Skip to main content

​
How Model Configuration Works

Dograh uses a two-level configuration system for AI models:
  1. Global configuration — A single set of model settings (LLM, TTS, STT) that applies to all agents by default.
  2. Agent-level overrides — Optional per-agent settings that override the global configuration for specific services.
If no overrides are set for an agent, it uses the global configuration as-is.
Agent-level overrides are selective — you can override only the services you want to change. For example, you can override just the LLM provider for a specific agent while keeping the global TTS and STT settings. There is no need to reconfigure every service.

​
Global Configuration

The global configuration is the default model setup shared across all your agents. Dograh ships with its own models by default — when you sign up on https://app.dograh.com or set up the platform on your self-hosted infrastructure, you get some Dograh model credits to start with. To configure the global models, go to Model Configurations in your dashboard:
  • Hosted: https://app.dograh.com/model-configurations
  • Self-hosted: http://localhost:3010/model-configurations
Model Configuration From here you can configure each service:
ServiceWhat it does
LLMThe language model that generates responses (e.g., OpenAI GPT-4.1, Anthropic Claude)
TTS (Voice)The text-to-speech model that converts responses to spoken audio (e.g., ElevenLabs, Cartesia)
STT (Transcriber)The speech-to-text model that transcribes user speech (e.g., Deepgram, AssemblyAI)
RealtimeA single speech-to-speech model that handles LLM, TTS, and STT in one (e.g., Gemini Live)
Select a provider from the dropdown and configure the API key, model, and any provider-specific settings. For Dograh’s own models, see Service Keys for instructions on creating Service Keys.

​
Agent-Level Model Overrides

You can override the global model configuration for any individual agent. This is useful when different agents have different requirements — for example, a customer support agent might use a faster, cheaper LLM while a sales agent uses a more capable one.

​
Configuring overrides

  1. Open the agent you want to customize.
  2. Go to Settings in the agent detail page.
  3. Select the Model Overrides tab.
  4. You will see tabs for each service: LLM, Voice (TTS), and Transcriber (STT).
  5. Toggle Override on for the service you want to change.
  6. Configure the provider, model, and other settings as needed.
  7. Save your changes.

​
Selective overrides

Each service can be toggled independently. When an override is off for a service, the agent inherits the global setting for that service. When an override is on, the agent uses the override setting instead.
LLM OverrideTTS OverrideSTT OverrideResult
OffOffOffAgent uses global config for all services
OnOffOffAgent uses custom LLM, global TTS and STT
OffOnOffAgent uses global LLM and STT, custom TTS
OnOnOnAgent uses custom config for all services
For example, if you only want to change the voice for a specific agent:
  1. Leave the LLM and Transcriber overrides off.
  2. Toggle the Voice override on.
  3. Select a different TTS provider or voice.
  4. The agent will use your custom voice while still using the global LLM and STT.

​
Realtime mode override

You can also switch an individual agent to use a Realtime provider (such as Gemini Live) even if the global configuration uses standard LLM + TTS + STT. Toggle the Realtime switch in the Model Overrides tab, then configure the realtime provider, model, and voice.
When an agent uses a Realtime provider, it replaces the separate TTS and STT services with a single speech-to-speech model. An LLM is still required alongside the Realtime model — it’s used for out-of-band tasks like variable extraction and QA analysis, which the realtime service does not handle. Context compaction is not applicable in Realtime mode and is ignored if enabled.

​
Gemini 3.1 Live

Gemini 3.1 Live is Google’s realtime multimodal API that handles both LLM and voice in a single model. Instead of configuring separate LLM, TTS, and STT services, Gemini Live acts as an all-in-one realtime provider — it processes speech input, generates a response, and speaks it back, all over a single streaming connection. Dograh supports Gemini 3.1 Live as a Realtime provider. The default model is gemini-3.1-flash-live-preview.

​
Available Voices

You can choose from the following built-in voices:
VoiceDescription
PuckDefault voice
Charon—
Kore—
Fenrir—
Aoede—

​
Getting a Gemini API Key

To use Gemini 3.1 Live with Dograh, you need a Google Gemini API key. Follow these steps:
  1. Go to Google AI Studio.
  2. Sign in with your Google account.
  3. Click on Get API Key in the left sidebar.
  4. Click Create API Key.
  5. Select an existing Google Cloud project or create a new one.
  6. Copy the generated API key and store it securely.
The Gemini API key is different from a Google Cloud service account key. You specifically need a Gemini API key from Google AI Studio for use with Dograh.

​
Configuring Gemini 3.1 Live in Dograh

  1. Go to Model Configurations in your Dograh dashboard (https://app.dograh.com/model-configurations for hosted or http://localhost:3010/model-configurations for local).
  2. Under the Realtime section, select google_realtime as the provider.
  3. Paste your Gemini API key.
  4. Select the model (gemini-3.1-flash-live-preview is available by default, or you can enter a model name manually).
  5. Choose a voice from the dropdown (default is Puck).
  6. Select the language (currently en is supported).
When using a Realtime provider like Gemini Live, you do not need to configure separate TTS and STT services — the realtime model handles speech in and out. However, you must still configure an LLM under the LLM tab: it powers variable extraction and QA analysis, which the realtime service does not perform.