Telnyx AI inference provides hosted, OpenAI-compatible access to open-weight models plus proxied frontier routes. The vendor-maintained external @telnyx/apus-provider plugin supplies the runtime; Apus carries its catalog and onboarding metadata. Authenticated discovery follows the complete model set enabled for your Telnyx account, while the offline fallback contains the models available when that plugin release was built.

Requirements

The catalog pin for @telnyx/apus-provider@0.2.0 requires Apus 2026.8.1 or newer. Upgrade Apus before installing the plugin:

apus update
apus --version

Install plugin

apus plugins install telnyx
apus gateway restart

Getting started

Create a key in the [Telnyx API keys portal](https://portal.telnyx.com/#/app/api-keys). Inference calls are usage-priced; see the [Telnyx inference docs](https://developers.telnyx.com/docs/inference/getting-started) for current models and rates.

apus onboard --auth-choice telnyx-api-key

apus onboard --non-interactive --accept-risk \
  --auth-choice telnyx-api-key \
  --telnyx-api-key "$TELNYX_API_KEY"

export TELNYX_API_KEY=...

In an interactive run, pick the **Telnyx API key** choice. For non-interactive setup, read the [security guidance](/security) first; `--accept-risk` acknowledges that setup can write credentials and daemon configuration without prompting.

```bash
apus models list --refresh --provider telnyx
```

With usable auth, the plugin requests the authenticated `/models` endpoint and lists every model returned for the account. Without auth, it stays offline and uses the bundled fallback.

Default model

telnyx/moonshotai/Kimi-K3 is the default model. In Apus it supports text and image input, tool calling, reasoning, a 1M-token context window, and up to 64k output tokens:

{
  agents: {
    defaults: {
      model: { primary: "telnyx/moonshotai/Kimi-K3" },
    },
  },
}

Use /model telnyx/moonshotai/Kimi-K3 to switch an existing chat.

Live model discovery

The plugin fetches the authenticated /models endpoint and caches the result for five minutes, then merges those rows over the bundled catalog. Newly launched Telnyx models appear automatically without waiting for a plugin release, including proxied frontier routes such as telnyx/openai/gpt-5.x, telnyx/anthropic/claude-haiku-4-5, and telnyx/google/gemini-2.5-flash.

Bundled fallback catalog

The authenticated live catalog is authoritative. These rows keep setup and model selection useful before discovery succeeds:

All bundled models support tool calling and reasoning. The openai/* and anthropic/* ids are Telnyx-proxied frontier routes; the rest are Telnyx-hosted open-weight models. Streaming usage is reported on the final chunk.

Manual config

Most setups only need the API key. To pin the provider explicitly:

{
  env: { vars: { TELNYX_API_KEY: "..." } },
  agents: {
    defaults: {
      model: { primary: "telnyx/moonshotai/Kimi-K3" },
    },
  },
  models: {
    mode: "merge",
    providers: {
      telnyx: {
        baseUrl: "https://api.telnyx.com/v2/ai/openai",
        apiKey: "${TELNYX_API_KEY}",
        api: "openai-completions",
        models: [
          {
            id: "moonshotai/Kimi-K3",
            name: "Kimi K3",
            reasoning: true,
            input: ["text", "image"],
            contextWindow: 1000000,
            maxTokens: 64000,
          },
        ],
      },
    },
  },
}

Related

Choosing providers, model refs, and failover behavior.

Select Apus reasoning effort levels.

List, inspect, and select discovered models.

Auth profiles and model-selection troubleshooting.