Agent: AI Assistant Service

agent is the core Go service that provides the Agent Web UI, HTTP APIs, and optional device-side voice interaction. It orchestrates model calls, tool execution (HID control, screenshots, audio playback), and conversation sessions.

Default Parameters

Parameter	Default Value
Config	/userdata/agent/agent.toml
Log	/userdata/agent/log/agent.log
Binary	/usr/lib/aiden/agent
Session data	/userdata/agent/

Startup

systemctl start aiden-agent.service
systemctl status aiden-agent.service
systemctl restart aiden-agent.service

Input Modes

The HTTP server and Web UI run in every input mode:

• input_mode = "stt": Runs the device audio loop in parallel with the HTTP server, using audio_service, VAD, the selected STT provider, the model, and the selected TTS provider.
• input_mode = "realtime": Runs the configured realtime voice model directly; realtime activation is selected explicitly by this mode.

Configuration

The service reads /userdata/agent/agent.toml. Provider tables define named provider instances, and the corresponding selection table references one of those names:

[voice_settings.mode]
input_mode = "stt"

[model_settings.providers.openai-main]
type = "openai"
api_key = "$OPENAI_API_KEY"
# base_url = "https://api.openai.com/v1"

[model_settings.model]
provider = "openai-main"
model = "gpt-5.5"
# temperature = 0.2
# max_response_tokens = 8192

[voice_settings.classic.audio]
# socket = "/run/audio_service/audio_service.sock"
# backend = "auto"

[advanced_settings.hardware.hid]
# keyboard_device = "/dev/hidg0"
# mouse_device = "/dev/hidg1"
# android_keyboard_device = "/dev/hidg2"
# frame_socket = "/run/frame_service/frame_service.sock"

[voice_settings.classic.stt.providers.openai-main]
type = "openai-whisper"
api_key = "$OPENAI_API_KEY"
model = "whisper-1"

[voice_settings.classic.stt]
provider = "openai-main"

[voice_settings.classic.tts.providers.minimax-main]
type = "minimax-cn"
api_key = "$MINIMAX_API_KEY"
voice_id = "male-qn-qingse"

[voice_settings.classic.tts]
provider = "minimax-main"
speed = 1.0

Built-in model provider types include openai, anthropic, openrouter, kimi, kimi-cn, volcengine, deepseek, gemini, and ollama. Anthropic Claude can be accessed directly through the native anthropic provider or through a compatible provider such as OpenRouter. Google Gemini models use the native gemini provider, which speaks the Interactions API.