LoginSign Up
← Back to all sparks
T

Telnyx

COMMS
Velocity6.3

Programmable voice, messaging, and connectivity APIs for developers.

telnyx.com ↗

Telnyx is assembling a multi-vendor AI voice stack on infrastructure it owns.

voice-aimodel-integrationson-network-inferenceagent-orchestrationstt-ttstelephony
◆Current state
Telnyx's release notes read like a procurement list for voice AI: in the last month it has added TTS engines (Inworld, Rime Coda), STT engines (Soniox, Deepgram Flux, Speechmatics), and LLMs (Minimax M3, Kimi K2.6, GPT-5.4) to its AI Assistants and Inference products. The common thread is on-network processing — every model runs on Telnyx-owned infrastructure rather than being stitched across vendors.
◆Where it's heading
The model menu is now broad enough that the differentiator has shifted from 'which models' to 'how you orchestrate them.' Conversation Workflows — multi-step assistants with conditional routing and per-step model and voice overrides — signals Telnyx moving up the stack from connectivity provider to agent-building platform. Expect the integration cadence to continue while orchestration features deepen.
◆Prediction
The next moves likely extend orchestration: more workflow node types, analytics on assistant performance, or first-party tooling that ties the STT/LLM/TTS chain into a single configurable pipeline.

◆Recent moves

  1. 1d ago

    Inworld Realtime TTS-2 Now Available for Telnyx Voice AI Assistants

    Adds another TTS option to the Voice AI Assistants menu, this time emphasizing conversational direction and crosslingual output. Consistent with the broad multi-vendor sourcing strategy rather than a departure from it.

    View source ↗
  2. 4d ago

    Minimax M3 is now available on Telnyx Inference

    Brings a frontier open-weight LLM onto Telnyx Inference, reinforcing the on-network model story. One more entry in a fast-growing menu of hosted models.

    View source ↗
  3. 6d ago

    Kimi K2.6 Now Available for Telnyx AI Assistants

    A 1T-parameter MoE model with a 256K context window joins AI Assistants in the US. Notable for the agentic-coding framing, but structurally it is another model added to the catalog.

    View source ↗
  4. 14d ago

    Soniox STT Now Available for Telnyx Voice AI

    Adds a multilingual, streaming speech-to-text option for voice AI workflows. Fills out the STT side of the stack alongside the TTS and LLM additions.

    View source ↗
  5. 14d ago

    Deepgram Flux Multilingual STT Now Available for LiveKit Agents on Telnyx

    Extends STT coverage to LiveKit agents running on Telnyx, with automatic language detection and low-latency turn-taking. Signals Telnyx supporting third-party agent frameworks on its network, not just its own Assistants.

    View source ↗
  6. 14d ago

    Conversation Workflows Now Available for Telnyx AI Assistants

    ⚡ SPARK

    The first clearly first-party orchestration layer in this batch: multi-step assistant design with conditional routing and per-step model and voice overrides. This is the 'how you wire the models together' capability the model-integration spree was building toward.

    View source ↗