Liuying v0.9.1

Critical fixes for first-launch and auto-update issues introduced in 0.9.0.

Fixed

• Fresh installs could not finish setup: first-launch dependency installation now uses a proper offline wheel bundle and succeeds reliably.
• Auto-updated users could get stuck: the built-in AI runtime (Ollama) is now upgraded automatically when an older version is detected, so required models download and run correctly.

Also in 0.9.x

• English UI polish across the launcher, diagnostics and prompts
• Default model upgraded to Qwen3.5 9B with BGE-M3 embeddings

Upgrade: run the app and accept the update prompt, or download this package and extract it over your installation folder.

v0.7.2

Lyfira KB v0.7.2 — First Release

Lyfira is a local-first AI knowledge base: a 100% offline RAG assistant for your documents and notes. Nothing leaves your machine.

Highlights

• Full English interface (Chinese also available, switch in Settings)
• Cross-conversation long-term memory — a "second brain" that remembers across chats
• Long-conversation quality hardening: structured rolling summaries, truncated-response recovery, honesty guardrails (no fabricated citations), clean final answers with reasoning kept separate
• Vault search: locate notes by path or filename; document chats stay free of unrelated personal entries
• Ollama isolation: bundled runtime, no conflict with system-wide Ollama installs

Install

1. Download & extract the zip, run wuxing-kb.exe
2. First launch auto-downloads Ollama (~2GB) and models (~3.5GB): gemma3:4b + bge-m3
3. Requires ~16GB RAM, runs fully offline after setup