Files
llmwiki-cli/README.md
doum1004 b19bd2e408 Add foundational concepts and entities related to LLMs and AI agents
- Create context-window.md to explain the significance of context window size in LLMs.
- Add llm-scaling-laws.md detailing the empirical relationships between model performance and resources.
- Introduce retrieval-augmented-generation.md to describe RAG architecture and its advantages.
- Add entity pages for key figures and organizations: andrej-karpathy.md, anthropic.md, google-deepmind.md, openai.md, sam-altman.md.
- Create sources for foundational papers: attention-is-all-you-need.md, claude-model-card.md, gpt4-technical-report.md, react-paper.md.
- Synthesize insights on AI agent patterns and RAG vs fine-tuning in dedicated pages.
- Update index.md to include new entities and concepts.
- Log all activities related to the wiki's development in log.md.
2026-04-13 00:05:30 -04:00

13 KiB
Raw Blame History

llmwiki-cli

npm version total downloads CI license bun ko-fi

A CLI tool for LLM agents to build and maintain personal knowledge bases.

Inspired by Andrej Karpathy's LLM Wiki.

Live Demo — interactive d3-force graph built from the example wiki in test-wiki-page/.

Overview

The CLI is the hands -- it reads, writes, searches, and manages wiki files. The LLM is the brain -- it decides what to create, update, and connect.

LLM Agent (Claude Code / Codex)
|
| shells out to:
|   $ wiki init my-wiki --domain "machine learning"
|   $ wiki write wiki/concepts/attention.md <<'EOF' ... EOF
|   $ wiki index add "concepts/attention.md" "Overview of attention"
|   $ wiki search "scaling laws"
|   $ wiki lint
|
v
wiki CLI (StorageProvider abstraction)
|
v
filesystem | git (auto-commit + auto-push) | supabase (database)

Key principle: The CLI never calls any LLM API. It is a pure storage tool with pluggable backends.

Install

npm install -g llmwiki-cli

This gives you two commands: wiki (primary, 4 chars) and llmwiki (fallback if wiki conflicts).

Storage Backends

Backend Description Init
filesystem (default) Plain markdown files on disk wiki init my-wiki
git Filesystem + auto-commit + auto-push to GitHub wiki init my-wiki --backend git --git-token <pat>
supabase Pages in a Supabase database table wiki init my-wiki --backend supabase --supabase-url <url> --supabase-key <key>

Quick Start

# Create a new wiki (filesystem backend, default)
wiki init my-wiki --name "My Notes" --domain "research"

# Or with git + GitHub sync
wiki init my-wiki --name "My Notes" --domain "research" --backend git --git-token ghp_xxx

# Write a page
wiki write wiki/concepts/attention.md <<'EOF'
---
title: Attention Mechanism
created: 2025-01-20
tags: [transformers, NLP]
---
The attention mechanism allows models to focus on relevant parts of the input.
See also [[transformers]] and [[self-attention]].
EOF

# Add to index and log
wiki index add "concepts/attention.md" "Overview of attention mechanisms"
wiki log append ingest "Attention mechanism page"

# Search and lint
wiki search "attention"
wiki lint

Wiki Structure

When you run wiki init (filesystem or git backend), it creates:

my-wiki/
├── .git/                  # Only with --backend git
├── .github/               # Only with --backend git (--viz)
│   └── workflows/
│       └── wiki-viz.yml   # GitHub Actions → GitHub Pages visualization
├── .gitignore             # Only with --backend git (--viz)
├── .llmwiki.yaml          # Wiki config (all backends)
├── SCHEMA.md              # Instructions for LLM agents
├── scripts/               # Only with --backend git (--viz)
│   ├── build-graph.js     # Builds graph.json from wikilinks
│   └── build-site.js      # Generates d3-force visualization
├── raw/                   # Immutable source documents
│   └── assets/            # Downloaded images
└── wiki/                  # LLM-generated pages
    ├── index.md           # Master index of all pages
    ├── log.md             # Chronological activity log
    ├── entities/          # People, orgs, products
    ├── concepts/          # Ideas, frameworks, theories
    ├── sources/           # One summary per ingested source
    └── synthesis/         # Cross-cutting analysis

For supabase backend, only .llmwiki.yaml is created locally. Pages are stored in the wiki_pages database table.

Schema and init: wiki init --backend supabase runs a quick check against wiki_pages (required columns, upsert on user_id,wiki_id,path with nullable user_id). If the table is missing or incompatible, the CLI prints ready-to-run SQL for PostgreSQL 15+ (unique nulls not distinct on user_id, wiki_id, path). Apply it in the Supabase SQL Editor, then run wiki init again. If initial seed writes still fail with a schema-style error, the same SQL is printed again as a hint.

user_id: Optional. NULL means a single shared partition (works well with the service role key for personal tooling). When set, the row is scoped to that auth user. The trigger fills user_id from auth.uid() when the client leaves it null and a user session exists.

RLS and auth: Policies allow the authenticated role to read and write rows where user_id is null or auth.uid() = user_id. Use the anon key in .llmwiki.yaml plus a Supabase Auth access token (JWT) per session: LLMWIKI_SUPABASE_ACCESS_TOKEN, or optional supabase.access_token in config (avoid committing secrets). The service role key bypasses RLS; do not rely on it for multi-tenant isolation. Older layouts (for example (wiki_id, path) only, or non-null user_id without the right unique constraint) need a migration or drop table public.wiki_pages cascade before applying the printed SQL.

Storage profiles (all backends): wiki profile use <slug>, --profile, LLMWIKI_PROFILE, or top-level profile in .llmwiki.yaml (legacy: supabase.profile). For filesystem and git, pages live under profiles/<slug>/ inside the wiki repo. For Supabase, the same slug selects the composite wiki_id. This is organizational separation only, not OS or cryptographic isolation.

Commands

Wiki Management

wiki init [dir] --name <name> --domain <domain> --backend <type>
wiki init [dir] --backend git --git-token <pat> [--git-repo owner/repo]
wiki init [dir] --backend git --no-viz              # Skip visualization scaffolding
wiki init [dir] --backend supabase --supabase-url <url> --supabase-key <key>
wiki init [existing-wiki-dir] --viz                 # Add visualization to existing git wiki
wiki registry                                       # List all wikis
wiki use [wiki-id]                                  # Set active wiki
wiki profile show                                   # Effective storage root / Supabase wiki_id
wiki profile use <slug>                             # Save profile in registry (all backends)
wiki profile clear                                  # Remove saved profile

Reading & Writing

wiki read <path>                                    # Print page to stdout
wiki write <path>                                   # Write stdin to page
wiki append <path>                                  # Append stdin to page
wiki list [dir] [--tree] [--json]                   # List pages
wiki search <query> [--limit N] [--all] [--json]    # Search pages

Index & Log

wiki index show                                     # Print master index
wiki index add <path> <summary>                     # Add entry to index
wiki index remove <path>                            # Remove entry
wiki log show [--last N] [--type T]                 # Print log entries
wiki log append <type> <message>                    # Append log entry
wiki lint [--json]                                  # Health check
wiki links <path>                                   # Outbound + inbound links
wiki backlinks <path>                               # Inbound links only
wiki orphans                                        # Pages with no inbound links
wiki status [--json]                                # Wiki overview stats

LLM Agent Skill Guide

Run wiki skill to print the full guide, or see docs/SKILL.md for the source. Covers workflows, command patterns, page format, and common gotchas for LLM agents.

How LLM Agents Use This

The generated SCHEMA.md in each wiki contains complete instructions. Here are the typical workflows:

Ingest a Source

# Save raw source
wiki write raw/paper.md <<'EOF'
<paste full text of paper>
EOF

# Create structured summary
wiki write wiki/sources/attention-paper.md <<'EOF'
---
title: Attention Is All You Need
created: 2025-01-20
tags: [transformers, attention, NLP]
source: https://arxiv.org/abs/1706.03762
---
Summary of the attention paper...
Links to [[transformers]] and [[self-attention]].
EOF

# Update bookkeeping
wiki index add "sources/attention-paper.md" "Attention Is All You Need (2017)"
wiki log append ingest "Attention paper"

Answer a Question

wiki search "attention mechanism"
wiki read wiki/concepts/attention.md
wiki links wiki/concepts/attention.md    # see related pages
wiki log append query "How does multi-head attention work?"

Maintain Wiki Health

wiki lint                  # find broken links, orphans, missing frontmatter
wiki orphans               # pages nobody links to
wiki status                # overview stats

Git backend and GitHub

  • New repo: With --git-token and no --git-repo, the CLI creates wiki-<name> as a public repository (straightforward to try GitHub Pages). Use --git-repo owner/existing to target a repo you already manage.
  • Token not committed: .llmwiki.yaml stores git.repo only, not your PAT, so GitHubs push secret scanning does not reject commits. For wiki write / wiki append (auto-push), set LLMWIKI_GIT_TOKEN, GITHUB_TOKEN, or GIT_TOKEN in your environment.
  • PAT permissions: The token must be allowed to update GitHub Actions workflow files (classic PAT: include the workflow scope; fine-grained: grant workflow/Actions write). Otherwise the first push after init can fail on .github/workflows/wiki-viz.yml.

Graph Visualization

Git-backend wikis automatically include a GitHub Actions workflow that builds an interactive d3-force graph visualization of your wiki's link structure and deploys it to GitHub Pages.

  • Auto-scaffolded: wiki init --backend git creates .github/workflows/wiki-viz.yml and scripts/ build scripts by default
  • On every push: GitHub Actions parses all [[wikilinks]], builds a force-directed graph, and deploys to Pages
  • Interactive: color-coded nodes by directory, zoom/pan, click-to-highlight connections, hover tooltips
  • Opt-out: use --no-viz to skip visualization scaffolding
  • Add to existing wiki: re-run wiki init <dir> --viz on an existing git wiki to add visualization files

After a successful first push, init also tries to enable GitHub Pages (build: GitHub Actions). You can still configure it under Settings → Pages → Source: GitHub Actions if needed.

Branch name: New git wikis rename the default branch to main before the first push so the github-pages Actions environment (which often only allows main) accepts deployment. If you see “Branch master is not allowed to deploy to github-pages”, rename locally with git branch -M main, push main, and under Settings → General set the default branch to main (and optionally delete the old master branch on GitHub).

Multi-Wiki Support

The CLI supports multiple wikis via a global registry at ~/.config/llmwiki/:

wiki init ~/wikis/ml --name ml --domain "machine learning"
wiki init ~/wikis/personal --name personal --domain "personal notes"
wiki registry        # lists both
wiki use ml          # switch active wiki
wiki --wiki personal read wiki/index.md   # target specific wiki
wiki search "neural networks" --all       # search across all wikis

Wiki resolution order: --wiki flag > cwd .llmwiki.yaml > walk up directories > registry default.

Requirements

  • Node.js >= 18 (or Bun)
  • Git (for --backend git only)
  • @supabase/supabase-js (for --backend supabase only — optional dependency)

Development

git clone https://github.com/doum1004/llmwiki-cli
cd llmwiki-cli
bun install
bun test            # 194 tests
bun run build       # bundle to dist/wiki.js
bun run dev -- --help

DATA

Visitors

License

MIT