How to use Telnyx AI Inference with OpenClaw

The Telnyx provider is an external, Telnyx-maintained OpenClaw plugin. It uses the OpenAI-compatible Telnyx Chat Completions API, supports streaming, and can retrieve model information from the Telnyx model catalog.

Primary references:

By the end of this guide, you will have:

The provider runtime is maintained in the public team-telnyx/openclaw-telnyx-provider repository.

The current repository package manifest declares:

It also declares:

An upstream OpenClaw pull request is seeking to add Telnyx to OpenClaw’s official external-provider catalog, onboarding flow, and provider documentation.

The direct installation path documented below does not require that catalog PR to be merged.

You need:

OpenClaw currently supports Node.js 22.22.3+, 24.15+, and 25.9+, including Node 26.

References:

Install the provider on the same machine that runs the OpenClaw Gateway. OpenClaw plugins are loaded by the Gateway runtime.

References:

Sign in to the Telnyx Mission Control Portal.

Then:

Telnyx displays the full value only when the key is created. Store it in a password manager or secrets vault.

Reference: Create and manage Telnyx API keys

Do not:

Telnyx recommends environment variables, regular rotation, and separate keys for development and production.

Reference: Telnyx API authentication and security practices

The current Telnyx package requires OpenClaw 2026.7.2-beta.7 or newer.

Reference: Telnyx provider compatibility declaration

Use the official installer and pin the tested version:

The --version option accepts an npm version, distribution tag, or package specification. The --no-onboard option installs OpenClaw without immediately starting the onboarding wizard.

References:

Use the official PowerShell installer and pin the same version:

The PowerShell installer’s -Tag option accepts an npm tag, version, or package specification.

Reference: OpenClaw Windows installer options

When Node.js and npm are already installed:

For npm 12, allow the OpenClaw lifecycle scripts explicitly:

Reference: Install OpenClaw using npm

Confirm that the result is 2026.7.2-beta.7 or a newer compatible version.

Create the standard OpenClaw state, configuration, workspace, and session directories without running the complete onboarding wizard:

Baseline setup creates the initial configuration and workspace and then exits. It does not run provider selection, Gateway setup, or channel onboarding.

References:

The default configuration path is:

The default workspace path is:

Reference: OpenClaw setup and workspace paths

Install the version declared by the current Telnyx-owned repository:

Restart the Gateway after installation:

References:

If you are replacing an earlier local or test installation, use:

The --force flag is supported by the OpenClaw plugin installer.

Reference: OpenClaw plugin-install command

List the installed plugins:

Inspect the Telnyx plugin:

Verify its runtime registration:

The runtime inspection should identify the telnyx provider and show that the plugin loaded successfully.

The --runtime option loads the plugin module and reports its registered runtime capabilities.

References:

There are two supported approaches:

Run:

Paste your Telnyx API key when prompted.

The upstream provider contribution documents this authentication flow and uses telnyx as the provider ID.

References:

Set the key in the shell that starts OpenClaw:

The Telnyx provider repository documents TELNYX_API_KEY as its environment-variable authentication method.

References:

After setting the variable, restart the Gateway from the same user environment:

For a managed Gateway service, make sure the environment variable is available to the service rather than only to an interactive terminal.

Reference: OpenClaw Gateway operations

Refresh the model metadata exposed by installed providers:

List the Telnyx models:

For JSON output:

References:

The provider includes an offline model catalog and can augment that information using authenticated live discovery from Telnyx.

The exact live model catalog can change. Retrieve the current list rather than permanently hardcoding every model into your application.

Reference: Current Telnyx inference models

The current provider manifest declares Kimi K3 as its default model:

Set it as the OpenClaw default:

Confirm the result:

References:

You can select any model returned by the Telnyx provider using:

For example:

Or:

The current Telnyx model guide describes:

Reference: Telnyx available models

Run OpenClaw’s onboarding wizard and install its managed Gateway service:

The onboarding process configures the Gateway, workspace, channels, skills, and health checks.

References:

Because Telnyx authentication and the default model were configured in the earlier steps, retain the existing Telnyx model when the wizard asks about model configuration.

Restart the managed Gateway after onboarding:

Check its status:

For a more detailed service scan:

References:

First, confirm the selected model:

Then run an isolated agent turn:

Expected response:

openclaw agent exec runs an embedded agent turn using the normal OpenClaw configuration and configured provider credentials.

Reference: OpenClaw agent exec command

For a more general test:

Launch the dashboard:

The command opens the OpenClaw dashboard using the active Gateway configuration.

References:

Send a test message such as:

The plugin registers telnyx as an OpenClaw model-provider ID.

A complete model reference has the following form:

It consists of:

Reference: OpenClaw model-provider reference format

The plugin manifest configures the following OpenAI-compatible base URL:

Telnyx documents its Inference API as OpenAI-compatible, allowing integrations to use OpenAI SDK and Chat Completions conventions with a different base URL and API key.

References:

The provider supports streaming through OpenClaw’s OpenAI-compatible completions transport.

References:

Check the installed OpenClaw version:

The provider currently requires:

Reference: Telnyx provider peerDependencies

Install the tested version:

Then restart the Gateway:

Inspect both its cold configuration and active runtime:

References:

Run the plugin diagnostics:

Then run the complete OpenClaw diagnostic tool:

For automated repair of supported issues:

References:

Repeat the interactive authentication flow:

Then inspect the model and authentication state:

Reference: OpenClaw models and authentication CLI

For environment-variable authentication, verify only whether the variable exists without printing its value:

Restart the Gateway after setting or changing the variable:

A 401 response normally means that the API key is missing or invalid.

Check that:

Reference: Telnyx authentication errors

Refresh and list the provider catalog:

Then restart the Gateway:

References:

Set the Telnyx model explicitly:

Then confirm the resolved default:

Reference: OpenClaw model-provider selection

Run the standard diagnostic sequence:

Look for:

References:

Inspect the currently installed provider:

Check for a newer version:

Install an explicitly reviewed version:

Restart and verify:

Reference: OpenClaw plugin update commands

For production environments, pin an explicitly reviewed provider version rather than automatically installing an unknown future version.

Before removing Telnyx, configure another usable OpenClaw model provider.

Preview the removal:

Remove the plugin:

Restart the Gateway:

Confirm the remaining model configuration:

Reference: OpenClaw plugin uninstall command

After Telnyx inference is working, you can:

OpenClaw does not use Hermes’s main-model and auxiliary-slot architecture.

Instead, OpenClaw supports a primary model, per-agent model configuration, and a separate model override for sub-agents. This lets you retain a higher-capability Telnyx model for the primary agent while assigning a faster or lower-cost model to repetitive delegated work.

OpenClaw’s voice-call plugin already supports Telnyx for inbound and outbound calls, including realtime voice capabilities and transcription-related call workflows.

The public Telnyx OpenClaw STT and TTS repositories should currently be treated as preview integrations rather than established production next steps: t