Chinese | Docs | Usage | FAQ | npm

Save what your AI reads. Ask it later. Keep everything local.

AIWiki is a local Markdown knowledge base for AI assistants.

Give your AI assistant a URL, article, file, or note. The assistant reads and understands it. AIWiki turns the result into structured, traceable, reusable Markdown knowledge files.

Use it to build a local AI knowledge base that can be queried, checked, maintained, and reused later.

Ask your AI assistant to install and configure AIWiki.

1. Choose a local folder for your knowledge base.

Windows: D:\AIWiki
macOS/Linux: ~/AIWiki
Project test: ./aiwiki-test

2. Copy this prompt into Codex, Claude Code, QClaw, OpenClaw, or another local coding assistant.

Before sending it, replace every <replace-with-my-aiwiki-path> with your own local folder path, such as D:\AIWiki or ~/AIWiki. Do not leave the placeholder in the commands.

Please install and configure AIWiki for me.

First check that Node.js is installed and that node --version is >=20.
If Node.js is missing or older than 20, stop and tell me how to upgrade before running npm install.

Use this knowledge base path:

<replace-with-my-aiwiki-path>

Replace every `<replace-with-my-aiwiki-path>` below with my own local folder path before running commands. If I left the placeholder unchanged, stop and ask me for the real path.

Run these commands:

npm install -g @itradingai/aiwiki@latest
aiwiki setup --path "<replace-with-my-aiwiki-path>" --yes
aiwiki agent sync --yes
aiwiki agent sync --path "<replace-with-my-aiwiki-path>" --yes
aiwiki agent check --json
aiwiki agent check --path "<replace-with-my-aiwiki-path>" --json
aiwiki doctor --path "<replace-with-my-aiwiki-path>"
aiwiki status --path "<replace-with-my-aiwiki-path>"

Then tell me:

1. whether AIWiki was installed successfully
2. which assistant targets were synced
3. whether workspace guidance was written
4. whether I need to restart or reload my assistant
5. what I should do next

3. Restart or reload your assistant if needed.

If assistant sync fails, open an Agent Integration issue and include the output of aiwiki agent check --json and aiwiki doctor --path "<workspace>".

Trying AIWiki for the first time? Use the short trial route in the Usage Guide.

Tell your assistant:

Ingest this into AIWiki:
<url>

Or:

Save this note into AIWiki:
<paste your note here>

The assistant reads the source and calls AIWiki to write the result into your local knowledge base.

Tell your assistant:

What does AIWiki know about <topic>?

The assistant should call:

aiwiki context "<topic>"

For direct terminal output, use:

aiwiki query "<topic>"

Tell your assistant:

Check and organize my AIWiki workspace.

The assistant should call:

aiwiki lint --json

A successful ingest creates a traceable knowledge package:

02-raw/articles/                  Raw source record
03-sources/article-cards/         Source card
05-wiki/source-knowledge/         Reusable wiki entry
09-runs/<run-id>/                 Processing record

Optional files may also be created when the assistant provides enough structured content:

04-claims/_suggestions/           Claim candidates
06-assets/_suggestions/           Reusable ideas or writing assets
07-topics/ready/                  Topic candidates
08-outputs/outlines/              Draft outlines

The Wiki Entry is the main reusable knowledge surface. The raw record and source card preserve traceability, so you can always go back from a summary to the source.

See:

• examples/demo-run/
• examples/obsidian-vault-sample/
• examples/public-trial-scenarios/

Most useful information dies in one of three places:

• bookmarks you never reopen
• chat summaries you cannot reuse
• notes that never become output

AIWiki helps your assistant turn reading into a durable local knowledge base.

Instead of saving links and losing context, you get Markdown files that your assistant can query, maintain, and reuse later.

• Build a personal research wiki from articles, PDFs, files, and notes.
• Turn useful reading into reusable ideas such as concepts, claims, topics, and outlines.
• Give Codex, Claude Code, or QClaw a stable local knowledge layer it can query before answering.
• Keep summaries traceable with raw records, source cards, Wiki Entries, and run records.

For a guided first trial, use the public-trial scenario pack. It includes article research, topic planning, and project decision memory samples, with commands, expected artifacts, query/context reuse examples, and WeChat-group-ready usage copy.

User gives a URL, file, note, or text
  -> AI assistant reads and understands it
  -> AIWiki writes structured Markdown files
  -> Assistant retrieves context later with aiwiki context
  -> AIWiki lint checks structure and consistency

AIWiki separates responsibilities:

• the assistant reads and understands sources
• AIWiki writes, links, queries, and checks the local knowledge base
• Markdown remains inspectable, editable, portable, and versionable

Technically, AIWiki is Agent-first: the host assistant reads and understands sources; AIWiki writes, links, queries, and checks the local Markdown knowledge base.

AIWiki is inspired by two useful ideas:

• LLM Wiki, popularized by Andrej Karpathy: compile sources into a persistent, maintainable wiki instead of rediscovering knowledge from raw documents every time.
• Content workflow thinking, seen in creator systems such as Dan Koe's: useful ideas should become reusable building blocks for topics, outlines, writing, and future work.

AIWiki does not simply copy either approach.

It turns them into a practical assistant-driven workflow:

source
  -> source card
  -> wiki entry
  -> reusable assets
  -> topics
  -> outlines
  -> future work

The goal is simple: make what your assistant reads useful again later.

AIWiki is built for assistant-driven workflows. aiwiki agent sync --yes syncs packaged AIWiki instructions into supported local assistant environments, and aiwiki agent sync --path "<workspace>" --yes writes workspace guidance so assistants entering the knowledge base use AIWiki commands first.

See Agent Handoff for the full command-first contract.

AIWiki writes plain Markdown and frontmatter.

Obsidian is optional but useful as a viewing surface. Dataview is an optional dashboard enhancement.

AIWiki does not require Obsidian, does not auto-install Dataview, and does not edit .obsidian.

Review Queue is not the main workflow. AIWiki creates Wiki entries first, then uses lint and assistant workflows to keep the workspace clean.

• AIWiki writes local Markdown and JSON files.
• AIWiki does not upload your knowledge base.
• AIWiki does not include a built-in LLM.
• AIWiki does not crawl the web by itself.
• npm install does not modify assistant configuration.
• Agent integration is explicit through aiwiki agent sync.

AIWiki currently focuses on:

• one local AIWiki knowledge base
• local Markdown and frontmatter retrieval
• assistant-driven ingest
• Source Cards and Wiki Entries
• context, query, lint, status, and doctor workflows

Semantic search, vector indexing, browser clipping, RSS collection, and enterprise permissions are intentionally out of scope for now.

AIWiki is not:

• a web crawler
• a WeChat reader
• a browser extension
• a built-in LLM
• a vector database
• a replacement for every RAG system
• an Obsidian plugin
• a default manual review queue
• a multi-knowledge-base manager
• an RSS or scheduled collection system

AIWiki receives content already read by your assistant and turns it into a local Markdown knowledge base.

AIWiki is developed by iTradingAI.

For Chinese users, scan the QR codes below to join the WeChat group or follow the official account.

• Docs
• Usage Guide
• Agent Handoff
• FAQ
• Showcase
• Public Trial Scenarios
• Trial Feedback Template
• Roadmap
• Release Notes

For local development:

npm install
npm run build
npm test
npm link

Use a temporary workspace for local testing:

aiwiki setup --path "./aiwiki-test" --yes
aiwiki doctor --path "./aiwiki-test"
aiwiki status --path "./aiwiki-test"
aiwiki ingest-agent --payload tests/fixtures/agent_payload.url.valid.json --path "./aiwiki-test"
aiwiki context "AI Agent" --path "./aiwiki-test"
aiwiki query "AI Agent" --path "./aiwiki-test"
aiwiki lint --path "./aiwiki-test"

MIT. See LICENSE.