The LLM decides what your agent says and which tools it calls. The default Edesy runtime supports OpenAI (GPT-4o mini to GPT-5.x), Azure OpenAI, Gemini 2.x, Groq and Cerebras for pipeline agents, plus native audio with Gemini Live and OpenAI Realtime. Anthropic Claude is available on request only.
Two ways to run an agent#
| Mode | How it works | Providers |
|---|---|---|
| Pipeline | STT turns speech into text, the LLM replies in text, TTS speaks it | OpenAI, Azure OpenAI, Gemini, Groq, Cerebras |
| Native audio | One model listens, reasons and speaks | Gemini Live (2.0, 2.5 HD, 3.1, 3.8), OpenAI Realtime (with your own key) |
The dashboard presets map to these: Low Latency, High Quality and Budget set up a pipeline; Native Audio uses one model.
Supported providers#
| Provider | Provider ID | Default model | Status |
|---|---|---|---|
| OpenAI | openai |
gpt-4o-mini |
Default runtime |
| Azure OpenAI | openai-azure |
gpt-4.1-mini |
Default runtime |
| Gemini | gemini / gemini-2.5 |
gemini-2.0-flash / gemini-2.5-flash-lite |
Default runtime |
| Groq | groq |
openai/gpt-oss-120b |
Default runtime |
| Cerebras | cerebras |
gpt-oss-120b |
Default runtime, set through the API |
| Gemini Live | gemini-live-2.5, gemini-live-3.1 and others |
Per version | Default runtime, native audio |
| OpenAI Realtime | openai-realtime |
- | Native audio, with your own OpenAI key |
| Anthropic Claude | - | - | Available on request |
If an agent names an LLM provider the runtime does not support, the call falls back to OpenAI gpt-4o-mini and the substitution is logged.
Choosing a model#
| Goal | Start with |
|---|---|
| Natural turn-taking with the fewest parts | Gemini Live (native audio) |
| Low cost, short turns | Gemini 2.5 Flash-Lite, or Groq |
| Careful instruction following and tool use | GPT-4.1 or GPT-5.x |
| Enterprise Azure hosting | Azure OpenAI |
| Open models | Groq or Cerebras (gpt-oss-120b) |
Test the same script on two models with evals before you switch a live agent.
Set the model#
- Open the agent and go to Model.
- Pick a preset or choose the provider and model yourself.
- Save and run a test call. Check that your tools are called correctly.
{
"llmProvider": "gemini-2.5",
"llmModel": "gemini-2.5-flash-lite"
}
FAQs#
Which LLM is the default?
OpenAI gpt-4o-mini for pipeline agents, unless you pick a preset or another model.
Can I use Claude? Not on the default runtime. It is available on request.
Can I use my own LLM keys? Yes, for OpenAI, Gemini, Azure and others. Google AI Studio keys can be added as a pool of up to 10. See BYOK.
Does the LLM affect cost? Yes. See provider costs.