# Telnyx > Telnyx setup for OpenAI-compatible AI inference - 网址：https://funcoding.ai/agents/openclaw/providers/telnyx/ - 来源：OpenClaw 官方文档原文（英文），MIT 许可，同步于 2026-10-11 - 官方原文：https://docs.openclaw.ai/zh-CN/providers/telnyx --- [Telnyx AI inference](https://developers.telnyx.com/docs/inference/getting-started) provides hosted, OpenAI-compatible access to open-weight models plus proxied frontier routes. The vendor-maintained external [`@telnyx/openclaw-provider`](https://github.com/team-telnyx/openclaw-telnyx-provider) plugin supplies the runtime; OpenClaw 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. | Property | Value | | --------------- | ---------------------------------------------------------------- | | Provider id | `telnyx` | | Plugin | vendor-maintained external package (`@telnyx/openclaw-provider`) | | Auth env var | `TELNYX_API_KEY` | | Onboarding flag | `--auth-choice telnyx-api-key` | | Direct CLI flag | `--telnyx-api-key ` | | API | OpenAI-compatible (`openai-completions`) | | Base URL | `https://api.telnyx.com/v2/ai/openai` | | Default model | `telnyx/moonshotai/Kimi-K3` | ## Requirements The catalog pin for `@telnyx/openclaw-provider@0.2.0` requires OpenClaw `2026.8.1` or newer. Upgrade OpenClaw before installing the plugin: ```bash openclaw update openclaw --version ``` ## Install plugin ```bash openclaw plugins install telnyx openclaw gateway restart ``` ## Getting started **Create a Telnyx account and API key** 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. **Run onboarding** ```bash Onboarding openclaw onboard --auth-choice telnyx-api-key ``` ```bash Direct flag openclaw onboard --non-interactive --accept-risk \ --auth-choice telnyx-api-key \ --telnyx-api-key "$TELNYX_API_KEY" ``` ```bash Env only export TELNYX_API_KEY=... ``` In an interactive run, pick the **Telnyx API key** choice. For non-interactive setup, read the [security guidance](https://docs.openclaw.ai/security) first; `--accept-risk` acknowledges that setup can write credentials and daemon configuration without prompting. **Verify the live catalog** ```bash openclaw 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 OpenClaw it supports text and image input, tool calling, reasoning, a 1M-token context window, and up to 64k output tokens: ```json5 { 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: | Model ref | Input | Context | Max output | | ----------------------------------- | ----------- | ------: | ---------: | | `telnyx/moonshotai/Kimi-K3` | text, image | 1M | 64k | | `telnyx/moonshotai/Kimi-K2.6` | text, image | 262k | 8k | | `telnyx/moonshotai/Kimi-K2.5` | text, image | 256k | 8k | | `telnyx/zai-org/GLM-5.2` | text | 1M | 131k | | `telnyx/zai-org/GLM-5.1-FP8` | text | 202k | 8k | | `telnyx/MiniMaxAI/MiniMax-M3-MXFP8` | text | 1M | 8k | | `telnyx/MiniMaxAI/MiniMax-M2.7` | text | 200k | 8k | | `telnyx/Qwen/Qwen3-235B-A22B` | text | 32k | 8k | | `telnyx/openai/gpt-5.4` | text | 400k | 8k | | `telnyx/openai/gpt-5.4-mini` | text | 400k | 8k | | `telnyx/openai/gpt-5.2` | text | 400k | 8k | | `telnyx/openai/gpt-5.1` | text | 400k | 8k | | `telnyx/openai/gpt-5` | text | 400k | 8k | | `telnyx/anthropic/claude-haiku-4-5` | text | 400k | 8k | 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.
Telnyx can add, remove, or change hosted models independently of plugin releases. The plugin refreshes model ids, context limits, output limits, and pricing from the authenticated API while retaining model-specific transport policy.
## Manual config Most setups only need the API key. To pin the provider explicitly: ```json5 { 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, }, ], }, }, }, } ```
If the Gateway runs as a daemon (launchd, systemd, Docker), make sure `TELNYX_API_KEY` is available to that process. A key exported only in an interactive shell is not visible to an already-running managed service.
## Related - [Model providers](https://funcoding.ai/agents/openclaw/concepts/model-providers/)：Choosing providers, model refs, and failover behavior. - [Thinking modes](https://funcoding.ai/agents/openclaw/tools/thinking/)：Select OpenClaw reasoning effort levels. - [Models CLI](https://funcoding.ai/agents/openclaw/cli/models/)：List, inspect, and select discovered models. - [Models FAQ](https://funcoding.ai/agents/openclaw/help/faq-models/)：Auth profiles and model-selection troubleshooting.