All in One LLM Wiki

中文版 | English

▶️ Video Demo

Export your digital life into an AI-readable LLM Wiki, so your Agent no longer has to start from scratch getting to know you.

This is not a "pretty knowledge base for humans." It's more like a long-term context layer for AI/Agents: organizing notes, AI memory, music, movies, video platforms, browsers, health, games, toolchains, NAS, and other data into a structured Markdown Wiki, so that Agents can understand your real background when answering, recommending, planning, and automating.

Just like you wouldn't read all your notes at once - you'd check the table of contents first, then flip to the page you need. The Wiki distills raw data from every platform into a structured index. The Agent loads on demand, no need to ingest everything.

Compatible Agents

This is a general Agent Skill. Any agent that can read SKILL.md and local files can use it:

• OpenClaw
• Hermes Agent
• WorkBuddy
• Claude Code
• HanaAgent
• Any other LLM Agent with Skill / Tool / MCP support

Quick Install

git clone https://github.com/perinchiang/all-in-one-llm-wiki

Or just send this to your LLM Agent:

Use https://github.com/perinchiang/all-in-one-llm-wiki to create a private .wiki/ in my Obsidian vault from my exported Bilibili/Spotify/AI memory files.

Before / After

Without LLM Wiki:

You: Recommend a movie
Agent: “Inception” is a classic Nolan film with high ratings... (generic)

With LLM Wiki:

You: Recommend a movie
Agent: You've marked 327 movies as “watched” on Douban, with ratings
concentrated in the 7-8 range. You prefer thriller and sci-fi, and
tend to rate pure romance films lower. Your recent Bilibili favorites
include some “high-IQ mind-bending” clips, let me take a look...
Found it! I'd recommend “The Invisible Guest” — a Spanish thriller
with tight pacing. A movie this perfect for you and you haven't seen
it yet? Or would you prefer something lighter to unwind today?

The Wiki transforms your Agent from "generic encyclopedia" into "assistant that actually knows you."

What Problem Does It Solve

The problem with ordinary AI conversations is: every time you have to re-explain who you are, what you're doing, what you like, what tools you have, and what stage you're at.

The goal of All in One LLM Wiki is:

• Organize data scattered across various platforms into a Wiki that AI can search and reference.
• Preserve source, confidence, and privacy boundaries to avoid turning into an un-auditable "personality guess."
• Let Agents act based on real context instead of just giving generic advice.

For example:

• After importing Spotify, the Agent can select playlists based on your music taste and current context.
• After importing Douban, the Agent no longer recommends movies and books based only on trending charts.
• After importing Bilibili / YouTube, the Agent can understand what learning and entertainment topics you're actually following recently.
• After importing Chrome / Google Takeout, the Agent can understand your toolchain, project context, and attention distribution.
• After importing Garmin / Apple Health, the Agent can combine sleep, steps, and exercise records to give more grounded lifestyle advice.
• After importing AI platform memories, the Agent can merge fragmented understandings of you from multiple AIs, reducing repeated explanations.
• After importing IMA knowledge bases, the Agent can search your subscribed public knowledge bases for more professional answers.

Data Ingestion Overview

Entry Point	Recommended Export Method	What to Distill	Privacy Level
Obsidian / Local Notes	Directly read Markdown folder	Learning stages, projects, concepts, terminology	Medium
AI Platform Memory	Ask platform / export long-term memory, manually save as Markdown	Cross-platform self-profile, preferences, long-term goals	High
Bilibili	bilibili-cli login then export history / favorites / following	Current interests, learning videos, creator preferences	Medium-High
Garmin	Python garminconnect script to pull summaries	Sleep, steps, exercise, recovery suggestions	High
Apple Health	iPhone Health App avatar → Export All Health Data	Long-term health trends, exercise history	High
Spotify	Spotify Web API / OAuth	Music taste, scene playlists, artist preferences	Medium
Douban	Save profile page / entry data or self-crawl, parse ratings and short reviews	Movie, book, game taste	Medium
Google Takeout	Chrome / YouTube export	Browser toolchain, YouTube subscriptions and playlists	High
Steam	Steam Web API	Game library, playtime, game genre preferences	Medium
NAS / Media Library	SSH scan + qBittorrent/Jellyfin API	Local media library, automation capabilities	High
IMA Knowledge Base	IMA client subscription + OpenAPI search	Public knowledge base content, tech docs, industry materials	Medium
Calendar	ICS / CalDAV / platform export	Schedule, low-frequency / high-frequency events	High
Service Accounts / Toolchain	Manual checklist, email summary, password manager category export	Available tools, cloud services, automation platforms	High

Click the links above for setup guides

Directory Structure

.wiki/                    # Private knowledge base, do not commit to Git by default
  SCHEMA.md
  index.md
  log.md
  raw/
    ai-memory/
    platform-exports/
    health/
    notes/
  entities/
  concepts/
  queries/
  _archive/

Basic workflow:

1. First, set up the Wiki structure: SCHEMA.md, index.md, log.md, raw/, entities/, concepts/.
2. For each source you import, first place the raw export into the corresponding raw/ directory.
3. Then generate one or more entities/ or concepts/ pages.
4. Write a one-line summary in index.md so the Agent can read the directory first.
5. Record the source, time, import method, and privacy handling in log.md.
6. For sensitive sources, only save aggregate summaries, not the raw data.

Security & Privacy

Default principles: local-first, privacy-first, anonymize before sharing.

Core measures: .wiki/ hidden folder + .gitignore excludes raw data + permission layering (Agent reads summary layer by default, not raw layer).

Full security plan, .gitignore template, permission layering, and encryption advice: docs/security_EN.md

Page Template

Each entity page should include:

---
title: Example Profile
created: YYYY-MM-DD
updated: YYYY-MM-DD
type: entity
tags: [profile, source-name]
sources: [raw/platform-exports/example.json]
confidence: high
---

# Example Profile

> One sentence explaining what this page helps the Agent understand.

## Overview

| Dimension | Value |
| --- | --- |
| Source | API / export / manual |
| Date range | YYYY-MM-DD to YYYY-MM-DD |
| Confidence | high |

## Findings

- Fact 1
- Fact 2
- Inference: must be labeled as inference

## Agent affordances

- What the Agent can now do.
- What the Agent should avoid.

## Open questions

- Points requiring user confirmation.

## Cross-links

- [[related-page]]

Recommended log.md Format

## [YYYY-MM-DD] ingest | Source name
- Source: export/API/manual source
- Method: command, script, or manual steps
- Created:
  - entities/example.md - summary
- Updated:
  - index.md - added source summary
- Privacy: raw data private; public version uses aggregates only
- Notes: parser assumptions, skipped files, anomalies

Appendix

• This project is inspired by Karpathy's LLM-Wiki concept and can serve as the data ingestion layer for llm-wiki.
• Before publishing, use PUBLISHING_CHECKLIST.md as a final pass.