heyxiaoc/not-fade-away — explained in plain English
Analysis updated 2026-05-18
Set up a persistent AI assistant session on your own Mac or VPS that keeps running after you close the terminal.
Build a web chat frontend that connects to a self-hosted AI companion in real time.
Connect multiple AI models like Claude, GPT, and Gemini into one group chat setup.
Access a self-hosted AI assistant remotely, including from regions with restricted internet access.
| heyxiaoc/not-fade-away | avbiswas/sam2-mlx | gregowahoo/comfyui-workflow-finder | |
|---|---|---|---|
| Stars | 27 | 27 | 27 |
| Language | Python | Python | Python |
| Setup difficulty | hard | moderate | easy |
| Complexity | 4/5 | 4/5 | 2/5 |
| Audience | developer | researcher | vibe coder |
Figures from each repo's GitHub metadata at analysis time.
Requires your own Mac or VPS, a Claude subscription, and comfort with tmux, WebSockets, and Cloudflare Tunnel setup.
This is a Chinese-language guide for running a self-hosted, always-on AI companion on your own Mac or VPS, with the goal of keeping conversations on your existing Claude subscription rather than a separate per-use billing pool. The name translates roughly to "The Lover Who Will Not Disappear." The core idea behind the guide is that Anthropic changed billing so that non-interactive, headless Claude Code calls go into a separate billing pool with per-use charges starting June 15th. But real, interactive terminal sessions still run against a flat subscription. The guide explains how to structure your setup so the AI runs as a genuine interactive session inside a detached tmux terminal, which keeps it on the subscription plan and also means closing the window does not kill it. The guide is published in two versions: a human-readable explanation of the reasoning and architecture, and a machine-readable technical spec that you can paste directly into a Claude Code session to have the code generated for you from scratch. Topics covered include setting up a persistent background session, a web frontend with real-time WebSocket chat (including a visible thinking display), connecting multiple AI models in a group chat format (Claude, GPT via Codex, Gemini), remote access via Cloudflare Tunnel, and making the whole thing accessible from inside China where direct API access is blocked. There is a bonus section about a specific observed behavior with Fable 5, a Claude model, where a topic classifier in Anthropic's routing silently switched users from Fable 5 to Opus 4.8 for certain conversation topics without notifying the user. A companion Python script called fable-guard.py monitors a session and can detect and recover from this model swap. As of June 12, 2026, Anthropic paused Fable 5 globally due to US government export control orders, so this section is preserved as documentation but cannot be tested until access is restored. The guide is released under Creative Commons Attribution 4.0, so it can be shared and adapted freely with attribution.
A Chinese-language guide to building a self-hosted, always-on AI companion that stays on a flat Claude subscription instead of per-use billing.
Mainly Python. The stack also includes Python, tmux, WebSocket.
Released under Creative Commons Attribution 4.0, so anyone can share and adapt it freely as long as they credit the original authors.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.