---
license: apache-2.0
base_model: Qwen/Qwen3-4B
library_name: transformers
pipeline_tag: text-generation
tags:
  - conversational
  - agent
  - qwen
  - gguf
  - eightly-os
  - nova
language:
  - en
---

# Nova Talker 4B

**Nova Talker 4B** is an optional conversational voice for **Nova**, the local AI assistant in [Eight.ly OS](https://eight.ly). It is a **talker**: it turns the results of a tool call into a clear, concise reply for the user. A Qwen3-4B model, it gives richer answers than the default 2B voice while still fitting an ~8 GB GPU.

## Role in Nova

Nova is **router-in-front**. The user only ever chooses a **talker** — this model is one of those choices. Tool selection is always handled by the hidden **[Nova Router 1.5B](https://huggingface.co/smashingtags/nova-router-1.5b)** (99%), which installs automatically alongside any talker. This model never picks tools; it writes the reply.

## Files

| File | Quant | Size |
|------|-------|------|
| `eightly-agent-q4b-Q4_K_M.gguf` | Q4_K_M | ~2.4 GB |

## Usage

```bash
ollama pull hf.co/smashingtags/nova-talker-4b:Q4_K_M
```

Within Eight.ly OS, selecting this talker in the Nova model catalog installs it and pulls the router + retrieval embedder automatically.

## License

Apache-2.0.