wikimem 0.2.3 → 0.8.0

package/README.md

@@ -1,228 +1,194 @@
-# llmwiki
+# `wikimem`
 
-[![npm version](https://img.shields.io/npm/v/llmwiki.svg)](https://www.npmjs.com/package/llmwiki)
-[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
-[![TypeScript](https://img.shields.io/badge/TypeScript-strict-blue.svg)](https://www.typescriptlang.org/)
-[![Tests](https://img.shields.io/badge/tests-passing-brightgreen.svg)](#tests)
-
-**Build self-improving knowledge bases with LLMs.**
+**Self-improving wiki IDE. Ingest anything. Query with any LLM. Three automations.**
 
-Drop files in. Get a structured, interlinked wiki out. It improves itself while you sleep.
+[![npm version](https://img.shields.io/npm/v/wikimem.svg)](https://www.npmjs.com/package/wikimem)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+[![Formats](https://img.shields.io/badge/formats-13%2B-blue.svg)](#13-format-ingestion)
 
 ```bash
-npx llmwiki init my-wiki
+npx wikimem@latest
 ```
 
-```
-raw/                        wiki/
-  2026-04-07/                 index.md ........... content catalog
-    paper.pdf     ──LLM──>   sources/paper.md ... summary + citations
-    podcast.mp3               entities/openai.md . people, orgs, tools
-    screenshot.png            concepts/rag.md .... ideas + frameworks
-    blog-url                  syntheses/ ......... cross-cutting analysis
-```
+## What is WikiMem?
 
-llmwiki processes any source (text, PDF, audio, video, images, URLs), compiles it into an interlinked markdown wiki with frontmatter and `[[wikilinks]]`, and opens directly in Obsidian.
+WikiMem takes [Karpathy's LLM wiki concept](https://x.com/karpathy/status/1908625766490001799) and turns it into a full IDE. Drop any file — PDF, audio, video, slides, spreadsheet, URL — and watch it compile into structured, interlinked wiki pages via Claude, GPT-4o, or Ollama. Three automations keep your knowledge base growing and self-improving while you sleep.
 
-Works with **Claude, OpenAI, or Ollama** (local). Your data stays on your machine.
-
-Inspired by [Andrej Karpathy's LLM Wiki pattern](https://x.com/karpathy/status/1908625766490001799).
-
-## Install
-
-```bash
-npm install -g llmwiki
 ```
-
-**Requirements:** Node.js >= 18 · An LLM API key (or Ollama running locally)
+raw/                        wiki/
+  paper.pdf                   index.md ........... content catalog
+  podcast.mp3    ──LLM──>    sources/paper.md ... summary + citations
+  screenshot.png              entities/openai.md . people, orgs, tools
+  meeting.docx                concepts/rag.md .... ideas + frameworks
+  blog-url                    syntheses/ ......... cross-cutting analysis
+```
 
 ## Quick Start
 
 ```bash
-# 1. Create a vault
-llmwiki init my-wiki
+# Create a vault and start the IDE
+npx wikimem init my-wiki
 cd my-wiki
-
-# 2. Ingest something
-llmwiki ingest https://en.wikipedia.org/wiki/Large_language_model
-llmwiki ingest ~/Documents/research-paper.pdf
-
-# 3. Ask questions
-llmwiki query "What are the key differences between RAG and compiled knowledge?"
+npx wikimem serve
 ```
 
-That's it. Your wiki is now a folder of markdown files you can open in Obsidian, VS Code, or any text editor.
+Open [http://localhost:3141](http://localhost:3141). That's it — you have a running wiki IDE.
 
-## Why llmwiki?
+```bash
+# Or ingest from the CLI
+wikimem ingest paper.pdf
+wikimem ingest https://en.wikipedia.org/wiki/Large_language_model
+wikimem query "What are the key themes across my sources?"
+```
 
-**The problem:** You have dozens of sources — papers, podcasts, articles, screenshots, meeting recordings. They sit in folders. You forget what's in them. When you need something, you search and re-read.
+## Features
 
-**RAG approach:** Chunk documents, embed them, retrieve at query time. Lossy, opaque, and the "knowledge" lives in a vector database you can't read.
+### 13+ Format Ingestion
 
-**llmwiki approach:** Compile sources into structured markdown pages with summaries, cross-references, and citations. The knowledge is readable, editable, version-controlled, and improves itself over time.
+Drop anything. WikiMem detects the file type, runs the right processor, and produces wiki pages with cross-references and citations.
 
-| | RAG | llmwiki |
-|---|---|---|
-| **Storage** | Vector embeddings | Plain markdown files |
-| **Readable?** | No (opaque vectors) | Yes (open in any editor) |
-| **Editable?** | Rebuild index | Edit any page |
-| **Version control** | Difficult | `git diff` |
-| **Self-improving** | No | Yes (LLM Council) |
-| **Works offline** | Depends | Yes (with Ollama) |
-| **Obsidian** | Plugin required | Native (`[[wikilinks]]`) |
+| Format | Extensions | Processor |
+|--------|-----------|-----------|
+| Text | `.md`, `.txt` | Direct read |
+| Structured | `.json`, `.csv`, `.yaml` | Schema-aware extraction |
+| PDF | `.pdf` | Built-in text extraction |
+| Office | `.docx`, `.pptx`, `.xlsx` | Document parsing |
+| HTML | `.html`, `.htm` | Tag stripping + content extraction |
+| Image | `.png`, `.jpg`, `.gif`, `.webp` | Claude Vision description |
+| Audio | `.mp3`, `.wav`, `.m4a`, `.ogg`, `.flac` | Whisper / Deepgram transcription |
+| Video | `.mp4`, `.mov`, `.avi`, `.mkv`, `.webm` | ffmpeg → Whisper transcription |
+| URL | `https://...` | Firecrawl / fetch → markdown |
 
-## Features
+### Knowledge Graph
 
-- [x] **Multi-format ingestion** — text, PDF, audio, video, images, URLs, Office docs
-- [x] **Multi-model support** — Claude, OpenAI, Ollama (local)
-- [x] **Obsidian-native** — `[[wikilinks]]`, YAML frontmatter, opens directly as vault
-- [x] **Semantic dedup** — rejects near-duplicate sources automatically
-- [x] **BM25 search** — zero-dependency full-text search, no external services
-- [x] **Auto-indexing** — `index.md` + `log.md` maintained automatically
-- [x] **Watch mode** — drop files into `raw/`, auto-ingested
-- [x] **Self-improvement** — LLM Council scores your wiki and fixes issues
-- [x] **External scraping** — pull from RSS, GitHub, URLs on a schedule
-- [x] **Health checks** — find orphan pages, broken links, missing summaries
-- [x] **File-back answers** — query results saved as synthesis pages
-- [x] **Schema co-evolution** — AGENTS.md evolves with your wiki
-- [x] **Domain templates** — personal, research, business, codebase
-- [x] **Local-first** — everything is files. No database. No cloud dependency.
+D3-powered interactive force-directed graph. Click a node to highlight its neighbors, double-click to open. Community detection clusters related pages. Hub nodes sized by connection count.
 
-## Architecture
+### Time-Lapse
 
-```
-┌────────────────────────────────────────────────────┐
-│                    llmwiki CLI                      │
-│                                                    │
-│  llmwiki init         Create a new vault           │
-│  llmwiki ingest       Process source → wiki pages  │
-│  llmwiki query        Ask questions with citations  │
-│  llmwiki lint         Health-check the wiki        │
-│  llmwiki watch        Auto-ingest on file drop     │
-│  llmwiki scrape       Fetch from external sources  │
-│  llmwiki improve      Self-improvement cycle       │
-│  llmwiki status       Vault statistics             │
-├──────────────────────┬─────────────────────────────┤
-│   Three Layers       │   Three Automations         │
-│                      │                             │
-│   raw/               │   A1: Ingest & Process      │
-│   (immutable)   <────│   file/URL → markdown       │
-│                      │   → place in wiki/          │
-│   wiki/              │                             │
-│   (LLM-owned)   <────│   A2: External Scrape       │
-│                      │   RSS, GitHub, web → raw/   │
-│   AGENTS.md          │                             │
-│   (schema)      <────│   A3: Self-Improve          │
-│                      │   LLM Council → score → fix │
-├──────────────────────┴─────────────────────────────┤
-│                  LLM Providers                     │
-│   Claude (Anthropic) · OpenAI (GPT) · Ollama      │
-├────────────────────────────────────────────────────┤
-│                  Processors                        │
-│   Text · PDF · Audio · Video · Image · URL · HTML  │
-└────────────────────────────────────────────────────┘
-```
+Watch your knowledge base grow commit-by-commit. Every wiki change is checkpointed in git — scrub through the timeline to see pages appear, links form, and the graph densify.
 
-### Three Layers
+### WYSIWYG Editing
 
-1. **`raw/`** &mdash; Immutable source documents. Date-stamped subdirectories. Never modified by the LLM.
-2. **`wiki/`** &mdash; LLM-generated markdown. Source summaries, entity pages, concept pages, synthesis pages. The LLM owns this entirely.
-3. **`AGENTS.md`** &mdash; Schema file. Tells the LLM how the wiki is structured, what conventions to follow, how to process sources. Co-evolved by you and the LLM.
+Click any wiki page to edit it inline. Markdown shortcuts, live preview, `Cmd+S` to save. Changes are auto-committed to git.
 
 ### Three Automations
 
-- **A1: Ingest & Process** &mdash; Detects file type, runs the appropriate processor (Whisper for audio, ffmpeg+Whisper for video, Claude Vision for images, text extraction for PDF), asks the LLM to produce wiki pages with cross-references.
-- **A2: External Scrape** &mdash; Fetches from RSS feeds, GitHub trending, web URLs. Deposits results in `raw/` and triggers A1 automatically.
-- **A3: Self-Improve** &mdash; LLM Council evaluates wiki quality across 5 dimensions (coverage, consistency, cross-linking, freshness, organization), proposes improvements, and applies them if below a configurable threshold.
-
-## All Commands
+| Automation | Trigger | What it does |
+|------------|---------|--------------|
+| **Ingest** | File watcher on `raw/` | New file detected → process → wiki pages → git commit |
+| **Scrape** | Cron schedule or manual | RSS feeds, GitHub trending, URLs → fetch → deposit in `raw/` → triggers Ingest |
+| **Observe** | Nightly or manual | LLM Council scores wiki quality (coverage, consistency, cross-linking, freshness, organization) → proposes and applies improvements |
 
-### `llmwiki init [directory]`
+### Git Checkpointing
 
-Create a new vault with the standard directory structure.
+Every change committed automatically. Browse history, restore snapshots, see diffs. Your wiki is a git repo from day one.
 
-```bash
-llmwiki init my-wiki                    # Create in my-wiki/
-llmwiki init .                          # Initialize current directory
-llmwiki init my-wiki --template research   # Use research template
-llmwiki init my-wiki --force            # Overwrite existing
-```
+### Pipeline Visualization
 
-Templates: `personal` (default), `research`, `business`, `codebase`
+See exactly how your document flows through the system — file detection, text extraction, LLM processing, page generation, cross-linking, indexing — step by step in the web UI.
 
-### `llmwiki ingest <source>`
+### Connectors
 
-Process a file or URL into wiki pages.
+Sync external sources into your vault automatically.
 
-```bash
-llmwiki ingest paper.pdf                # PDF → extract text → wiki pages
-llmwiki ingest podcast.mp3              # Audio → Whisper transcription → wiki
-llmwiki ingest screenshot.png           # Image → Claude Vision description → wiki
-llmwiki ingest lecture.mp4              # Video → ffmpeg → Whisper → wiki
-llmwiki ingest article.md               # Markdown → wiki pages
-llmwiki ingest data.json                # JSON → code block in wiki
-llmwiki ingest page.html                # HTML → strip tags → wiki
-llmwiki ingest report.docx              # Office → basic extraction → wiki
-llmwiki ingest https://example.com/post # URL → Firecrawl/fetch → wiki
-llmwiki ingest raw/2026-04-07/file.md   # Re-ingest from raw/
-```
+| Connector | Status |
+|-----------|--------|
+| Local folders | ✅ Shipped |
+| Git repos | ✅ Shipped |
+| GitHub | ✅ Shipped |
+| Webhooks | ✅ Shipped |
+| Slack | 🔜 Coming soon |
+| Gmail | 🔜 Coming soon |
 
-Each source is auto-detected by file type, copied to `raw/{date}/`, checked for duplicates, compiled into wiki pages by the LLM, and indexed. Use `-p` to pick a provider, `-m` for a specific model, `--verbose` for detailed output.
+### MCP Server
 
-### `llmwiki query <question>`
-
-Ask a question and get an answer synthesized from your wiki.
+Use WikiMem as a tool inside Claude Code, Cursor, or any MCP-compatible client.
 
 ```bash
-llmwiki query "What are the main themes across my sources?"
-llmwiki query "Compare approaches to knowledge management" --file
-llmwiki query "Who is mentioned most frequently?" -p openai
+wikimem mcp
 ```
 
-Use `--file` to save the answer as a synthesis page in `wiki/syntheses/`. The query engine uses BM25 search to find relevant pages, reads the top 10, and synthesizes an answer with `[[wikilink]]` citations.
-
-### `llmwiki lint`
-
-Health-check the wiki for structural issues.
-
-```bash
-llmwiki lint                  # Check for issues
-llmwiki lint --fix            # Auto-fix where possible
+### Multiple LLMs
+
+| Provider | Flag | Default Model |
+|----------|------|---------------|
+| Claude | `-p claude` | `claude-sonnet-4-20250514` |
+| OpenAI | `-p openai` | `gpt-4o` |
+| Ollama | `-p ollama` | `llama3.2` |
+
+Ollama runs fully local — no API keys, no network, no data leaves your machine.
+
+## CLI Reference
+
+| Command | Description |
+|---------|-------------|
+| `wikimem init [dir]` | Create a new vault (`--template research\|business\|codebase`, `--from-folder`, `--from-repo`) |
+| `wikimem serve` | Start the web IDE on port 3141 |
+| `wikimem ingest <source>` | Process a file or URL into wiki pages |
+| `wikimem search <term>` | BM25 full-text search across wiki pages |
+| `wikimem ask <question>` | Ask a question, get an answer from your wiki |
+| `wikimem query <question>` | Ask a question and optionally save as synthesis page (`--file`) |
+| `wikimem lint` | Health-check: orphan pages, broken links, missing summaries (`--fix`) |
+| `wikimem status` | Vault statistics: pages, words, sources, links, orphans |
+| `wikimem watch` | Auto-ingest files dropped into `raw/` |
+| `wikimem scrape` | Fetch from configured RSS/GitHub/URL sources |
+| `wikimem improve` | Run self-improvement cycle (`--dry-run`, `--threshold 90`) |
+| `wikimem export` | Export wiki to other formats |
+| `wikimem open` | Open vault in Obsidian |
+| `wikimem history` | Browse audit trail, restore snapshots |
+| `wikimem mcp` | Start MCP server for Claude Code / Cursor |
+| `wikimem duplicates` | Detect and manage near-duplicate sources |
+
+## Web UI
+
+`wikimem serve` opens a full IDE at [localhost:3141](http://localhost:3141):
+
+- **File tree** — browse wiki pages with collapsible folders
+- **Tabbed editor** — open multiple pages, WYSIWYG markdown editing
+- **Knowledge graph** — interactive D3 force-directed visualization
+- **Pipeline view** — drag-and-drop file ingestion with step-by-step progress
+- **Time-lapse** — scrub through git history to watch your wiki grow
+- **Search** — `Cmd+K` fuzzy search across all pages
+- **Command palette** — `Cmd+P` for quick actions
+- **Settings** — configure API keys, models, and automations from the UI
+- **Ask your knowledge** — query your wiki from the browser
+
+## MCP Server
+
+WikiMem ships with a built-in MCP server so Claude Code and Cursor can read, search, and query your wiki directly.
+
+**Add to Claude Code** (`.mcp.json`):
+
+```json
+{
+  "mcpServers": {
+    "wikimem": {
+      "command": "npx",
+      "args": ["-y", "wikimem", "mcp"],
+      "env": {
+        "WIKIMEM_VAULT": "/path/to/your/vault"
+      }
+    }
+  }
+}
 ```
 
-Checks for:
-- Orphan pages (no inbound `[[wikilinks]]`)
-- Broken wikilinks (links to non-existent pages)
-- Pages missing frontmatter summaries
-- Near-empty pages (< 10 words)
-
-Reports a quality score out of 100.
-
-### `llmwiki watch`
-
-Watch the `raw/` directory and auto-ingest new files.
+**Or run standalone:**
 
 ```bash
-llmwiki watch                 # Watch current vault
-llmwiki watch -v ./my-wiki    # Watch a specific vault
+wikimem-mcp
 ```
 
-Uses `chokidar` for reliable cross-platform file watching. Waits for writes to stabilize before ingesting (2-second debounce). Press `Ctrl+C` to stop.
-
-### `llmwiki scrape`
-
-Fetch content from configured external sources and deposit in `raw/`.
-
-```bash
-llmwiki scrape                # Run all configured sources
-llmwiki scrape -s "HN Top"   # Run a specific source
-```
+## Configuration
 
-Sources are configured in `config.yaml`:
+After `wikimem init`, your vault contains `config.yaml`:
 
 ```yaml
+provider: claude                    # claude | openai | ollama
+model: claude-sonnet-4-20250514
+
 sources:
-  - name: "HN Top Stories"
+  - name: "HN Front Page"
     type: rss
     url: "https://hnrss.org/frontpage"
 
@@ -230,55 +196,11 @@ sources:
     type: github
     query: "stars:>100 created:>7d language:typescript"
 
-  - name: "Company Blog"
-    type: url
-    url: "https://example.com/blog"
+improvement:
+  threshold: 80
+  schedule: "0 3 * * *"            # 3am nightly
 ```
 
-Supported source types: `rss`, `github`, `url`
-
-### `llmwiki improve`
-
-Run the self-improvement cycle (Automation 3).
-
-```bash
-llmwiki improve                   # Evaluate and improve
-llmwiki improve --dry-run         # Show what would change
-llmwiki improve --threshold 90    # Stricter quality bar
-```
-
-The improvement cycle:
-
-1. **Score** &mdash; Evaluates 5 quality dimensions (coverage, consistency, cross-linking, freshness, organization)
-2. **Decide** &mdash; If score < threshold (default 80), improvements are needed
-3. **Improve** &mdash; Proposes actions: add cross-links, create missing pages, expand stubs, flag contradictions
-4. **Log** &mdash; Records what changed and why in `log.md`
-
-### `llmwiki status`
-
-Show vault statistics at a glance.
-
-```bash
-llmwiki status
-```
-
-```
-llmwiki vault status
-────────────────────────────────────
-  Pages:        42
-  Words:        18,340
-  Sources:      15
-  Wiki links:   127
-  Orphan pages: 2
-  Last updated: 2026-04-07
-```
-
-## Configuration
-
-After `llmwiki init`, your vault contains a `config.yaml` where you set the LLM provider, external sources, self-improvement schedule, and processing options.
-
-See [docs/configuration.md](docs/configuration.md) for the full reference.
-
 ### Environment Variables
 
 | Variable | Purpose |
@@ -286,113 +208,53 @@ See [docs/configuration.md](docs/configuration.md) for the full reference.
 | `ANTHROPIC_API_KEY` | Claude API access (default provider) |
 | `OPENAI_API_KEY` | OpenAI API access |
 | `OLLAMA_BASE_URL` | Ollama server URL (default: `http://localhost:11434`) |
-| `FIRECRAWL_API_KEY` | Enhanced URL-to-markdown (optional, falls back to fetch) |
-| `DEEPGRAM_API_KEY` | Audio transcription API (optional, falls back to Whisper) |
-
-## Multi-Model Support
+| `FIRECRAWL_API_KEY` | Enhanced URL-to-markdown (optional) |
+| `DEEPGRAM_API_KEY` | Audio transcription (optional, falls back to Whisper) |
 
-llmwiki works with any major LLM provider. Choose at init time or per-command.
-
-| Provider | Flag | Default Model | Env Variable |
-|----------|------|---------------|-------------|
-| **Claude** | `-p claude` | `claude-sonnet-4-20250514` | `ANTHROPIC_API_KEY` |
-| **OpenAI** | `-p openai` | `gpt-4o` | `OPENAI_API_KEY` |
-| **Ollama** | `-p ollama` | `llama3.2` | `OLLAMA_BASE_URL` |
-
-```bash
-# Use Claude (default)
-llmwiki ingest paper.pdf
-
-# Use OpenAI
-llmwiki ingest paper.pdf -p openai -m gpt-4o-mini
-
-# Use Ollama (fully local, no API keys)
-llmwiki ingest paper.pdf -p ollama -m llama3.2
-```
-
-## Multi-Format Support
-
-| Format | Extensions | Processor | Requirements |
-|--------|-----------|-----------|-------------|
-| **Text** | `.md`, `.txt`, `.csv` | Direct read | None |
-| **PDF** | `.pdf` | Built-in text extraction | None |
-| **Audio** | `.mp3`, `.wav`, `.m4a`, `.ogg`, `.flac`, `.aac` | Whisper / Deepgram | `whisper` CLI or `DEEPGRAM_API_KEY` |
-| **Video** | `.mp4`, `.mov`, `.avi`, `.mkv`, `.webm` | ffmpeg + Whisper | `ffmpeg` + `whisper` |
-| **Image** | `.jpg`, `.png`, `.gif`, `.webp` | Claude Vision | `ANTHROPIC_API_KEY` |
-| **HTML** | `.html`, `.htm` | Tag stripping | None |
-| **JSON** | `.json` | Code block wrapping | None |
-| **Office** | `.docx`, `.pptx`, `.xlsx` | Basic extraction | None (enhanced coming) |
-| **URL** | `https://...` | Firecrawl / fetch | Optional `FIRECRAWL_API_KEY` |
-
-When a processor's requirements are not met (e.g., Whisper not installed for audio), llmwiki creates a reference page noting the source file and suggests installing the missing tool. The raw file is always preserved.
-
-## Obsidian Integration
-
-llmwiki vaults are Obsidian vaults. Open any llmwiki directory in Obsidian and you get:
-
-- **Graph view** showing all pages and their `[[wikilinks]]`
-- **YAML frontmatter** rendered as page metadata
-- **Backlinks** panel showing what links to each page
-- **Search** across all wiki content
-- **Tag view** from frontmatter `tags:` arrays
-
-No plugins required. No configuration. Just `Open folder as vault` in Obsidian.
+## Architecture
 
 ```
-# Every wiki page has frontmatter like this:
----
-title: "Attention Is All You Need"
-type: source
-created: "2026-04-07"
-updated: "2026-04-07"
-tags: [transformers, attention, nlp]
-sources: ["raw/2026-04-07/attention-paper.pdf"]
-summary: "Foundational transformer architecture paper introducing self-attention"
----
+vault/
+├── wiki/           ← LLM-generated pages (sources/, entities/, concepts/, syntheses/)
+├── raw/            ← Immutable source documents (date-stamped subdirectories)
+├── AGENTS.md       ← Schema — wiki structure + conventions
+├── config.yaml     ← Configuration — provider, sources, schedules
+└── index.md        ← Content catalog (auto-maintained)
 ```
 
-## Vault Structure
+**Three layers:** `raw/` (immutable sources) → LLM processing → `wiki/` (structured knowledge). `AGENTS.md` is the schema file that tells the LLM how to structure output — it co-evolves with your wiki.
 
-```
-my-wiki/
-├── AGENTS.md             # Schema — wiki structure + conventions
-├── config.yaml           # Configuration — provider, sources, schedules
-├── .gitignore
-├── raw/                  # Immutable source archive
-│   ├── 2026-04-07/
-│   │   ├── paper.pdf
-│   │   ├── podcast.mp3
-│   │   └── blog-post.md
-│   └── 2026-04-08/
-│       └── meeting-notes.md
-└── wiki/                 # LLM-generated knowledge base
-    ├── index.md          # Auto-maintained content catalog
-    ├── log.md            # Chronological operation record
-    ├── sources/          # One summary per ingested source
-    ├── entities/         # People, organizations, tools
-    ├── concepts/         # Ideas, frameworks, patterns
-    └── syntheses/        # Cross-cutting analyses, query results
-```
+**Three automations:** Ingest (file watcher → process → wiki pages), Scrape (RSS/GitHub/URLs → raw/), Observe (LLM Council → score → improve).
 
-## Tests
+## Obsidian Integration
 
-```bash
-cd /path/to/llmwiki && pnpm test
-```
+WikiMem vaults are Obsidian vaults. Open any wikimem directory in Obsidian — no plugins, no configuration:
 
-## Contributing
+- `[[wikilinks]]` rendered as backlinks
+- YAML frontmatter as page metadata
+- Graph view showing all connections
+- Tag view from frontmatter `tags:` arrays
 
-See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines.
+## Privacy
 
-## See Also
+- Everything runs locally. Your wiki is a folder of markdown files.
+- No data sent anywhere except LLM API calls (and those are optional with Ollama).
+- `raw/` excluded from git by default — your source documents stay private.
+- `config.yaml` excluded from git — API keys never committed.
 
-- **[agentgrid](https://github.com/naman10parikh/agentgrid)** &mdash; Manage grids of AI coding agents in tmux
-- **[agentdial](https://github.com/naman10parikh/Energy)** &mdash; Universal agent identity protocol across 8 messaging channels
+| Path | Safe to commit? | Why |
+|------|:-:|-----|
+| `wiki/` | ✅ | LLM-generated summaries, no raw personal data |
+| `AGENTS.md` | ✅ | Schema file, no personal data |
+| `raw/` | ❌ | Original source files |
+| `config.yaml` | ❌ | May contain API keys |
 
 ## Credits
 
-Inspired by [Andrej Karpathy's LLM Wiki pattern](https://x.com/karpathy/status/1908625766490001799) &mdash; the idea that LLMs should compile knowledge into structured, interlinked wikis rather than just answering questions from raw chunks.
+Inspired by [Andrej Karpathy's LLM Wiki pattern](https://x.com/karpathy/status/1908625766490001799) — the idea that LLMs should compile knowledge into structured, interlinked wikis rather than just answering questions from raw chunks.
+
+Built with [Express](https://expressjs.com/), [D3](https://d3js.org/), [simple-git](https://github.com/steveukx/git-js), and the [Claude](https://docs.anthropic.com/) / [OpenAI](https://platform.openai.com/) / [Ollama](https://ollama.com/) APIs.
 
 ## License
 
-MIT &mdash; see [LICENSE](LICENSE).
+MIT — see [LICENSE](LICENSE).

package/dist/cli/commands/ask.d.ts

@@ -0,0 +1,3 @@
+import { Command } from 'commander';
+export declare function registerAskCommand(program: Command): void;
+//# sourceMappingURL=ask.d.ts.map

package/dist/cli/commands/ask.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ask.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/ask.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAiBpC,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,OAAO,GAAG,IAAI,CA2DzD"}

package/dist/cli/commands/ask.js

@@ -0,0 +1,63 @@
+import chalk from 'chalk';
+import ora from 'ora';
+import { resolve } from 'node:path';
+import { existsSync } from 'node:fs';
+import { queryWiki } from '../../core/query.js';
+import { getVaultConfig } from '../../core/vault.js';
+import { createProviderFromUserConfig } from '../../providers/index.js';
+import { loadConfig } from '../../core/config.js';
+export function registerAskCommand(program) {
+    program
+        .command('ask <question>')
+        .description('Ask a question — LLM-powered Q&A against the wiki')
+        .option('-v, --vault <path>', 'Vault root directory', '.')
+        .option('-p, --provider <provider>', 'LLM provider (claude, openai, ollama)')
+        .option('-m, --model <model>', 'Model to use')
+        .option('--file', 'Save the answer as a new wiki page')
+        .action(async (question, options) => {
+        const vaultRoot = resolve(options.vault ?? '.');
+        const config = getVaultConfig(vaultRoot);
+        const userConfig = loadConfig(config.configPath);
+        if (!existsSync(config.schemaPath)) {
+            console.error(chalk.red('Not a wikimem vault. Run `wikimem init` first.'));
+            process.exit(1);
+        }
+        const provider = createProviderFromUserConfig(userConfig, {
+            providerOverride: options.provider,
+            model: options.model,
+        });
+        const spinner = ora('Thinking…').start();
+        try {
+            const result = await queryWiki(question, config, provider, {
+                fileBack: options.file ?? false,
+            });
+            spinner.stop();
+            console.log();
+            const formatted = result.answer
+                .replace(/^(#{1,3})\s+(.+)$/gm, (_m, hashes, text) => {
+                if (hashes === '#')
+                    return chalk.bold.underline(text);
+                if (hashes === '##')
+                    return chalk.bold(text);
+                return chalk.italic(text);
+            })
+                .replace(/\[\[([^\]]+)\]\]/g, (_m, link) => chalk.cyan(`[[${link}]]`))
+                .replace(/`([^`]+)`/g, (_m, code) => chalk.yellow(code))
+                .replace(/\*\*([^*]+)\*\*/g, (_m, bold) => chalk.bold(bold));
+            console.log(formatted);
+            console.log();
+            if (result.sourcesConsulted.length > 0) {
+                console.log(chalk.dim(`Sources: ${result.sourcesConsulted.join(', ')}`));
+            }
+            if (result.filedAs) {
+                console.log(chalk.green(`Saved to: ${result.filedAs}`));
+            }
+        }
+        catch (error) {
+            spinner.fail('Failed');
+            console.error(chalk.red(error instanceof Error ? error.message : String(error)));
+            process.exit(1);
+        }
+    });
+}
+//# sourceMappingURL=ask.js.map

package/dist/cli/commands/ask.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ask.js","sourceRoot":"","sources":["../../../src/cli/commands/ask.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,MAAM,OAAO,CAAC;AAC1B,OAAO,GAAG,MAAM,KAAK,CAAC;AACtB,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACrC,OAAO,EAAE,SAAS,EAAE,MAAM,qBAAqB,CAAC;AAChD,OAAO,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AACrD,OAAO,EAAE,4BAA4B,EAAE,MAAM,0BAA0B,CAAC;AACxE,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AASlD,MAAM,UAAU,kBAAkB,CAAC,OAAgB;IACjD,OAAO;SACJ,OAAO,CAAC,gBAAgB,CAAC;SACzB,WAAW,CAAC,mDAAmD,CAAC;SAChE,MAAM,CAAC,oBAAoB,EAAE,sBAAsB,EAAE,GAAG,CAAC;SACzD,MAAM,CAAC,2BAA2B,EAAE,uCAAuC,CAAC;SAC5E,MAAM,CAAC,qBAAqB,EAAE,cAAc,CAAC;SAC7C,MAAM,CAAC,QAAQ,EAAE,oCAAoC,CAAC;SACtD,MAAM,CAAC,KAAK,EAAE,QAAgB,EAAE,OAAmB,EAAE,EAAE;QACtD,MAAM,SAAS,GAAG,OAAO,CAAC,OAAO,CAAC,KAAK,IAAI,GAAG,CAAC,CAAC;QAChD,MAAM,MAAM,GAAG,cAAc,CAAC,SAAS,CAAC,CAAC;QACzC,MAAM,UAAU,GAAG,UAAU,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;QAEjD,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,UAAU,CAAC,EAAE,CAAC;YACnC,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,gDAAgD,CAAC,CAAC,CAAC;YAC3E,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,MAAM,QAAQ,GAAG,4BAA4B,CAAC,UAAU,EAAE;YACxD,gBAAgB,EAAE,OAAO,CAAC,QAAQ;YAClC,KAAK,EAAE,OAAO,CAAC,KAAK;SACrB,CAAC,CAAC;QAEH,MAAM,OAAO,GAAG,GAAG,CAAC,WAAW,CAAC,CAAC,KAAK,EAAE,CAAC;QAEzC,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,SAAS,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,EAAE;gBACzD,QAAQ,EAAE,OAAO,CAAC,IAAI,IAAI,KAAK;aAChC,CAAC,CAAC;YAEH,OAAO,CAAC,IAAI,EAAE,CAAC;YACf,OAAO,CAAC,GAAG,EAAE,CAAC;YAEd,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM;iBAC5B,OAAO,CAAC,qBAAqB,EAAE,CAAC,EAAE,EAAE,MAAc,EAAE,IAAY,EAAE,EAAE;gBACnE,IAAI,MAAM,KAAK,GAAG;oBAAE,OAAO,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;gBACtD,IAAI,MAAM,KAAK,IAAI;oBAAE,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;gBAC7C,OAAO,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;YAC5B,CAAC,CAAC;iBACD,OAAO,CAAC,mBAAmB,EAAE,CAAC,EAAE,EAAE,IAAY,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,CAAC;iBAC7E,OAAO,CAAC,YAAY,EAAE,CAAC,EAAE,EAAE,IAAY,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;iBAC/D,OAAO,CAAC,kBAAkB,EAAE,CAAC,EAAE,EAAE,IAAY,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;YAEvE,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;YACvB,OAAO,CAAC,GAAG,EAAE,CAAC;YAEd,IAAI,MAAM,CAAC,gBAAgB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACvC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,YAAY,MAAM,CAAC,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC;YAC3E,CAAC;YAED,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;gBACnB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,aAAa,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YAC1D,CAAC;QACH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YACvB,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;YACjF,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;IACH,CAAC,CAAC,CAAC;AACP,CAAC"}