Skip to main content
Bring your own key (BYOK) runs a model on your own provider account instead of an SLNG-hosted one. The request still flows through the SLNG Execution Layer, so routing, fallbacks, and caching work the same way; only the model runs on your account. You register the provider key once, and it is available to every project in your organization.

​
When to use your own key

Prefer an SLNG-hosted model where one exists: it runs in your region and skips the hop to an outside provider, which is the lowest-latency path. Reach for your own key when one of these applies:
  • Provider relationship. You already have contracts, pricing, or quota with the provider that you want to keep using.
  • Coverage. A model or voice you need is not SLNG-hosted in your region.
  • Procurement or compliance. The provider account must stay in your name.
A BYOK route leaves SLNG for your provider, so its latency depends on that provider and its region, not on your SLNG region.
Need a model or provider that is not an SLNG model or available through BYOK? Email support@slng.ai and we will look at adding it.

​
Register a key

You register a provider key once, scoped to your organization, so every project can use it. For speech, BYOK covers the providers the vault lists; a model from any other provider runs SLNG-hosted. For LLMs you register a custom model of your own, so you are not limited to a list. See Add a custom LLM model.
1

Open the BYOK vault

In your project, go to Project Vault, then BYOK. The page splits into Speech models and LLM providers. Click Manage in SLNG Vault.
The BYOK page in Project Vault, with the Manage in SLNG Vault button highlighted

Project Vault, BYOK. Open Manage in SLNG Vault.

2

Add a key in the organization vault

The organization vault opens on SLNG Vault → BYOK, with the same Speech models and LLM providers tabs. Click Add key.
The organization BYOK vault with the Add key button highlighted

The organization BYOK vault. Click Add key.

3

Choose a model and paste your key

Set the Type (speech to text, text to speech, or LLM), then pick the Model. The API key field appears once a model is selected. Paste the key from your provider’s dashboard, leave Apply to on Organisation (default) so every project can use it, and click Add key. The key is stored encrypted and used automatically whenever that model is selected.
The Add a speech model key dialog with a model selected and the API key field ready for a key

After you pick a model, the API key field appears. Paste your provider key.

4

Select it on your agent

The key now shows up on the BYOK page and in the BYOK tab of each model dropdown on Listen, Think, and Speak. Pick that model and the agent uses your key.

​
Add a custom LLM model

The LLM providers tab is open-ended: you register any provider endpoint rather than picking from a list. The model runs on your provider account, and the Context Router routes to it.
1

Open the LLM providers tab

In the BYOK vault, open the LLM providers tab and click Add key.
2

Fill in the model details

In Add an LLM provider, enter:
  • Model: the model name your provider expects, such as gpt-4o.
  • API key: your provider key, from the provider’s dashboard.
  • Endpoint URL: the provider’s base URL.
  • Provider (optional): leave it on Auto-detect, or pick the provider.
Click Add key. The key is stored encrypted, and you can update it anytime.
The Add an LLM provider dialog with Model, API key, Endpoint URL, and Provider fields

The Add an LLM provider dialog: model, API key, endpoint URL, and an optional provider.

3

Select it on your agent

The model appears in the BYOK tab of the Think section’s LLM dropdown. Pick it and the agent routes to your model.
For the full set of provider options (Azure, Vertex, and custom auth headers), see Create a client model. For how the Context Router uses your saved keys, see Saved in the dashboard.

​
Use a BYOK model in code

Model IDs name where the model runs. A BYOK model uses the bare provider route, such as deepgram/nova:3, with your credential attached. For every route, see Where a model can run. Supply a BYOK credential one of two ways, and never attach one to an slng/ model:
  • Registered credential (recommended). A key stored in the vault is resolved automatically. Managed agents just select the model; direct calls send only Authorization: Bearer $SLNG_API_KEY.
  • Per-request header (direct model calls). Send X-Slng-Provider-Key with X-Slng-Agent-Id and X-Slng-Session-Id, which are required. Managed agents never send this header.

​
Next steps