# Advanced Topics

#### 🎙️ Voice Integration (Coming Soon)

Oxyde is built to support real-time **speech-to-text** and **text-to-speech** pipelines:

* Plug in services like:
  * 🗣️ OpenAI Whisper
  * 🧠 Deepgram, AssemblyAI
  * 🗣️ ElevenLabs (TTS)
* Route audio input → transcript → LLM → audio output

Future agent config will include:

```json
"voice_profile": {
  "tts": "elevenlabs",
  "stt": "whisper"
}
```

***

#### 🌐 Streaming Output

You can enable real-time LLM streaming to get **token-by-token output**, ideal for:

* NPCs talking while typing
* Dynamic cutscenes
* Interruptible generation

The async pipeline in `llm_service.rs` supports streaming via:

* `reqwest` and `futures`
* WebSocket-compatible forwarders (WIP)

***

#### 🗃️ Persistent Storage (Memory DB)

Persistent long-term memory is being designed to store:

* ⌛ Session data
* 🧠 High-priority long-term memories
* 🔄 Recalled insights across playthroughs

Pluggable backends:

* 🪵 JSON (dev mode)
* 📄 SQLite (lightweight)
* 🛢️ Postgres (production)

Future config block:

```json
"storage": {
  "type": "sqlite",
  "path": "./data/velma.db"
}
```

***

#### 🖥️ CLI Deployment

Use Oxyde to power:

* NPC simulation servers
* Discord agents
* Networked agent brains via socket
* Inference daemons behind REST APIs

`main.rs` and `cli.rs` already support:

```bash
cargo run --example rpg_demo
```

Soon: `oxyde serve` for headless server deployment.

***

#### 🧩 Feature Flags

For lighter builds, Oxyde supports compile-time toggles:

| Feature       | Flag                |
| ------------- | ------------------- |
| LLM Streaming | `--features stream` |
| TTS/STT       | `--features voice`  |
| SQLite        | `--features db`     |

***

#### 🔗 Related Pages

* 8\. API Reference
* 9\. Examples & Demos