DEV.co
RAG Frameworks · PeterH0323

Streamer-Sales

Streamer-Sales is an open-source LLM-based sales presentation system that generates product pitches, voices, and digital avatars for e-commerce. It integrates RAG for product knowledge, ASR for voice input, TTS for emotional speech synthesis, and Agent capabilities for real-time queries, deployed via Docker with a Vue/FastAPI stack.

Source: GitHub — github.com/PeterH0323/Streamer-Sales
3.7k
GitHub stars
565
Forks
Python
Primary language
AGPL-3.0
License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

FieldValue
RepositoryPeterH0323/Streamer-Sales
OwnerPeterH0323
Primary languagePython
LicenseAGPL-3.0 — OSI-approved
Stars3.7k
Forks565
Open issues14
Latest releasev0.10.0 (2024-11-04)
Last updated2025-03-08
Sourcehttps://github.com/PeterH0323/Streamer-Sales

What Streamer-Sales is

Built on InternLM2-Chat (7B) with instruction tuning via xtuner, the system combines LMDeploy inference acceleration, PostgreSQL persistence, RESTful APIs, RAG-augmented generation, and ComfyUI-based digital human synthesis. Supports multiple deployment modes: Docker-Compose orchestration or native Ubuntu installation with Uvicorn/Vue frontend.

Quickstart

Get the Streamer-Sales source

Clone the repository and explore it locally.

terminalbash
git clone https://github.com/PeterH0323/Streamer-Sales.gitcd Streamer-Sales# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

E-commerce live-streaming automation

Generate contextual product narratives, emotionally-resonant voiceovers, and avatar videos for livestream hosts in real-time, reducing content creation time and scaling to multiple product listings.

Self-service product description at scale

Automatically create persuasive product pitches from minimal input (specs, images) with RAG-enhanced knowledge, enabling small retailers to compete on presentation quality without copywriters.

Multilingual retail support

Combine ASR input, LLM generation, and emotional TTS to support interactive product inquiries in multiple languages, bridging customer questions to real-time product knowledge bases.

Implementation considerations

  • GPU requirement: Minimum 6GB VRAM for 7B model inference (4-bit quantization available for 4GB). Docker-Compose setup assumes NVIDIA GPU; CPU fallback not documented.
  • RAG setup: Requires manual ingestion of product documentation into vector DB; no automated web scraping or bulk import tools provided—plan for data preparation pipeline.
  • LLM calibration: Model trained on sales/persuasion language; may generate overly aggressive pitches. Review and tune system prompts or fine-tune on brand-specific guidelines.
  • TTS emotional quality: Relies on external TTS service (voice selection, emotion tags); latency and cost depend on provider. ComfyUI digital human generation is offline but computationally expensive.
  • Agent API cost: Network query features (express tracking, pricing) require paid third-party APIs; no budget controls or fallback mechanisms documented.

When to avoid it — and what to weigh

  • Compliance-critical retail (pharma, financial products) — LLM-generated content carries hallucination risk; use only where manual review gates exist. No built-in audit trails or liability frameworks evident.
  • Offline or ultra-low-latency requirements — Requires continuous inference (GPU), PostgreSQL, and internet for Agent APIs. Not suitable for edge-only or latency-sensitive (<100ms) deployments.
  • High-volume concurrent users without scaling plan — While Docker-Compose enables multi-instance deployment, the project lacks load-testing data, caching strategies, or documented horizontal scaling limits.
  • Non-Chinese retail markets without localization — Trained on Chinese product/sales data; TTS, NER, and RAG may degrade on English or other languages without additional fine-tuning.

License & commercial use

Licensed under AGPL-3.0 (GNU Affero General Public License v3.0). This is a copyleft license requiring any derivative works or network services to open-source modifications under the same license. Commercial use of the software itself is permitted, but if you modify and deploy as a service, source code must be disclosed. Proprietary fine-tuned models or datasets trained on this codebase do not inherit the AGPL obligation, but any changes to the codebase do.

Caution: AGPL-3.0 imposes copyleft obligations on any deployed modifications. Using the unmodified codebase as-is in production is legally permitted for commercial purposes. However, if you customize the backend, integrate custom modules, or fork the project, you must open-source those changes and offer them to users. This limits proprietary customizations. If you plan to build a closed-source product, review the license with legal counsel or consider requesting a commercial license from the maintainer. Fine-tuned models derived from InternLM2 are not restricted by AGPL (model weights are separate), but the training/inference code is.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

SignalAssessment
MaintenanceActive
DocumentationAdequate
License clarityNeeds review
Deployment complexityHigh
DEV.co fitGood
Assessment confidenceMedium
Security considerations

No formal security audit evident. Key concerns: (1) AGPL license requires disclosure of modifications, limiting proprietary hardening; (2) FastAPI backend includes JWT auth but lacks rate-limiting, input validation, and injection protections (SQL, prompt); (3) PostgreSQL credentials appear in docker-compose.yaml (no secrets management docs); (4) LLM generation can be exploited for prompt injection or data exfiltration via RAG queries; (5) TTS/ASR/Agent APIs require external API keys—no encryption or vault integration documented; (6) No HTTPS enforced in examples; (7) Front-end Vue app lacks CSRF/XSS guards in README. Recommend: security review before production, secrets manager (e.g., HashiCorp Vault), API gateway with rate-limiting, and input sanitization.

Alternatives to consider

Dify / LLMOps platforms

No-code or low-code LLM orchestration (RAG, tools, TTS) with proprietary or permissive licenses. Better for non-engineers; less customizable code but faster time-to-market.

Synthesia / D-ID (digital avatar APIs)

Commercial SaaS with polished avatar/TTS. Higher cost, closed-source, but eliminates infrastructure burden and ComfyUI tuning complexity.

OpenAI GPT-4 + custom integrations

Stronger base model, fewer fine-tuning concerns, hosted inference. Trade-off: API costs, data privacy (models trained on API queries), and vendor lock-in. No local control.

Software development agency

Build on Streamer-Sales with DEV.co software developers

Fork Streamer-Sales on GitHub, deploy via Docker-Compose, and start generating persuasive product pitches—but review the AGPL-3.0 license with your legal team if you plan to customize for proprietary use.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

Related on DEV.co

Explore the category and the services that help you build with it.

Streamer-Sales FAQ

Can I use this for a closed-source commercial product?
Not without legal review or a commercial license from the maintainer. AGPL-3.0 requires you to open-source any modifications and provide source code to customers. The unmodified codebase can be used commercially, but any customizations must be disclosed.
What GPU do I need?
Minimum 6GB VRAM for the 7B full model, or 4GB with 4-bit quantization. Recommended: A100 (40GB) or RTX 4090 for multi-user / concurrent inference. CPU-only is not supported.
Can I switch to a different LLM (e.g., Llama 2, Qwen)?
Technically yes, but requires re-training with xtuner and adjusting the inference pipeline. This is non-trivial and not officially documented. The sales/persuasion fine-tuning is baked into the InternLM2 model weights.
How do I keep product data up-to-date?
Use the RAG feature: ingest product docs (PDFs, HTML) into the PostgreSQL vector DB via the provided scripts. No real-time web scraping is built-in; you manage data freshness manually or via scheduled ETL.

Software development & web development with DEV.co

From first prototype to production, DEV.co delivers software development services around tools like Streamer-Sales. Our software development agency staffs experienced software developers and web developers for custom software development, web development, integrations, and ongoing support across rag frameworks and beyond.

Ready to automate your sales narratives?

Fork Streamer-Sales on GitHub, deploy via Docker-Compose, and start generating persuasive product pitches—but review the AGPL-3.0 license with your legal team if you plan to customize for proprietary use.