Agent Personas¶
Omni supports multiple AI personas — each with a unique system instruction, voice, model, and tool set. Each persona is wrapped as an AgentTool on the root agent, running via Runner.run_async() with the generateContent API (not the Live API).
Built-in Personas¶
| Persona | Description | Capabilities | Model |
|---|---|---|---|
| Claire (assistant) | General-purpose assistant | Search, tasks, image gen | gemini-2.5-flash |
| Dev (coder) | Software development expert | Code execution, desktop, search | gemini-2.5-flash |
| Nova (analyst) | Data analysis specialist | Code execution, search | gemini-2.5-flash |
| Sage (researcher) | Deep research assistant | Search, tasks | gemini-2.5-flash |
| Muse (creative) | Creative collaborator | Image generation, media | gemini-2.5-flash |
| Pixel (genui) | Visual data renderer | Code execution, GenUI schema | gemini-2.5-flash-lite |
Creating a Custom Persona¶
Via REST API¶
POST /personas
{
"name": "My Persona",
"system_instruction": "You are a helpful marketing assistant...",
"voice": "Kore",
"model_override": null,
"capabilities": ["search", "task"]
}
Via Dashboard¶
- Open the Persona panel in the sidebar
- Click + New Persona
- Fill in name, instruction, voice, and capabilities
- Click Save
Capability Tags¶
Each persona declares which tool categories it can use:
| Tag | Tools Included |
|---|---|
search | Google Search |
code_execution | E2B code interpreter |
desktop | E2B virtual desktop |
image_gen | Imagen image generation |
task | Planned tasks, scheduling |
cross_client | Cross-device actions |
genui | GenUI schema lookup |
wildcard | MCP plugins |