About
An LLM wiki for AI work.
Agents capture what they learn, you add sources you trust, and Wenlan keeps source-backed wiki pages current across AI work.
Project status
v0.15.8
macOS, Linux, Windows
Apache-2.0
Built by Qi-Xuan Lu
01
Why Wenlan exists
AI work has become serious work, but most sessions still end like disposable conversations. Decisions, debugging lessons, project constraints, and handoffs get buried in old chats.
Wenlan is built so the work can compound into an LLM wiki. Agents can save what matters, recall it later, and keep refined, source-backed context available across MCP-compatible tools.
02
Design principles
Local-first
Memory starts on your machine. Cloud sync, telemetry, local models, and API keys are opt-in choices rather than the default source of truth.
Human-readable
Memory, page, and session writes leave Markdown artifacts in local git. The daemon database powers retrieval, while the source-backed artifacts stay inspectable.
Session rhythm
Wenlan follows how AI work actually happens: load context, capture durable facts, write handoffs, and bring the right context into the next run.
Deliberate distillation
Between sessions, Wenlan deduplicates repeat facts and links related ideas. Run /distill when a topic deserves a source-backed page; local models or API keys can add automatic page distillation and richer graph work.
03
Built by Qi-Xuan Lu
Wenlan is built and maintained by Qi-Xuan Lu (GitHub @7xuanlu). Background in AI infrastructure, knowledge graphs, and local-first systems.
The work focuses on an LLM wiki agents can build and humans can inspect: hybrid retrieval on libSQL, real git versioning for readable pages, session handoffs and status artifacts, mandatory provenance on distilled pages, and one daemon serving multiple AI tools.
Project channels: GitHub Issues for bugs and feature requests, SECURITY.md for vulnerabilities, and the Wenlan release notes for changes.
04
Current status
Wenlan v0.15.8 ships a macOS Apple Silicon desktop preview plus native runtime artifacts for macOS, Linux (x86_64, aarch64; glibc), and Windows (x86_64). The macOS preview is not yet notarized, and Windows remains a runtime ZIP rather than a desktop app. The daemon, CLI, MCP server, Claude Code plugin, and Codex plugin are Apache-2.0; the desktop app crate is AGPL-3.0-only.