offers, max, is sent as xhigh. Otari has no none: an effort off, or none picked at all, leaves the field out and the route's own default applies.

Test

From this repository's root:

.github/test-cell.sh providers/nodecode-otari

Apache-2.0, the same licence as the original. See NOTICE.