Offline semantic search engine for documentation. Embeddings are generated locally at build time. The browser runtime is pure math - no model weights, no cloud APIs, no data leaving your machine.

Zero config

Run npx docmd-search ./docs in any folder. No setup, no API keys, no config file needed.

Two ways to use it

docmd-search is a completely standalone tool. It can also integrate with docmd when you want semantic search in your documentation site.

Standalone CLI

Run docmd-search ./any-folder to index any directory and get an interactive terminal search. Add --ui to launch a web interface powered by docmd.

docmd plugin

Add semantic: true to your docmd config and docmd-search powers the search. No code changes - just a config flag.

How it works

Build time (Node.js)                    Search time (Browser, <3KB)
───────────────────                     ──────────────────────────
 Crawl files                             Load manifest.json
   → Chunk by heading                      → Load batch 000 (instant)
     → Embed via ONNX                        → Background-load rest
       → Quantize Float32 → Int8               → Keyword + cosine
         → Compress (ternary/PQ)                 → Ranked results
           → Save multi-batch index

All embedding computation happens at build time using ONNX Runtime on your machine. The browser receives only pre-computed integer vectors and performs keyword matching + cosine similarity - no neural network inference, just arithmetic.

Key capabilities

Offline by default

All embeddings generated locally with ONNX Runtime. No data leaves your machine. No cloud API keys needed.

Instant search

Progressive indexing means search is available from the first batch. Incremental re-indexing only processes changed files.

Tiny client

Browser runtime under 3KB gzipped. No model weights shipped to the browser. Hybrid keyword + vector scoring.

Resumable

Interrupted indexing resumes from the last checkpoint. Batch-based output means partial indexes are always usable.

Quick start

Standalone
docmd plugin
# Install globally
npm install -g docmd-search

# Install embedding dependencies (one-time)
npm install -g @huggingface/transformers onnxruntime-node

# Index any directory
docmd-search ./my-folder

# Or launch web UI (requires docmd)
docmd-search ./my-folder --ui

# In your docmd project
npm install docmd-search

Then enable semantic search in your config:

// docmd.config.js
export default {
  plugins: {
    search: {
      semantic: true,  // ← activates docmd-search
    }
  }
};

On first run, a setup wizard lets you choose an embedding model. After that, files are crawled, chunked, embedded, and search is ready.

What’s next

Page Description
Getting Started Installation, first run, and model selection
Configuration Global, project, and CLI configuration options
How It Works Architecture, chunking, compression, and scoring
CLI Reference All commands, flags, and examples
Programmatic API Use docmd-search in scripts and CI pipelines
Browser Client Integrate search into any web page

Architecture: two independent tools

docmd-search and docmd are completely independent projects from the same brand (docmd.io). They can work together but neither depends on the other.

┌─────────────────────────────────────────────────────────────────────┐
│                        docmd-search (standalone)                    │
│                                                                     │
│  CLI → Index directory → .docmd-search/ batches → TUI search       │
│                              │                                      │
│                              │ --ui flag?                           │
│                              ▼                                      │
│                    Spawn docmd with config                          │
│                    (docmd is just the UI shell)                     │
└─────────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────────┐
│                        docmd (documentation engine)                 │
│                                                                     │
│  Config → Build docs → plugin-search runs                           │
│                              │                                      │
│                              │ semantic: true?                      │
│                              ▼                                      │
│                    Import docmd-search, build index                 │
│                    (docmd-search is just the indexer)               │
└─────────────────────────────────────────────────────────────────────┘

When docmd-search launches docmd as its UI (--ui flag), it:

  1. Builds the semantic index first (its own job)
  2. Generates a docmd config pointing at the pre-built index
  3. Spawns docmd - which reads the index but never calls back to docmd-search

When docmd uses docmd-search as a plugin (semantic: true), it:

  1. Dynamically imports docmd-search at build time
  2. Runs the indexer to build .docmd-search/ batches
  3. Serves the semantic client bundle - no runtime dependency on docmd-search
next Getting Started