RagClaw Knowledge Base Skill

Local-first knowledge base for OpenClaw.

Description

Index and search your documents, code, and web pages locally. Zero external APIs, offline embeddings, SQLite-based storage.

Commands

/kb add <source>

Index a file, directory, or URL.

Examples:

/kb add ./docs/
/kb add https://docs.example.com
/kb add ~/projects/my-app/src/
/kb add https://docs.example.com --crawl --crawl-max-depth 2

Options:

• --db <name> — Knowledge base name (default: "default")
• --recursive — Recurse into directories (default: true)
• --embedder <preset> — Embedder preset: nomic|bge|mxbai|minilm (default: nomic)
• --include <pattern> — Regex filter: include only matching filenames
• --exclude <pattern> — Regex filter: exclude matching filenames
• --max-depth <n> — Maximum directory recursion depth
• --max-files <n> — Maximum number of files to index
• --crawl — Follow links from a seed URL
• --crawl-max-depth <n> — Link traversal depth (default: 3)
• --crawl-max-pages <n> — Max pages to fetch (default: 100)
• --crawl-same-origin — Stay on the same domain (default: true)
• --crawl-include <patterns> — Comma-separated URL path prefixes to include
• --crawl-exclude <patterns> — Comma-separated URL path prefixes to exclude
• --crawl-concurrency <n> — Concurrent fetchers (default: 1)
• --crawl-delay <ms> — Delay between requests in ms (default: 1000)
• --enforce-guards — Enable path/URL security guards

/kb search <query>

Search the knowledge base.

Examples:

/kb search how to configure authentication
/kb search async function error handling
/kb search "memory leak" --mode hybrid --limit 10

Options:

• --db <name> — Knowledge base name (default: "default")
• --limit <n> — Max results (default: 5)
• --mode <mode> — Search mode: vector|keyword|hybrid (default: hybrid)
• --json — Machine-readable JSON output

/kb reindex

Re-process changed sources and keep vectors up to date.

Options:

• --db <name> — Knowledge base name (default: "default")
• -f, --force — Force full rebuild (ignore hashes)
• -p, --prune — Remove sources that no longer exist on disk
• --embedder <preset> — Switch embedder and rebuild all vectors

/kb merge <source.sqlite>

Merge another knowledge base into the local one.

Options:

• --db <name> — Destination knowledge base (default: "default")
• --strategy <strict|reindex> — strict copies vectors verbatim (same embedder required); reindex re-embeds locally (default: strict)
• --on-conflict <skip|prefer-local|prefer-remote> — Conflict resolution (default: skip)
• --dry-run — Preview changes without writing
• --include <paths> — Comma-separated path prefixes to import
• --exclude <paths> — Comma-separated path prefixes to skip

/kb status

Show knowledge base statistics (chunks, sources, vector backend, embedder).

Options:

• --db <name> — Knowledge base name (default: "default")

/kb list

List indexed sources.

Options:

• --db <name> — Knowledge base name (default: "default")
• -t <file|url> — Filter by source type

/kb remove <source>

Remove a source from the index.

Options:

• --db <name> — Knowledge base name (default: "default")
• -y — Skip confirmation prompt

/kb embedder list

List all available embedder presets with RAM requirements and status.

/kb embedder download [preset]

Pre-download a model for offline use.

Options:

• --all — Download all built-in presets

/kb doctor

Check system health: Node.js version, RAM, sqlite-vec status, embedder compatibility, loaded plugins.

/kb plugin list

List discovered plugins with enabled/disabled status.

/kb plugin enable <name>

Enable a plugin (use --all for all discovered plugins).

/kb plugin disable <name>

Disable a plugin.

/kb config list

Show all configuration values and their source (env / config file / default).

/kb config get <key>

Show a single config value.

/kb config set <key> <value>

Persist a config value to ~/.config/kbclaw/config.yaml.

Supported Formats

Embedder Presets

Run /kb doctor to check which presets fit your available RAM.

Storage

Knowledge bases are stored as SQLite files following XDG conventions:

• Default data dir: ~/.local/share/kbclaw/
• Config file: ~/.config/kbclaw/config.yaml
• Backwards compat: if ~/.openclaw/kbclaw/ exists it will be used automatically.

How It Works

1. Extract — Pull text from documents (PDF, DOCX, HTML, code, images via OCR)
2. Chunk — Split into semantic units (paragraphs, functions, classes)
3. Embed — Generate vectors using a configurable local model (default: nomic-embed-text-v1.5, 768 dims)
4. Store — SQLite with FTS5 for keyword search; embedder info written to DB metadata
5. Search — Hybrid: 70% vector similarity + 30% BM25 keyword; embedder auto-detected from DB

All processing happens locally. No API keys required.