Skip to main content
Keep your agent on Deepgram and route only its language model through the SLNG context router. You get smart routing, context optimisation, in-region execution, and compliant caching on every turn. Speech to text and text to speech stay on Deepgram.

​
Prerequisites

​
Set up in the dashboard

1

Choose Deepgram

Start a new project, choose Improve my agent, and under On a voice platform pick Deepgram.
New project chooser with the Deepgram voice platform highlighted

Pick Deepgram on the voice platform path.

2

Pick your model

Open the Think section and choose your LLM model. Keep Context Router on so the model routes through SLNG. Speech to text and text to speech stay on Deepgram.
Think section with the LLM model card highlighted, routed through SLNG

Choose your model in the Think section.

​
In your Deepgram dashboard

Click Integrate in the top right. The modal fills in your router URL and model, then walks you through the steps below in Deepgram.
Deepgram Voice Agent LLM tab with the custom model name, URL, API format, and headers fields

The LLM settings in the Deepgram Voice Agent.

1

Open the LLM tab

Open your Deepgram Voice Agent and go to the LLM tab.
2

Select a custom model

In the Model dropdown, select Other - Custom model.
3

Set the custom model name

In Custom Model Name, enter the model you picked in the Think section, for example slng/auto.
4

Set the custom model URL

In Custom Model URL, paste https://au.context-router.slng.ai/v1/chat/completions. Swap au for the region closest to your callers.
5

Set the API format

In Custom Model API Format, select OpenAI.
6

Add the authorization header

Under Custom Model Headers, add a header Authorization : Bearer followed by your actual SLNG API key value.
7

Save and test

Save and run a test conversation to confirm the LLM is routing through SLNG.
The router also requires X-Slng-Agent-Id and X-Slng-Session-Id headers that identify the agent and call. Deepgram’s custom model headers take fixed values, so it cannot map a per-call session variable the way ElevenLabs and Vapi do. See Integrating the context router.

​
Next steps