Streamer-Sales
Streamer-Sales is an open-source AGPL-licensed Chinese LLM project that generates product-pitch scripts and a digital-human livestream presenter for e-commerce sales.
- GitHub stars
- 3.8k
- Last commit
- 1 yr ago
- Latest release
- v0.10.0
- Licence
- AGPL-3.0
- Self-hosted
- Yes
- Hosted version
- Available

Streamer-Sales is an open-source large language model application built to generate sales pitches for livestream and in-store product promotion. Given a product's features, it produces narration intended to highlight selling points and encourage purchase, aimed at e-commerce sellers and streamers who want AI-assisted sales copy and presentation.
The model is instruction fine-tuned from InternLM2 using xtuner, and the system integrates several components around it: LMDeploy for accelerated inference, retrieval-augmented generation so pitches can draw on up-to-date product manuals without retraining, an agent that can look up shipping information online, text-to-speech generation with emotional tone, automatic speech recognition for voice interaction, and generation of a digital-human presenter video so the pitch is not limited to text.
The stack uses a Vue, Pinia and Element Plus frontend with a FastAPI, Uvicorn and PostgreSQL backend exposing RESTful APIs with JWT authentication, and Docker Compose for one-command deployment. It is written in Python and released under the AGPL-3.0 license.
Key features
- AI-generated product sales scripts
- Digital-human presenter video generation
- Retrieval-augmented generation from product manuals
- Text-to-speech with emotional tone
- Speech-to-text voice interaction
- Docker Compose one-command deployment
Pricing: Free and open source under the AGPL-3.0 license; a hosted demo is linked from the project.

















