Turn your notes, files, and AI chats into a living wiki.

Wenlan pages cite their sources, get checked against them, and update when they change. Claude, Codex, and more can put them to work.

Download GitHub

Free
Open source
Local-first

Notes, files, and saved conversations become an editable launch plan with links to its sources.

It tells you what it can't back up.

When your AI tool saves a page, Wenlan compares each cited sentence with its source. If most citations don't match, the page isn't saved and the AI is told why, so it can fix them.

An AI tool session in four steps: an AI tool writes a page; Wenlan refuses the write with the error "write_page refused, nothing was written: too few cited sentences match their source"; the AI fixes its citations so each sentence cites the memory that says it; the page is saved. — Wenlan's real error when a write is refused · sample data

Your knowledge grows and connects.

From personal notes to research and projects, explore the connections across your knowledge.

The same annotated Wenlan graph used in the README: personal context, Wenlan development, MCP and AI integration, AI memory systems, and illustrative regions for urban history, genetics and evolution, and human–AI collaboration research. Memories, topics and wiki pages are labeled in the image.

Wiki page
Memory
Files · PDF / Markdown
Web links · URL

Know what needs an update.

Change a source, and only the pages built on it get flagged. Wenlan can update pages it wrote; pages you wrote ask you first.

A source-file diff replaces test-only completion with desktop and mobile UI checks. The V1/V2 records show the revised completion criteria, and Steep updates an untouched AI-organized knowledge page automatically while leaving a user-edited page for review. Edited workflow illustration based on App material.

Put your knowledge to work.

Find context through keywords, meaning, and knowledge connections. Let your connected AI use that evidence to answer, write, and act.

WenlanSearch & answer

Keep my reading position steady when the sidebar opens.

››

Illustrative Claude interface, answers, and interactive outcomes, not a measured test. Wenlan sources come from a real demo library; follow citations to inspect them.

See examples

Illustrative workflow · not live product output

You ask AI

This request failed. How should we retry it?

AI consults Wenlan

API retry policyCompiled from API docs, decision records and a runbook.ExpandCollapse

Three records, one retry policy.

Read requests: Retry GET up to 3 times⁠^[1].
Write requests: Never retry POST automatically⁠^[2].
Still open: Timeout duration⁠^[3].

Next update: add the timeout once a decision is recorded.

Read requests: Retry GET up to 3 times⁠^[1].
Write requests: Never retry POST automatically⁠^[2].

AI answers from the wiki

Retry failed GET requests up to 3 times; never retry POST automatically⁠^[1][2]. The timeout boundary is still unspecified⁠^[3].

View sources(3)

We already confirmed that reads and writes need different retry rules.

[1]api-v1.mdAPI retry rule

Supplied excerptGET /reports: after a failed first attempt, allow up to 3 retries. POST /payments: do not retry automatically.

[2]decision-07.mdClient decision

Supplied excerptUse the API's GET retry cap of 3 to limit repeated requests. Automatic POST retries are disabled because duplicate payments are unsafe.

[3]runbook-v1.mdLogging and timeouts

Supplied excerptRecord the endpoint, attempt number, and final failure in the local test log. The request timeout has not been decided.

You ask AI

Based on past interviews, what should we improve in our AI assistant next?

AI consults Wenlan

AI assistant product researchExample note: past interviews and a product discussion.ExpandCollapse

Turn interview findings and a prior discussion into test directions for the next release.

Interview findings: Review before sending; trace changes afterward⁠^[1][2].
Prior discussion: Test a send preview and change history⁠^[3].
To validate: Will this make users more willing to delegate⁠^[3]?

Next update: add test results, then decide whether to adopt.

Interview findings: Review before sending; trace changes afterward⁠^[1][2].
Prior discussion: Test a send preview and change history⁠^[3].

AI answers from the wiki

Interviewees want to review emails before sending and trace file changes⁠^[1][2]. Build on the prior discussion: test a send preview and change history; the benefit is still untested⁠^[3].

View example records(3)

One interview asks to review emails before sending; another asks to see file changes. The design discussion proposes testing both controls.

[1]interview-01.mdAI assistant interview

Example recordI let the AI draft emails, but I want to see the recipient and message before it sends anything.

[2]interview-02.mdAI assistant interview

Example recordAfter the AI organizes my files, I cannot tell which files it changed, so I check them myself.

[3]product-review.mdProduct review

Example recordTest a recipient and message preview before sending, plus a history of file changes. Whether users will delegate more remains untested.

You ask AI

The client asked about data export again. Did we promise it?

AI consults Wenlan

Client delivery scopeCompiled from project conversations, scope docs and decisions.ExpandCollapse

Three records, one clear delivery scope.

Phase one includes: Sign-in and payments⁠^[1].
Outside phase one: Data export⁠^[1].
To confirm: Export approval, added scope and delivery date⁠^[2][3].

Next update: record the export scope and date if it is approved.

Phase one includes: Sign-in and payments⁠^[1].
Outside phase one: Data export⁠^[1].

AI answers from the wiki

Phase one covers sign-in and payments; export is not committed⁠^[1][2]. The project owner must approve new scope before setting a delivery date⁠^[3].

View sources(3)

Phase one covers sign-in and payments; data export is a separate discussion.

[1]scope-v2.mdScope boundary

Supplied excerptPhase one includes sign-in and payments; data export is outside this scope.

[2]meeting-notes.mdMeeting minutes

Supplied excerptThe client asked whether data export could be added; delivery is not confirmed.

[3]decision-log.mdDecision log

Supplied excerptNew scope needs the project owner’s approval; delivery timing for the added item is not decided.

You ask AI

Does remote work improve productivity? What do my readings say?

AI consults Wenlan

Remote-work researchCompiled from two studies and a reading log.ExpandCollapse

Two studies and a reading log, organized for comparison.

Call-centre trial: Higher output when working from home⁠^[1].
Hybrid-work trial: No meaningful difference in performance reviews⁠^[2].
Compare by: Job type, work arrangement and performance measure⁠^[3].

Next update: add studies from other industries and compare their settings.

Call-centre trial: Higher output when working from home⁠^[1].
Hybrid-work trial: No meaningful difference in performance reviews⁠^[2].

AI answers from the wiki

Not necessarily: compare job types and work arrangements⁠^[1][2][3]. The call-centre trial found higher output at home; the hybrid-work trial found no meaningful performance change⁠^[1][2].

View sources(3)

Two studies and a reading log, brought together for a report.

[1]Call-centre study (2015)Bloom et al. · QJE

Study summary (paraphrased)The Ctrip experiment found a 13% increase in performance among call-centre employees assigned to work from home, compared with office staff.Read the study

[2]Hybrid-work study (2024)Bloom, Han & Liang · Nature

Study summary (paraphrased)At Trip.com, employees could work from home two days a week. The trial found no effect on performance-review grades.Read the study

[3]Reading log (example)Comparison checklist

Illustrative reading logRecord the job type, days working from home and how performance was measured. These studies examine different settings; do not treat their findings as interchangeable.

You ask AI

What style should I keep for this newsletter?

AI consults Wenlan

Newsletter editorial guidanceIllustrative editorial feedback for a sample newsletter.ExpandCollapse

Three example records, one clear writing direction.

Style: Plain language; no hype⁠^[1].
Example: Include one practical example⁠^[2].
Open: The next topic is undecided⁠^[3].

Next update: add the topic once it is chosen.

Style: Plain language; no hype⁠^[1].
Example: Include one practical example⁠^[2].

AI answers from the wiki

Use plain language and add one practical example⁠^[1][2]. Avoid hype; the topic is still undecided⁠^[1][3].

View example records(3)

Earlier editorial feedback set a plain-language tone and called for a practical example.

[1]editorial-decisions.mdSample style decision

Example recordUse plain language and avoid promotional claims.

[3]draft-review.mdSample open question

Example recordThe next newsletter topic is still undecided.

You ask AI

Before tomorrow's meeting, what is still unconfirmed?

AI consults Wenlan

Prepare for the next meetingFrom meeting minutes, follow-ups and a saved AI discussion.ExpandCollapse

Three illustrative records, one budget check before the launch date.

Launch: Launch moved to October⁠^[1].
Budget: Quote received; budget not approved⁠^[3].
Next step: Confirm budget, then settle the launch date⁠^[2].

Next update: record the date after the budget is confirmed.

Launch: Launch moved to October⁠^[1].
Budget: Quote received; budget not approved⁠^[3].

AI answers from the wiki

Confirm the hosting budget first, then set the October launch date⁠^[1][2]. The quote arrived, but the budget is still unapproved⁠^[3].

View example records(3)

Meeting minutes moved launch to October; a later follow-up says the hosting quote arrived, but the budget remains unapproved.

[1]meeting-notes.mdMeeting minutes

Example recordThe launch moved to October; hosting budget confirmation is still pending.

[2]ai-discussion.mdSaved meeting prep

Example recordFor the next meeting, confirm the hosting budget first, then choose the launch date.

[3]follow-up.mdBudget follow-up

Example recordWe received the hosting quote, but the budget is still unapproved.

Let what you learn carry forward.

Keep your sources, enrich their context, and use the knowledge in your everyday tools.

Your AI tool writes the pages

Your AI tool turns notes, files, and chats into cited pages as it works, and /wenlan steep finishes the rest. No extra model needed. Prefer the background? Automatic uses a model you pick.

Your own writing stays yours

Pages you wrote or edited are set to Ask me first: AI changes wait for your OK. Pages AI wrote keep updating on their own.

Launch plan

You wrote this page · Ask me first

Current
Launch on November 14.
AI suggests
Launch on November 14, after the payment test passes.

Your code gets a wiki too

Run /wenlan wiki in a repository. Your AI tool writes pages with diagrams, each tied to the code lines it relies on. When that code changes, the pages are flagged and your AI tool updates them, daily on GitHub if you like.

App
Daemon
Database

Tied to these lines

src/main.rs:12–48
crates/server/lib.rs:30–75

Bring what you already have

Notes apps, files, AI chats, links, and code. Wenlan keeps each source with its link, so pages can cite it.

Saved by your AI tool

Gmail
Notion
Google Drive
Slack
GitHub

Notes apps

Obsidian
Logseq
OpenWiki (OKF)

AI chats

ChatGPT (export)
Claude (export)
Claude Code
Codex

Exports, plus what your AI tool saves as memories while you work

Files

PDF (text-based)
Word
Markdown
Text
Folders

Code

Repositories

Import guide

Keep using your Obsidian vault

Connect your vault and Wenlan follows changes in the background. It never edits your existing notes without your OK.

My Obsidian vault

Reading & research

Attention and learning.md
Reading notes.md

Edits notes only with your OK

Search and cite in Wenlan

Connection guide

Switch tools. Keep your knowledge.

Claude Code, Codex, Cursor, and other AI tools use the same pages and memories through plugins or MCP. Change tools anytime; your knowledge stays.

Claude Code
Codex
Cursor
Claude Desktop
VS Code
Antigravity

Connection guide

Web Access

Let supported web AI tools search your Wenlan knowledge on your computer, and write to it if you allow. Keep Wenlan online while you work.

Web AI

ChatGPTClaude

Your computer

Set up Web Access

A space for each project

Keep projects, research, and personal knowledge in separate Spaces. Choose the context your AI retrieves.

In this SpaceReading layout decisions

A home for each topic

Gather related pages. Follow connections across topics.

Origin of Species

See how your knowledge changed

Revisit page revisions and earlier memories.

Delivery checklist

v2 · Earlier
Pass the tests.
v3 · Current
Pass the tests, then check desktop and mobile views.

App captures & capability illustrations · demo content

Your files stay on your computer.

Your pages, memories, and graph are stored on your computer. Pages are readable Markdown you can open in your own editor. You decide which AI tools can read them.

How local storage works

Putting retrieval to the test.

Two separate tests of Wenlan: context size and retrieval coverage.

Evaluation & methodology

Context per query

LoCoMo · historical, not rerun

Send the whole conversation

4,505 tokens / query

Find the relevant passages

168 tokens / query

Tokens measure text length.

Retrieval coverage

LongMemEval (oracle) · 500 questions

93.6%

Average coverage of labeled relevant material in the top 5.

Not answer accuracy.

Try Wenlan with your own notes.

Add a page or file, then connect the AI tool you already use.

Download

Turn your notes, files, and AI chats into a living wiki.

Scattered notes. Connected knowledge.

Agent Loop

Find the evidence

Keep a decision that can be reused

Find the evidence

Keep a decision that can be reused

How the issue surfaced

Keep a decision that can be reused

No need to rethink it next time

It tells you what it can't back up.

Your knowledge grows and connects.

Know what needs an update.

Put your knowledge to work.

Keep my reading position steady when the sidebar opens.

Keyword+

Semantic+

Graph+

The client asked about data export again. Did we promise it?

Let what you learn carry forward.

Your AI tool writes the pages

Your own writing stays yours

Launch plan

Your code gets a wiki too

Bring what you already have

Links

Saved by your AI tool

Notes apps

AI chats

Files

Code

Keep using your Obsidian vault

Switch tools. Keep your knowledge.

Web Access

A space for each project

A home for each topic

See how your knowledge changed

Delivery checklist

Your files stay on your computer.

Putting retrieval to the test.

Context per query

Retrieval coverage

Try Wenlan with your own notes.