Voice & Text Input
Contop supports two input methods for sending commands to the AI agent.
Text Input
Type commands directly in the execution bar at the bottom of the session screen. Press Send or hit Enter to submit.
The text input supports multi-line messages and preserves your input history within the session.
Voice Input
Tap the microphone button to start recording. Contop supports multiple speech-to-text providers: Gemini (default), OpenAI Whisper, and OpenRouter Whisper. You can switch providers or disable voice entirely in AI Settings.
How Voice Works
- Tap microphone - Recording begins with an aurora voice effect animation
- Speak naturally - Audio is captured at 16 kHz mono PCM and sent to your configured STT provider for transcription
- Release or tap stop - Recording ends and the transcribed text appears
- Review and send - The transcribed text is shown in the input bar for review before sending
Conversation History
The mobile app includes recent conversation context with each request, managed by token limits rather than a fixed turn count. When the conversation model on your phone detects that a command requires tool execution (not just a text response), it bundles the relevant conversation context and sends it to the server execution agent as a user_intent message.
Intent Routing
The mobile conversation model (default: Gemini 2.5 Flash) classifies your input locally:
- Text response - Handled entirely on the phone (no server round-trip)
- Tool execution - Routed to the server agent for autonomous execution
After the server completes execution, the result is polished through the mobile conversation model for a natural response.
Related: Mobile App · Agent Execution · Model Selection