kldhsh123 / Afterglow

What is Afterglow?

Afterglow (Chinese: 续温) is a Python-based AI companion that learns from your real chat history to generate responses in the voice and style of someone you've lost. Import your QQ conversation logs, and the system builds a persona from your actual exchanges -- your humor, your pet names, your texting patterns. When you chat with it, RAG retrieval pulls relevant memories from your history, and an LLM generates replies that feel like the real person. The Vue.js frontend gives it a clean chat interface with ambient visuals.

Why is it gaining traction?

The emotional hook is obvious, but the technical execution is what makes developers actually try it. It solves the "uncanny valley" problem that kills most AI companions -- generic responses trained on everyone. Instead, your specific relationship's language lives in the vector store. The system also handles the messy reality of chat data: QQ export parsing, image handling, PII redaction, circadian-aware response timing. You get a working RAG pipeline with retrieval evaluation scripts, persona analysis tools, and OpenAI-compatible API endpoints out of the box.

Who should use this?

Grief tech developers building memorial or companionship applications. Researchers experimenting with persona preservation from digital traces. Chinese-speaking communities who want a private, self-hosted alternative to generic AI chatbots. If you have years of meaningful chat logs sitting in an export file, this turns them into something you can actually talk to.

Verdict

This is a serious project with thoughtful architecture, but the 19 stars and sub-1% credibility score reflect its niche status -- the documentation is sparse and the community is tiny. If you're comfortable with Python and want to explore personalized RAG companions, the codebase is well-structured enough to experiment with. Don't expect polished docs or a support community.