nc://telnyx

Telnyx: hosted open-weight models and proxied frontier routes, on an OpenAI-compatible API. A Nodecode rewrite of team-telnyx/hermes-inference-provider by team-telnyx.

Telnyx serves inference at https://api.telnyx.com/v2/ai/openai (chat completions: streaming, tool calling, usage on the final chunk) plus a standard authenticated /models catalog that mixes models on Telnyx's own GPU infrastructure (Kimi, GLM, MiniMax) with proxied frontier routes (openai/..., anthropic/...). With this cell, /connect offers Telnyx, /models lists what your key can see, and a turn on moonshotai/Kimi-K3 (or any other catalog id) goes to that base:

• the live catalog is the core's own listing (GET /v2/ai/openai/models with your key as its Bearer); the models in this folder are the fallback the picker shows before that answers, and the ones the original ships
• no output cap is volunteered. A cap combined with tool calls trips Telnyx's error 10015 on hosted models, and no catalog row predicts which models reject it, so a Telnyx round carries max_tokens only when you configured a ceiling yourself

Get it

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

Set it up

Make a key in the Telnyx portal at https://portal.telnyx.com/#/app/api-keys, then save it with /connect, or set TELNYX_API_KEY. A saved key wins over the variable. The key is required for the catalog too: an anonymous /models request answers 401, and the picker then shows only the fallback rows.

The address is optional, for a relay that speaks the same wire:

{
  "telnyx": {
    "base_url": "https://api.telnyx.com/v2/ai/openai"
  }
}

Use it

Pick Telnyx in /models and a model from the list, then talk as usual. Models are named by their catalog id, e.g. moonshotai/Kimi-K3, zai-org/GLM-5.2 or MiniMaxAI/MiniMax-M3-MXFP8. Pricing is per 1M tokens on Telnyx's model docs.

Test

From this repository's root:

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

MIT licensed. See NOTICE.