Get started
Start with one decision in Wenlan.
Install Wenlan, connect the AI tool you use, then save and find one useful decision. You do not need to set up every client.
01
Install and open Wenlan
Download the desktop app for macOS Apple silicon or Windows x64, then follow its setup guide. Your pages and sources stay in your local Wenlan library.
Choose your downloadA new library opens on a getting-started checklist. Pick how you want to start, then follow the steps: write a page, connect an AI tool, bring in sources, and ask your AI about what you saved.
Steep turns sources into pages. On demand, the default, your connected AI tool does it as you work, with no model to set up. For Automatic, choose a local model or configure a cloud provider in Settings. Cloud processing sends the relevant content to that provider.
Prefer the command line, or use Linux?
Install the local service without the desktop app. Choose only your operating system below. On Windows, extract the complete archive into a folder on PATH and keep its libraries beside the executables.
# macOS Apple silicon
npx -y wenlan setup# Linux x64 or ARM64
curl -fsSL https://raw.githubusercontent.com/wenlan-app/wenlan/main/install.sh | bash
wenlan setup --basic
wenlan background on
wenlan status# Windows x64, after extracting the ZIP and adding it to PATH
wenlan setup --basic
wenlan background on
wenlan status02
Connect one AI tool
Choose the tool you already use. You only need one of these paths. If you prefer to read and organize in the desktop app, you can connect an AI tool later.
Claude Code plugin
Install the Wenlan plugin inside Claude Code with the first two commands below. It adds the session commands and connects Claude Code to Wenlan.
Restart Claude Code if prompted, then run /wenlan to check the connection. The first setup may need a few minutes to download the search model. Everything else also starts from /wenlan: add what you want in plain words, such as /wenlan lint.
/plugin marketplace add wenlan-app/wenlan/plugin install wenlan@7xuanlu-wenlan/wenlanCodex
Install the Wenlan runtime using the command-line option above, then run these two commands in your terminal to add the Codex plugin.
Start a new Codex task and type $wenlan to check the connection. If you only want MCP tools, wenlan connect codex is an alternative; it does not install the Wenlan slash skills.
codex plugin marketplace add wenlan-app/wenlan
codex plugin add wenlan@7xuanlu-wenlanChatGPT, Claude.ai or Muse (advanced)
In the Wenlan desktop app, open Settings > Connections > Web access and turn on the switch. If you have Spaces, choose whether to share the whole library or one Space. Then use Add to Claude or Add to ChatGPT, or for Muse and other apps, add https://relay.wenlan.app/mcp as a custom OAuth MCP connector.
Return to Wenlan and click Allow to approve the pairing request. If you share one Space, other and future Spaces remain private. Keep Wenlan online during use, and revoke access in the app when no longer needed.
Remote queries pass through the relay and your AI provider; local access activity is recorded. This is a custom connection, not a public ChatGPT Apps Directory listing.
Web access guide (English)Cursor, Claude Desktop, or another local client
First install the Wenlan runtime using the command-line option above. Then run the matching connection command and restart your AI client.
This connects the client to Wenlan through MCP. It does not add the Claude Code or Codex plugin commands.
With the desktop app, Settings > Connections > Add a tool connects detected tools with one click, including Muse Code.
wenlan connect cursor
# or: claude-desktop, vscode, gemini03
Save one decision. Find it in a new chat.
With the Claude Code or Codex plugin connected, try the fictional decision below. In Claude Code, use /wenlan:handoff and /wenlan:recall; in Codex, use /handoff and /recall.
After saving, start a new chat and ask for the decision. Check that the answer preserves “drafts only” and points back to the saved record. Saving a handoff and organizing a page are separate steps.
Local MCP clients can save and search using ordinary language. Browser connections search by default; turn on Search and write in Web Access to let them save too. To test search, first add a page or file in the desktop app, then ask your AI to find it.
1. Save this fictional decision:
/handoff Demo project: the email assistant may draft replies, but sending requires my approval.2. In a new chat:
/recall What can the demo email assistant do without my approval?