@telnyx/ai-sdk-provider weekly downloads statistics

@telnyx/ai-sdk-provider

Telnyx provider for the Vercel AI SDK, with support for chat, embeddings, speech generation, and transcription.

Requirements

• Node.js >= 22 • AI SDK 7 (ai@^7) • Zod (zod@^3.25 || ^4)

Setup

zod is a required peer dependency. This package imports zod/v4 internally for speech and transcription model schemas, so consumers must install it even if they are not using tool calling.

Set your Telnyx API key:

Usage

Chat (LLM)

Streaming

Tool Calling

Structured Output

Embeddings

Text-to-Speech (TTS)

You can also pass additional speech options:

Provider-specific options are also supported via providerOptions.telnyx:

Transcription (STT)

Note: AI SDK 7 auto-detects the audio media type via detectMediaType(). The mediaType parameter is no longer needed (and not accepted) in v7.

You can also use provider-specific options:

Custom Instance

Available Models

Chat Models

| Model ID | Best For |

| moonshotai/Kimi-K2.5 | General use, voice AI | | zai-org/GLM-5.1-FP8 | Highest intelligence open-source | | MiniMaxAI/MiniMax-M2.7 | Cost-effective, high intelligence | | Qwen/Qwen3-235B-A22B | Function calling, reasoning | | openai/gpt-4o | General purpose | | anthropic/claude-haiku-4-5 | Fast, efficient | | google/gemini-2.5-flash | Multimodal, fast |

See src/telnyx-models.ts for the full list of supported model IDs.

Embedding Models

| Model ID | Dimensions |

| thenlper/gte-large | 1024 |

Speech Models (TTS)

The modelId passed to telnyx.speech() is used as the AI SDK model identifier for metadata/logging. The actual synthesis is controlled by the voice parameter and providerOptions.telnyx.

| Voice | Provider |

| Telnyx.NaturalHD.astra | NaturalHD | | Telnyx.KokoroTTS.af_alloy | KokoroTTS |

See the Telnyx TTS docs for available voices.

Transcription Models

Examples: • distil-whisper/distil-large-v2 • openai/whisper-large-v3-turbo • deepgram/nova-3

Exports

Notes

• telnyx('model-id') is an alias for telnyx.languageModel('model-id') • telnyx.speech() and telnyx.speechModel() are equivalent • telnyx.transcription() and telnyx.transcriptionModel() are equivalent • imageModel() is not supported and throws NoSuchModelError • The default telnyx export is lazy, so importing it does not require TELNYX_API_KEY until first use

Changelog

v2.0.0

Breaking change — upgraded to AI SDK 7 (Provider V4).

• Breaking: Requires ai@^7 (previously ai@^6) • Breaking: Requires Node.js >= 22 (previously >= 18) • Migrated all provider interfaces from V3 to V4 (ProviderV3 → ProviderV4, LanguageModelV3 → LanguageModelV4, etc.) • specificationVersion changed from 'v3' to 'v4' • mediaType parameter removed from transcribe() (AI SDK 7 auto-detects via detectMediaType) • LanguageModelUsage inputTokens/outputTokens are now optional (may be undefined in streaming) • Security fixes: vitest 3→4, npm audit fixes (nanoid, postcss, vite)

v1.0.0

Initial release. AI SDK 6 / Provider V3.

License

MIT