Skip to content

Wenlan

A living personal knowledge library for the AI-native age. Agents capture what they learn, you add sources you trust, and Wenlan keeps source-cited pages current.

Get startedView on GitHub
Claude Code pluginClaude Code memoryMCP serverLocal daemon

The problem

Every new AI session starts cold.

The work happened, but the context did not survive. Decisions, fixes, and project instincts stay trapped in old chats instead of helping the next agent.

One missing handoff is enough to make the next conversation repeat the last one.

What Wenlan brings

A handoff loop for AI work.

Wenlan captures decisions, lessons, and next steps as work happens, then loads the handoff when the next agent starts.

The next conversation starts from the handoff instead of reconstructing the past.

Deliberate distillation

Wenlan turns repeated context into source-backed pages.

Run /distill when repeated captures should become a readable page. Optional model or API-key paths can add background extraction and page refresh work.

The next run starts from cited context, not transcript residue.

Knowledge pages

The work becomes reusable pages.

Cleaned decisions and lessons become durable pages instead of buried chat logs. They are organized enough for agents to use and concrete enough for humans to read.

Your work stops being transcript history and starts becoming project knowledge.

Hybrid storage

The daemon owns recall. Readable artifacts stay inspectable.

Wenlan keeps raw captures in the local daemon store for retrieval, then projects pages, handoffs, and status files you can open, diff, and move.

Agents recall from the daemon. You inspect the readable files.

Hybrid retrieval, measured

96% fewer tokens. Honest retrieval metrics.

Hybrid retrieval finds the right local context without replaying chat history.

SurfaceScopeResult
Full replayNo retrieval4,505 tokens / query
LME_OracleCE-reranked, 500 Q168 tokens / query · 93.6% R@5 · 0.883 NDCG@10
LME_SCE-reranked, N=90 deep-S168 tokens / query · 87.7% R@5 · 0.822 NDCG@10

Retrieval-only snapshots. LME_Oracle also records 0.857 MRR; LME_S records 0.815 MRR on 84 gradeable rows from the 90-question deep-S fixture. Token comparison is full replay vs retrieved context. Run the harness yourself.

FAQ

Common questions.

What is Wenlan?+
Wenlan is a living personal knowledge library for AI work. Agents capture what they learn, you add sources you trust, and the local daemon keeps source-cited pages current across chats, tools, projects, and time.
How is Wenlan different from built-in AI memory?+
Built-in memory stores what the AI decided was important. You usually cannot trace it, correct it, or use it from another tool. Wenlan keeps memory local, visible, correctable, and traceable. Readable pages, session logs, and project status are versioned in ~/.wenlan/.git/, and every distilled page cites the source memory IDs that produced it.
What retrieval quality does Wenlan reach?+
Hybrid retrieval combines vector search (BGE-Base-EN-v1.5-Q, 768-dim), FTS5, reciprocal-rank fusion, knowledge-graph context, and the local BGE reranker. LME_Oracle is 93.6% Recall@5, 0.857 MRR, and 0.883 NDCG@10 on the 500-question snapshot. LME_S is 87.7% Recall@5, 0.815 MRR, and 0.822 NDCG@10 on the stratified N=90 deep-S snapshot. The eval harness ships in the repo at crates/wenlan-core/src/eval/.
Is my data private?+
Yes. Wenlan runs on your machine and stores its database locally. No cloud sync or telemetry by default. Local memory setup works without a model or API key. On-device models or an Anthropic key are opt-in for automatic page distillation, recaps, and richer graph work.
Is Wenlan just another memory MCP?+
No. The MCP server is the connector. Wenlan also includes the local daemon, manual /distill, optional model-backed background extraction and page work, a libSQL store with DiskANN vectors, FTS5 + knowledge graph, mandatory provenance, real git versioning for memory, page, and session artifacts, and readable Markdown export paths.
What AI tools work with Wenlan?+
Claude Code has a marketplace plugin. MCP-compatible clients such as Cursor, Codex, Claude Desktop, VS Code, Gemini CLI, and others connect through Wenlan's MCP server.
Is Wenlan a replacement for Notion or Obsidian?+
No. Wenlan is not a notes app or a writing tool. It captures and refines what you learn from AI conversations. The Markdown projection under ~/.wenlan/ can be symlinked into Obsidian if you want to read it there.
How do I set it up?+
In Claude Code, run /plugin marketplace add 7xuanlu/claude-plugins, then /plugin install wenlan@7xuanlu, then /init. For other MCP clients, run npx -y wenlan setup first, then ~/.wenlan/bin/wenlan mcp add codex, cursor, claude-desktop, vscode, or gemini.
Does Wenlan work on Windows or Linux?+
Yes. The daemon builds and runs on macOS (arm64, x64), Linux (x86_64, aarch64; glibc), and Windows (x86_64). Service registration uses launchd on macOS, systemd-user on Linux, and Task Scheduler (schtasks) on Windows.
Can I keep work and personal memory separate?+
Yes. Memories, pages, and recalls belong to a space (for example, work, personal, or client-X). Set the active space per shell with WENLAN_SPACE, or declare them in ~/.wenlan/spaces.toml. The auto-detector also picks a space from the current repo or workspace.
Is Wenlan free?+
Yes. Wenlan is open-source. The local runtime, CLI, MCP server, and Claude Code plugin files in the Wenlan repo are Apache-2.0.

Open source

Open where it matters.

The local runtime, CLI, MCP server, and Claude Code plugin are Apache-2.0.

Get startedView on GitHub

Get release updates.