Repository navigation
Chatbot
aivrar edited this page Nov 18, 2026
·
1 revision
A built-in AI assistant for writing scripts, narration, prompts and lyrics, wired straight into the engines. Under each reply, one click sends the text to XTTS 🎙️, Fish 🐟, Kokoro ❤️, Stable Audio 🎵 or ACE-Step ⚡.

Shown: Qwen2.5-3B-Instruct (Q4_K_M GGUF, about 2 GB) running locally with llama.cpp.
| Backend | Runs where | Setup |
|---|---|---|
| Llama.cpp (local) | Your GPU/CPU, fully offline | Put .gguf models in the folder set by LLM_DIRECTORY in config.py, pick one, set context size and GPU layers, click Load
|
| LM Studio | Your LM Studio app | Start LM Studio's local server; LMSTUDIO_API_BASE in config.py (default http://127.0.0.1:1234/v1) |
| OpenRouter | Cloud | Put your key in OPENROUTER_API_KEY in config.py, then pick a model from the list |
Small instruct models (1–8B, Q4–Q8) are plenty for scripts and prompts, and leave VRAM free for the audio engines.
- Pick a backend and load or select a model.
- Type a request and click Send, for example:
- "Write a 60-second YouTube intro about the history of lighthouses."
- "Give me five Stable Audio prompts for a haunted-house ambience."
- "Write ACE-Step lyrics: first line a style tag for upbeat synthwave, then a verse and a chorus."
- Click an engine icon under the reply to drop the text into that panel, then press Generate there.
- The System Prompt panel on the right sets the assistant's role. Save several (e.g. Lyricist, Narration writer, SFX prompt engineer) and switch between them.
- Conversations are kept in
brain/context_history/. Save archives a conversation, Delete clears it, and saved chats can be reloaded from the list. - Inference settings (temperature, max tokens, top P/K, penalties) are under the gear.
More in chapter 11 of the manual.
LocalSoundsAPI · Latest release · PDF manual · Issues · MIT licensed
Get started
Speech
Music & sound
Production
Developers
Help