How do I install @wei840222/qmd?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does @wei840222/qmd need an API key?

No required environment variable is listed on this record.

What transport does @wei840222/qmd use?

@wei840222/qmd runs locally over stdio.

When was @wei840222/qmd last checked?

Last checked 2026-11-10. Hybrid search for markdown with local BM25 and remote embedding, expansion, and reranking APIs

Máy chủ MCP @wei840222/qmd

QMD - Query Markup Documents

QMD indexes Markdown notes, meeting transcripts, documentation, and source files. Search and retrieve them from the terminal, an MCP client, or an application.

Documents and keyword/CJK search stay local in SQLite. Semantic search, query expansion, and reranking use optional remote APIs; QMD does not download or run local inference models.

Start with Quick Start, Architecture, Configuration, or MCP Server. For an existing installation, read Upgrading before opening an old index with a new build. Changes are recorded in the changelog.

Quick Start

Install QMD, add a folder, and search it with keywords. This path needs no model configuration or API key.

pnpm add --global @wei840222/qmd
qmd collection add ~/notes --name notes
qmd context add qmd://notes "Personal notes and ideas"
qmd search "project timeline" -c notes

Search results include a file path and a short document ID. Retrieve the document or a section of it:

qmd get "notes/meeting.md"
qmd get "#abc123"
qmd get "notes/meeting.md:50:40"

For semantic or hybrid search, first configure your remote model APIs, then explicitly build embeddings:

qmd embed
qmd vsearch "how to deploy" -c notes
qmd query "quarterly planning process" -c notes

vsearch sends the query to the configured embedding provider. query can also use remote expansion and reranking. Use typed lexical queries with reranking disabled when an MCP client needs model-free search.

Architecture

QMD separates the local index from optional model services:

Part	Responsibility
Local index	SQLite stores document content, collection/path metadata, BM25/CJK indexes, and vectors.
Remote models	Explicit embedding operations send document chunks to the configured provider. Model-backed searches can send query text and candidate text for expansion, embedding, or reranking.
Interfaces	The CLI and SDK manage the index and retrieve documents. MCP exposes retrieval tools and opens existing indexes read-only.

Choose a search mode according to the information you have:

Command	How it searches	Model API needed
`qmd search`	Keywords through BM25, with local CJK analysis	No
`qmd vsearch`	Similarity against stored document vectors	Remote query embedding
`qmd query`	Keyword and vector retrieval, rank fusion, optional expansion/reranking	Remote embeddings and any enabled expansion/reranking providers

The hybrid pipeline combines those stages:

flowchart LR
  Q[User Query] --> X[Query Expansion]
  Q --> FTS[BM25 Search]
  Q --> VS[Vector Search]
  JEV[Jev Intent Hints] -.-> X
  X --> HYDE[HyDE]
  X --> VEC[Vec dense sentences]
  X --> LEX[Lex BM25 keywords]
  HYDE --> VS
  VEC --> VS
  LEX --> FTS
  VS --> RRF[Reciprocal Rank Fusion]
  FTS --> RRF
  RRF --> RR[Reranker]
  API[rerank_api_*] --> RR
  JR[Jev] -.->|none configured or failed| RR
  RR --> OUT[Final ranked results]

Typed lex searches use BM25; vec and hyde use the vector index. Hybrid results are combined with reciprocal rank fusion before optional reranking. When configured, Jev supplies intent hints to the remote generator. A configured rerank API is preferred; Jev can rerank when that API is absent or fails.

Vectors are partitioned by collection, so scoped searches scan that collection's vectors. Adding an embedded document to another collection reuses its stored vectors; a missing chunk is embedded again on its own. Explicit cleanup compacts sparse vector chunks without contacting a model. See Index Maintenance for commands and Score Normalization & Fusion for ranking details.

Requirements

System Requirements

Node.js >= 26.0.0
macOS: Homebrew SQLite (for extension support)
```
brew install sqlite
```

Remote-only runtime

QMD runs on Node.js with explicitly configured remote model APIs. There is no bundled or automatically downloaded inference model and no GPU runtime to configure. BM25 (qmd search), document reads, collection/context management, and maintenance work without model configuration. Semantic operations require remote model APIs. qmd pull is retired: it prints migration guidance and exits nonzero without network access or downloads.

Existing local-model databases remain readable through BM25 and document tools. QMD never re-embeds old vectors on its own or silently sends their documents to a provider. The automatic layout upgrade described below can discard vectors and chunk metadata in the cases listed there. To use remote vectors, configure the desired remote model and run an explicit qmd embed --force after reviewing the provider and data boundary. This clears and rebuilds vectors as a user-requested operation; lexical content is retained.

Installation

pnpm add --global @wei840222/qmd

Development

git clone https://github.com/wei840222/qmd
cd qmd
pnpm install --frozen-lockfile
pnpm run build
pnpm qmd --version
pnpm qmd --help

For contributor workflows, source locations, and validation commands, see AGENTS.md.

Configuration

Collections, contexts, and optional model settings live in index.yml. Start with example-index.yml; its model section is commented out so keyword search works without enabling remote requests.

Configuring `index.yml`

The collection and context commands read and write a single YAML config file — you can also edit it directly. Everything QMD knows about your collections (paths, masks, exclusions, per-collection update hooks, contexts, and optional model overrides) lives here. A fully-commented starter template ships as example-index.yml in this repo.

Location: ~/.config/qmd/index.yml by default. The directory honors XDG_CONFIG_HOME (→ $XDG_CONFIG_HOME/qmd/index.yml) and QMD_CONFIG_DIR. A named index uses {name}.yml — qmd --index work … reads/writes work.yml. A project-local index created with qmd init lives at .qmd/index.yml (.qmd/index.yaml is also accepted) alongside a project-local index.sqlite, so config and index stay inside the project instead of ~/.config / ~/.cache.

# ~/.config/qmd/index.yml

# Context applied to every collection (system-message style). Optional.
global_context: "Knowledge base for my projects"

# Terminal hyperlink template for search results. Optional.
# Overridden by the QMD_EDITOR_URI env var. See "Editor Links" below.
editor_uri: "vscode://file{path}:{line}:{col}"

# Remote models are optional for BM25 and document reads.
models:
  embed_api_url: https://api.openai.com/v1
  embed_api_model: text-embedding-3-small
  embed_dimension: 1536
  generate_api_url: https://api.example.com/v1
  generate_api_model: your-expansion-model
  rerank_api_url: https://api.example.com/v1/rerank
  rerank_api_model: your-reranker

# One entry per collection. The key is the collection name.
collections:
  notes:
    path: /Users/me/notes        # absolute path to index (required)
    pattern: "**/*.md"           # glob mask (default: **/*.md)
    ignore:                      # glob patterns to exclude from indexing
      - "Archive/**"
      - "**/drafts/**"
    update: "git pull --rebase"  # bash command run before each `qmd update`
    includeByDefault: true       # include in unscoped queries (default: true)
    context:                     # path prefix → description; longest match wins
      "/": "Personal notes and ideas"
      "/work": "Work-related notes"

Key	Scope	Purpose
`global_context`	top-level	Context prepended for every collection. Set via `qmd context add /`.
`editor_uri` (alias `editor_uri_template`)	top-level	Hyperlink template for clickable result paths; `QMD_EDITOR_URI` overrides.
`models.embed` / `.rerank` / `.generate`	top-level	Legacy model identity fields retained for diagnostics; local inference is retired. Configure remote `_api_` fields instead.
`collections.<name>.path`	per-collection	Absolute directory to index.
`collections.<name>.pattern`	per-collection	Glob mask. Set via `qmd collection add --mask`. Default `*/.md`. Comma-separated lists and brace groups (`{a,b}`) are a union of patterns.
`collections.<name>.ignore`	per-collection	Glob patterns excluded from indexing — useful to stop nested collections double-indexing. YAML-only — no CLI command sets this. Additive with QMD's built-in exclusions (`node_modules`, `.git`, `.cache`, `vendor`, `dist`, `build`), which you cannot un-ignore.
`collections.<name>.update`	per-collection	Bash command run before `qmd update` re-indexes this collection. Set via `qmd collection update-cmd`.
`collections.<name>.includeByDefault`	per-collection	Whether unscoped queries search it. Toggle with `qmd collection include`/`exclude`. Default `true`.
`collections.<name>.context`	per-collection	Path-prefix → description map; the most specific (longest) matching prefix wins. Set via `qmd context add`.

Note: Editing index.yml changes which directories and models QMD uses, but does not re-index on its own. Run qmd update after changing path, pattern, or ignore. Changing remote embedding identity requires the explicit whole-index rebuild described above; legacy models.embed no longer enables inference.

Automatic update commands

A collection's update field is QMD's built-in refresh hook: when you run qmd update, each collection's update command runs first, then the collection is re-indexed. This keeps a collection in sync with an upstream source (a git remote, a sync script) without wrapping qmd yourself.

collections:
  wiki:
    path: ~/reference/wiki
    update: "git pull --ff-only"

$ qmd update
[1/3] wiki (**/*.md)
    Running update command: git pull --ff-only
    Already up to date.
Collection: ~/reference/wiki (**/*.md)
Indexed: 0 new, 2 updated, 340 unchanged, 0 removed

The command runs via bash -c in the collection's own directory (its path), not your current working directory. If it exits non-zero, qmd update prints the failure and aborts the entire run — collections after the failing one are not re-indexed. Set or clear it from the CLI instead of editing YAML by hand:

qmd collection update-cmd wiki 'git pull --ff-only'   # set
qmd collection update-cmd wiki                         # clear

Checked-in `.qmd` config is not trusted by default

A project-local .qmd/index.yml travels with a git clone, and QMD adopts it automatically for any command run inside the tree. Three fields in that file can reach outside the project, and QMD will not use them unattended:

update commands — somebody else's shell script, run by qmd update
collections.*.path pointing outside the project directory
models.embed / models.rerank / models.generate other than the built-in defaults (legacy model identity fields)

In-project collection paths (for example ./docs) still index. On a terminal qmd update (and qmd embed / qmd query) lists the gated fields and asks. Approving records the approval in ~/.config/qmd/trusted.json. With no terminal to ask — agents, CI, MCP — those fields are skipped and in-project indexing continues.

Approvals cover the exact gated set you saw. Editing a command, pointing a collection outside the project, or changing a custom model URI asks again.

qmd trust           # review and approve this project's gated fields
qmd trust list      # show every approved project config
qmd trust revoke    # drop the approval for this project

Set QMD_TRUST_LOCAL_CONFIG=1 (or QMD_TRUST_UPDATE_HOOKS=1) for CI that should allow them unattended. Your own ~/.config/qmd/*.yml — including anything qmd collection update-cmd or qmd collection add writes — is never gated.

Remote Embedding & Models Configuration (Explicit Opt-in)

Configure remote embeddings and remote LLM models in index.yml under the unified models: section. Configuration alone sends no requests; embedding builds and model-backed searches are explicit operations.

models:
  # Remote embeddings: set both embed_api_url (or embed_url / embed_base_url) and embed_api_model.
  # Provider is auto-detected from the model name / URL (voyage* → voyageai, otherwise openai-compatible).
  # Optional override: embed_provider: voyageai | openai
  embed_api_url: https://api.openai.com/v1
  embed_api_model: text-embedding-3-small           # or text-embedding-3-large / voyage-4 / ...
  embed_dimension: 1536                             # Optional; required for unknown custom OpenAI models

  # Optional: Remote LLM Query Expansion (aliases: generate_url, generate_base_url, generate_api_url)
  generate_api_url: https://your-model-gateway.example/v1/chat/completions
  generate_api_model: your-expansion-model
  # generate_api_key: <key>  # required if this endpoint authenticates requests

  # Optional: Remote Reranking (aliases: rerank_url, rerank_base_url, rerank_api_url)
  rerank_api_url: https://your-reranker.example/v1/rerank  # Cohere-shaped `results`
  rerank_api_model: your-reranker
  # rerank_api_key: <key>    # required if this endpoint authenticates requests
  # For LLM scoring, use a /v1/chat/completions endpoint instead.

  # Optional: TypeSafe Jev. Guides query expansion; reranks only when no rerank_api_* is configured, or as a fallback.
  # jev_api_key: <key>                              # or TYPESAFE_API_KEY
  # jev_base_url: https://jev.example.com           # or TYPESAFE_BASE_URL
  # jev_api_model: jev-1.13                         # or TYPESAFE_DEFAULT_MODEL

# Optional: Custom User Dictionary for CJK segmentation
dictionary: ~/.config/qmd/dictionary.txt

OpenAI-compatible embeddings

models:
  embed_api_url: https://api.openai.com/v1
  embed_api_model: text-embedding-3-small   # or text-embedding-3-large
  # embed_dimension: 1536                   # optional for known models; required for unknown custom models

export OPENAI_API_KEY="..."   # or models.embed_api_key in a private config
qmd embed

Voyage AI embeddings

Voyage is a first-class provider (voyageai). Model names starting with voyage (or an explicit embed_provider: voyageai) select it. Known models resolve default dimensions automatically (voyage-4 / voyage-4-large / voyage-3 → 1024, voyage-3-lite / voyage-3.5-lite → 512, etc.). Voyage requests use native output_dimension and input_type (query vs document); they do not send OpenAI encoding_format.

Official Voyage endpoint:

models:
  embed_api_url: https://api.voyageai.com/v1   # optional when embed_provider: voyageai (default base URL)
  embed_api_model: voyage-4                    # or voyage-4-large, voyage-3, voyage-code-3, ...
  # embed_provider: voyageai                   # optional; inferred from voyage* model names
  # embed_dimension: 1024                      # optional override / Matryoshka truncation when supported

export VOYAGE_API_KEY="..."   # or models.embed_api_key in a private config
qmd embed

OpenAI-compatible proxy hosting Voyage (Bifrost, LiteLLM, etc.):

models:
  embed_api_url: https://your-proxy.example/v1
  embed_api_model: voyage-4
  # embed_provider: voyageai   # set if the model name alone is not enough to infer Voyage

# Prefer VOYAGE_API_KEY. OPENAI_API_KEY is accepted only for non-official Voyage base URLs
# (custom proxies). Official api.voyageai.com never falls back to OPENAI_API_KEY.
export VOYAGE_API_KEY="..."
# export OPENAI_API_KEY="..."  # proxy-only fallback when VOYAGE_API_KEY is unset
qmd embed

URL fields: choose one alias per operation: *_url takes precedence over *_base_url, then *_api_url. Embedding URLs are base URLs: the provider appends /embeddings. Expansion and reranking also accept complete endpoints ending in /chat/completions or /rerank; a base URL gets /chat/completions for expansion or /rerank for reranking. Configure the full /chat/completions URL explicitly for chat scoring; QMD does not discover a sibling chat endpoint automatically. Dedicated rerank endpoints must return results[].{index,relevance_score}. Voyage's rerank API returns data, which needs a gateway that translates it to this shape.

Credentials: OpenAI-compatible embeddings use models.embed_api_key, then OPENAI_API_KEY. Voyage uses models.embed_api_key, then VOYAGE_API_KEY, with OPENAI_API_KEY accepted only for custom Voyage proxies. Expansion and reranking use their own models.generate_api_key and models.rerank_api_key; they do not inherit OPENAI_API_KEY. An unset key omits the Authorization header, for keyless gateways. YAML does not interpolate ${ENV_VAR} placeholders. Keep actual keys out of checked-in files; use the supported embedding/Jev environment variables or an explicitly managed private config.

CLI, SDK & MCP Integration: Remote LLM query expansion and LLM Chat Reranking are automatically wired into CLI (qmd query), SDK (createStore), and MCP. Query expansion automatically enforces query language & script consistency (e.g., Traditional Chinese queries generate Traditional Chinese lex, vec, and hyde variations). LLM Chat Reranking features prompt-tail Recency Enforcement and strict JSON sanitization to ensure safe execution with any remote LLM backend.

qmd init writes index and collection defaults only; it does not prompt for or generate model configuration. To use a remote endpoint, edit index.yml manually and set credentials when required.

API Key Note: Remote embeddings require both an embedding endpoint (embed_url, embed_base_url, or embed_api_url) and embed_api_model (unless embed_provider: voyageai supplies the Voyage default model/URL). Official OpenAI needs models.embed_api_key or OPENAI_API_KEY; official Voyage needs models.embed_api_key or VOYAGE_API_KEY. Custom OpenAI-compatible proxies may omit a key when the upstream is keyless. For Voyage-shaped traffic through a custom proxy only, OPENAI_API_KEY may fill in when VOYAGE_API_KEY is unset — never for https://api.voyageai.com/v1. QMD never writes keys to SQLite, diagnostics, logs, or error messages.

Configure embed_url, embed_base_url, or embed_api_url together with embed_api_model to enable remote embeddings. OpenAI-compatible providers receive POST /embeddings; Voyage AI uses its native embeddings payload (input_type, optional output_dimension). Changing the endpoint, provider, or model changes the remote embedding identity and requires a vector rebuild.

Remote document embeddings use a bounded memory cache by default, shared by stores using the same loaded QMD module instance. To share results across processes, plugin reloads, and process restarts, add an optional cache directory to the existing models configuration:

models:
  embed_api_url: https://api.openai.com/v1
  embed_api_model: text-embedding-3-small
  embed_cache_dir: ~/.cache/qmd/embeddings
  # Linux tmpfs alternative: /dev/shm/qmd-embeddings

The SDK accepts the same field in createStore({ config: { collections: ..., models: ... } }); callers such as plugins must forward it into QMD's config. Paths support ~. Relative paths resolve from the YAML directory, except project-local .qmd/index.yml, where they resolve from the project root; inline SDK paths resolve from the working directory. Project-local cache paths outside the project follow the existing qmd trust policy. Omitting the field keeps memory-only caching; a blank value is invalid.

With a directory configured, OpenAI-compatible and Voyage providers use document-embeddings-v1.sqlite as the authoritative document cache. Processes pointing to the same local directory share atomic writes and a crash-released SQLite lock, including separate OpenClaw plugin generations. Cache-miss batches in one directory are serialized; use separate directories for unrelated workloads if independent throughput matters. A normal disk retains the cache across restarts; tmpfs works too, but its contents disappear on reboot and consume RAM. QMD creates the directory as needed and does not mount tmpfs.

Both modes reuse results for up to two hours, with at most 2,000 entries and a 64 MiB payload budget (SQLite files and journals add overhead). Reuse requires matching provider configuration, credentials, build identity, and exact formatted input. Memory mode also separates different fetch transports; disk mode deliberately shares across transports and processes using that configured directory. Only vectors and hashed keys are persisted, never input text or raw API keys. Each store retains its own authorization, build lease, and index. Forced rebuilds bypass reuse and refresh the shared vectors. Query embeddings keep their separate cache, and read-only SDK stores do not open the disk document cache.

Corrupt individual entries are recomputed. An unavailable or corrupt cache database emits a warning and falls back to normal computation without deleting the file or blocking indexing. Expiration, eviction, forced rebuilds, failed requests, and crash recovery can require new API calls; this is deduplication, not an exactly-once billing guarantee.

The vector index can store only one dimension. A dimension or identity change requires a forced rebuild:

qmd embed --force

Remote document chunks use deterministic UTF-8 byte windows. Query embedding is allowed only when the same identity is ready and compatible vectors exist; a vector or hybrid query then sends the formatted query text to the configured remote provider. After a vector reset deletes the old vectors, there is no vector rollback; lexical search remains available while rebuilding.

Embedding identity, recovery, and data boundary

The vector index holds one vector dimension at a time. An embedding identity includes the provider, model, dimension, local/remote mode, and prompt/chunking profiles; changing it requires qmd embed --force. A build lease with an owner ID and generation sequence prevents concurrent writers. If an embedding request fails or the lease expires, the index enters Partial; rerunning qmd embed with the same identity resumes committed work and requests only missing, vector-only, metadata-only, or layout-incomplete chunks. Vector search reads only ready vectors with a compatible fingerprint.

Remote document chunks and formatted query text are sent to the configured remote embedding provider. Remote embedding failures never fall back to a local embedding provider, because vectors from a different identity are incompatible. API keys follow the per-operation config/environment rules above and are never stored in SQLite, diagnostics, logs, or errors. Errors redact response bodies, request payloads, credentials, and sensitive URLs; identity and lease state are revalidated inside SQLite write locks before vectors are cleared or published.

Remote LLM routing is independent per operation: expansion and reranking may each use separately configured remote endpoints. Remote failures do not fall back to local inference; a dedicated rerank URL returning HTTP 404 retries with a chat-scoring payload at the same URL. It does not rewrite the URL to /chat/completions; configure that endpoint explicitly when the backend expects chat scoring. This behavior does not apply to remote embeddings.

qmd status and qmd doctor report provider/model/dimension, short fingerprint, key presence as a boolean, pending/inconsistent chunks, CJK channel readiness, and actionable repair commands without loading local models or issuing remote HTTP requests. qmd status reads the existing index without migration. qmd doctor reconciles a valid YAML config into an existing database and can initialize its schema, including vector layout conversion; follow the upgrade backup procedure before running it on an older index.

Expansion and reranking need their own URL/model settings. Jev supplies intent hints, not a replacement generator. A failed API request does not permanently disable that endpoint; later requests can retry. When Jev is available as the rerank fallback, QMD waits 60 seconds by default before trying a failed rerank API again. Cancellation does not start a fallback request.

Usage

Collection Management

# Create a collection from current directory
qmd collection add . --name myproject

# Create a collection with explicit path and custom glob mask
qmd collection add ~/Documents/notes --name notes --mask "**/*.md"

# Comma-separated masks are a union (brace form `{a,b}` also works)
qmd collection add ~/notes --name notes --mask "sources/**/*.md,CO - *.md"

# List all collections
qmd collection list

# Remove a collection
qmd collection remove myproject

# Rename a collection
qmd collection rename myproject my-project

# List files in a collection
qmd ls notes
qmd ls notes/subfolder

# Show collection details (path, glob mask, include status, context count, top metadata keys)
qmd collection show notes

# Discover metadata keys, types, and value counts (see Metadata Discovery)
qmd collection metadata notes
qmd collection metadata notes --match '{"field":"key","operator":"eq","value":"topics"}'
qmd collection metadata notes --match '{"field":"value","operator":"eq","value":"docs-team"}'

# Include or exclude a collection from default (unscoped) queries
qmd collection include notes
qmd collection exclude notes

# Run a command before every `qmd update` (e.g. git pull); empty arg clears it
qmd collection update-cmd notes 'git pull --rebase'
qmd collection update-cmd notes

Generate Vector Embeddings

# Embed all indexed documents (900 tokens/chunk, 15% overlap)
qmd embed

# Force re-embed everything
qmd embed -f

# Enable AST-aware chunking for code files (TS, JS, Python, Go, Rust)
qmd embed --chunk-strategy auto

# Also works with query for consistent chunk selection
qmd query "auth flow" --chunk-strategy auto

# Memory control for large corpora / constrained systems
qmd embed --max-docs-per-batch 50   # cap docs per embedding batch
qmd embed --max-batch-mb 64         # cap batch size in MB

AST-aware chunking (--chunk-strategy auto) uses tree-sitter to chunk code files at function, class, and import boundaries instead of arbitrary text positions. This produces higher-quality chunks and better search results for codebases. Markdown and other file types always use regex-based chunking regardless of strategy.

The default is regex (existing behavior). Use --chunk-strategy auto to opt in. Run qmd status to verify which grammars are available.

Note: Tree-sitter grammars are optional dependencies. If they are not installed, --chunk-strategy auto falls back to regex-only chunking automatically.

Context Management

Context adds descriptive metadata to collections and paths, helping search understand your content.

# Add context to a collection (using qmd:// virtual paths)
qmd context add qmd://notes "Personal notes and ideas"
qmd context add qmd://docs/api "API documentation"

# Add context from within a collection directory
cd ~/notes && qmd context add "Personal notes and ideas"
cd ~/notes/work && qmd context add "Work-related notes"

# Add global context (applies to all collections)
qmd context add / "Knowledge base for my projects"

# List all contexts
qmd context list

# Remove context
qmd context rm qmd://notes/old

Search Commands

┌──────────────────────────────────────────────────────────────────┐
│                        Search Modes                              │
├──────────┬───────────────────────────────────────────────────────┤
│ search   │ BM25 full-text search only                           │
│ vsearch  │ Vector semantic search only                          │
│ query    │ Hybrid: FTS + Vector + Query Expansion + Re-ranking  │
└──────────┴───────────────────────────────────────────────────────┘

# Full-text search (fast, keyword-based)
qmd search "authentication flow"

# Vector search (semantic similarity)
qmd vsearch "how to login"

# Hybrid search with re-ranking (best quality)
qmd query "user authentication"

Two aliases exist for the semantic/hybrid modes: vector-search (→ vsearch) and deep-search (→ query).

Query Syntax

Use a plain query when you want QMD to choose search variants, or supply typed queries when you want to control keyword and semantic retrieval yourself.

qmd search '"rate limiter" -sports'                 # model-free keyword search
qmd query "how does auth work"                     # automatic expansion policy
qmd query --expand "how does auth work"            # force remote expansion
qmd query "lex: authentication"                    # skip expansion; other remote stages can still run

For plain qmd query input, the expansion policy is:

Input	Policy	Behavior
Unprefixed text	`auto`	A strong keyword match can skip expansion. CJK queries skip expansion when no expansion-capable service is configured; with one configured, they can expand. Context and relative-time wording can prevent the strong-match shortcut.
`expand: text` or `--expand`	`force`	Request expansion even when an automatic shortcut would apply. Fail if the service cannot produce usable variants.
Standalone `lex: text`	`skip`	Skip expansion, while retaining original-query keyword retrieval and vector retrieval when a searchable vector index exists.

Skipping expansion is not an offline guarantee: vector retrieval can still call the embedding API, and reranking is enabled by default. --no-rerank disables reranking only. Use qmd search for model-free CLI search.

Keyword operators work in qmd search and typed lex queries:

Syntax	Meaning	Example
`word`	Prefix match; positive terms are AND-combined	`perf` matches `performance`
`"phrase"`	Match consecutive tokens in order	`"rate limiter"`
`-word`	Exclude a term	`auth -oauth`
`-"phrase"`	Exclude a phrase	`auth -"test data"`

Include at least one positive term; a query containing only exclusions returns no matches. Separator-containing terms such as multi-agent and src/lib/i18n.ts match adjacent tokens rather than a word prefix. CJK uses character-phrase matching instead of ordinary word-prefix matching.

Typed query documents bypass automatic expansion. Each non-empty line names the retrieval method and the text to search:

Type	Supply	Retrieval
`lex:`	Keywords, identifiers, or quoted phrases	Local BM25
`vec:`	A natural-language question or description	Remote embedding and vector search
`hyde:`	A hypothetical answer passage	Remote embedding and vector search

# Bash/Zsh: $'...' turns \n into actual newlines
qmd query $'lex: rate limiter\nvec: how does the API limit burst traffic'
qmd query 'hyde: The API uses a token bucket to limit bursts and returns HTTP 429 when the quota is exhausted.'

# Optional intent provides context for reranking
qmd query $'intent: web page load times\nlex: performance\nvec: how to improve performance'
qmd query --intent "web page load times" "performance"

Put your strongest search first: it receives 2x weight in structured rank fusion. vec and hyde need remote embedding configuration and a compatible vector index; reranking still defaults to enabled. A single vec: or hyde: line is also a typed document; a single lex: line uses the plain-query skip policy above.

Every non-empty document line must use lex:, vec:, hyde:, or intent: with non-empty text. Blank lines are ignored and surrounding whitespace is trimmed; a final newline is optional.
At most one intent: line is allowed, in any position, alongside at least one search line. CLI --intent overrides inline intent and supplies reranking context; intent is not a search term.
Do not mix plain text or expand: with typed document lines. --expand with a standalone lex: policy is an error.
Keep each search on one line. Lexical quotes must be balanced; -term and -"phrase" exclusions belong in lexical queries, not vec or hyde text.

Use -c docs -c notes to scope collections; see Collection Filtering. For MCP, send typed lines as a searches array rather than a multiline query string. The MCP parameter reference covers the tool schema; the bundled agent query reference covers validation, context precedence, and request examples.

Options

# Search options
-n <num>           # Number of results (default: 5, or 20 for --files/--json)
-c, --collection   # Restrict search to a specific collection
--all              # Return all matches (use with --min-score to filter)
--min-score <num>  # Minimum score threshold (default: 0)
--full             # Show full document content
--line-numbers     # Add line numbers to output
--explain          # Include retrieval score traces (query, JSON/CLI output)
--filter <json>    # Metadata filter (recursive JSON AST; see Metadata Filtering)
--index <name>     # Use named index
--intent "<text>"  # Background context for ambiguous queries, same as an `intent:` line
                   # (CLI query reranking context; e.g. "web page load times")
--no-rerank        # Skip remote reranking (RRF scores only)
-C, --candidate-limit <n>  # Max candidates to rerank (default: 40)
--full-path        # Emit on-disk filesystem paths instead of qmd:// URIs
                   # (a result whose file has moved or been deleted since
                   #  indexing keeps its qmd:// URI + docid, and a notice is
                   #  printed to stderr — run `qmd update` to refresh)

# Output formats (for search and multi-get)
--format <kind>    # cli (default) | json | csv | md | xml | files
                   # (--json, --csv, --md, --xml, --files are legacy aliases)

# Get options
qmd get <file>[:from[:count]]  # Get document; optional start line and count
-l <num>                       # Maximum lines to return
--from <num>                   # Start line (overrides the :from suffix)
--no-line-numbers              # Disable line numbering (on by default)

# Multi-get options
-l <num>           # Maximum lines per file
--max-bytes <num>  # Skip files larger than N bytes (default: 64KB)

Collection Filtering

The -c/--collection flag filters results by collection name (as shown by qmd collection list). Collections are a global registry — you can search any collection from any directory:

qmd search "auth" -c notes           # single collection
qmd search "auth" -c notes -c docs   # multiple collections (OR)

With no -c flag, all default-included collections are searched. Collections marked excluded (qmd collection exclude <name>) are skipped unless named explicitly with -c.

Note: Multiple -c flags are combined with OR and applied before each lexical and vector candidate cutoff. Matching candidates from the selected collections are then ranked together.

Metadata Filtering

Documents can opt into typed metadata through a namespaced frontmatter block. A document without qmd.metadata behaves exactly as before, and the frontmatter stays ordinary searchable content (no chunking, embedding, or line-number changes):

---
qmd:
  metadata:
    topics:
      - typescript
      - programming
    status: published
    priority: 3
    reviewed: true
---

# Document body starts here

Supported values are strings, numbers, booleans, and flat homogeneous arrays of one of those. Nested objects, nulls, empty arrays, and mixed-type arrays are rejected (the document still indexes; it is excluded from filtered search until corrected). Metadata keys are user-defined data — tags, topics, and labels are all ordinary keys with no special semantics.

Every search surface (CLI, SDK, MCP, HTTP) accepts the same recursive filter, a JSON AST discriminated by operator. A condition is a predicate over one field of the document's metadata: field names the metadata key, operator says how to compare, and value is what to compare against:

# One condition
qmd search "authentication" \
  --filter '{"field":"status","operator":"eq","value":"published"}'

# Composed conditions — works with search, vsearch, and query
qmd query "dependency injection" --filter '{
  "operator": "and",
  "operands": [
    { "field": "topics", "operator": "all", "value": ["typescript", "programming"] },
    { "field": "status", "operator": "nin", "value": ["draft", "archived"] },
    { "operator": "or", "operands": [
      { "field": "priority", "operator": "gte", "value": 3 },
      { "field": "reviewed", "operator": "eq", "value": true }
    ] },
    { "field": "topics", "operator": "prefix", "value": "sql", "caseInsensitive": true },
    { "operator": "not", "operand": { "field": "audience", "operator": "eq", "value": "internal" } }
  ]
}'

Node	Shape
Logical group	`{ "operator": "and" \| "or", "operands": […] }`
Negation	`{ "operator": "not", "operand": {…} }`
Comparison	`{ "field", "operator": "eq" \| "ne" \| "gt" \| "gte" \| "lt" \| "lte", "value" }`
Membership	`{ "field", "operator": "in" \| "nin" \| "all", "value": […] }`
Text	`{ "field", "operator": "contains" \| "prefix" \| "suffix", "value": "…" }`
Type	`{ "field", "operator": "type", "value": "string" \| "number" \| "boolean" }`
Presence	`{ "field", "operator": "exists", "value": true \| false }`

Any condition whose value is a string or an array of strings may add "caseInsensitive": true.

Semantics:

Matching is typed and exact — no string/number/boolean coercion, and a type mismatch never matches (including ne and nin). Text operators match string values only. type matches the stored type of a key's values, which is how a filter reaches one side of a key whose documents disagree on type.
Array-valued metadata is a set: a condition matches when any element satisfies it, all requires every filter value to be present, and ne/nin require that no element equals the operand (with at least one element of the operand's type present).
Missing keys do not match ne/nin; combine with { "operator": "exists", "value": false } in an or group to include them.
Matching is case-sensitive unless a condition sets caseInsensitive, which folds ASCII letters on both sides. Non-ASCII letters compare exactly.
Multiple conditions require an explicit and group — there is no implicit AND, and no $-prefixed shorthand.

Guarantees and limits:

Every returned result satisfies the filter, before RRF fusion and reranking.
Highly selective filters can return fewer than limit results. Lexical search filters a bounded over-fetch window; vector search scans the filter-eligible set exactly when it holds at most 20,000 chunk vectors, and above that over-fetches and post-filters.
Filtered search only considers documents whose metadata has been extracted (run qmd update after upgrading; qmd status shows the pending count).

JSON output (--format json), the SDK, MCP structured results, and the HTTP endpoints include each result's indexed metadata.

Metadata Discovery

Filtering is only useful if you know what to filter on. Discovery reports the metadata keys, types, and value counts already in the index, so a filter can be written from what is indexed instead of guessed. It reads the same tables filtering reads: no re-indexing, and every value it reports is one an eq filter can match.

The command is one sentence: show metadata matching X for documents filtered by Y. --filter narrows documents by their metadata (same AST as search) and decides which are counted. --match narrows the metadata itself and decides which entries are reported. It takes the same AST, evaluated against each metadata entry instead of each document: a condition's field is "key" (the entry's key name) or "value" (its value). Every operator applies, including type, the text operators, caseInsensitive, and and/or/not. Only exists and all are rejected, as they have no meaning for a single entry.

`--match`	Question answered
	Which keys exist, with a window of values each
`{"field":"key","operator":"eq","value":"topics"}`	Everything about one key
`{"field":"key","operator":"prefix","value":"mem-"}`	A family of keys
`{"field":"key","operator":"in","value":["tags","topics","labels"]}`	Which of these key names exist
`{"field":"value","operator":"eq","value":"docs-team"}`	Which keys hold this value
`{"field":"value","operator":"prefix","value":"2025-"}`	Which keys hold values shaped like this
`{"field":"value","operator":"type","value":"boolean"}`	Which keys hold booleans
`and` of `key eq priority` and `value gte 3`	Values of one key above a threshold
`and` of `key eq priority` and `value type number`	The numeric side of a key whose documents disagree on type

Start wide and narrow:

# Which keys does this collection use? (`qmd collection show notes` previews the top five)
qmd collection metadata notes

# Everything about one key: coverage, distinct count, top values
qmd collection metadata notes --match '{"field":"key","operator":"eq","value":"topics"}'

# Reverse lookup: which keys hold this value
qmd collection metadata notes --match '{"field":"value","operator":"eq","value":"docs-team"}'

# Values of one key matching a pattern
qmd collection metadata notes --match '{
  "operator": "and",
  "operands": [
    { "field": "key", "operator": "eq", "value": "topics" },
    { "field": "value", "operator": "prefix", "value": "sql" }
  ]
}'

# What remains after a filter, before committing to it in a query
qmd collection metadata notes \
  --match '{"field":"key","operator":"eq","value":"topics"}' \
  --filter '{"field":"status","operator":"eq","value":"published"}'

# Then search with the filter you just validated
qmd query "dependency injection" -c notes --filter '{
  "operator": "and",
  "operands": [
    { "field": "status", "operator": "eq", "value": "published" },
    { "field": "topics", "operator": "all", "value": ["typescript"] }
  ]
}'

The drill-down prints one block per key, in coverage order:

qmd collection metadata notes

topics  string[]  388 of 480 documents  1,204 distinct
  typescript    140
  sqlite         92
  search         77
  architecture   61
  sqlite-vec     44
  mcp            39
  embeddings     35
  agents         31
  cli            28
  testing        26
1,194 more values, use --value-limit <n>, --value-offset <n>, or --all-values

priority  number  205 of 480 documents  5 distinct
  min 1  median 3  max 5
  1 (12)  2 (40)  3 (88)  4 (50)  5 (15)

reviewed  boolean  480 of 480 documents
  true 61  false 419

With --filter, the output opens with a filter: line stating how many documents pass, and every coverage count is measured against that population rather than the whole collection — the numbers a filtered search would see:

qmd collection metadata notes \
  --match '{"field":"key","operator":"eq","value":"topics"}' \
  --filter '{"field":"status","operator":"eq","value":"published"}' \
  --value-limit 5

filter: 312 of 480 documents

topics  string[]  260 of 312 documents  811 distinct
  typescript    104
  sqlite         70
  search         58
  architecture   40
  mcp            31
806 more values, use --value-limit <n>, --value-offset <n>, or --all-values

Two windows page the key and value lists. Each windowed list ends with the exact remainder and the flags that reach it:

--key-limit <n>      # Keys reported, in coverage order (default 50)
--key-offset <n>     # Keys skipped before the window
--all-keys           # Remove the key window
--value-limit <n>    # Values reported per key and type (default 10)
--value-offset <n>   # Values skipped per key and type, in --sort order
--all-values         # Remove the value window
--sort count|value   # Order values by document count (default) or by value
--min-count <n>      # Drop values held by fewer documents

Omitting the collection name covers the default collections, exactly as an unscoped search does.

Reading the output:

Counts are documents, not values. A document with topics: [a, b] contributes one to each. Coverage is "documents declaring this key", out of the documents the filter admits when there is one.
A bare string is exactly the value. A string whose bare form could be read as something else ("42", "", "a, b") prints as a JSON string, so paste it into a filter as the JSON string it is.
Numbers report min, median, and max, plus the enumerated values when they fit, which is enough to write a gt/lt threshold in one call.
Discovery sees exactly what filtering sees. Same extraction gate, same active-document rule, same collection scope, and every count in a result comes from one database snapshot. Documents still pending extraction are reported on stderr and excluded until qmd update runs.
Type conflicts are reported, not resolved. Metadata is validated one document at a time, so priority: 3 in one file and priority: high in another both index, within one collection or across several. Discovery splits such a key by type, each with its own document count and (across collections) its contributing collections:

qmd collection metadata --match '{"field":"key","operator":"eq","value":"priority"}'

priority  number | string  1,427 of 2,100 documents
  number    223 docs  min 1  median 3  max 5               notes, work
  string  1,204 docs  high (700), medium (380), low (124)  work

To report one side only, add a type condition on the value field. The filter that reaches exactly those documents is the same condition with the metadata key in field:

qmd collection metadata --match '{
  "operator": "and",
  "operands": [
    { "field": "key", "operator": "eq", "value": "priority" },
    { "field": "value", "operator": "type", "value": "number" }
  ]
}'
qmd query "release checklist" --filter '{"field":"priority","operator":"type","value":"number"}'

qmd collection list names each collection's top keys, qmd collection show <name> details the top five with a value preview, and qmd status summarizes how many keys and files carry metadata.

The CLI prints text and rejects unsupported format flags. Structured discovery is available through the SDK, MCP, and HTTP with the same options (collection, match, filter, keyLimit, keyOffset, valueLimit, valueOffset, sort, minCount) and the same result shape:

// SDK: one flat result, keys in coverage order, each split per type
const discovery = await store.listMetadata({
  collection: "notes",
  match: { field: "key", operator: "eq", value: "topics" },
  valueLimit: 5,
})
discovery.documents            // active documents in scope
discovery.totalKeys            // keys with a matching entry, before the key window
discovery.remainingKeys        // keys after the window
discovery.keys[0].types[0]     // { type, multiValued, documents, distinctValues, values, remainingValues, range?, collections }

// Narrowed by a filter: how many documents pass, and what is left to filter on
const published = await store.listMetadata({
  collection: "notes",
  filter: { field: "status", operator: "eq", value: "published" },
})
published.filteredDocuments    // the denominator for every coverage count in this result

// Page through a wide vocabulary
const nextPage = await store.listMetadata({ keyLimit: 50, keyOffset: 50 })

filter is a MetadataFilter and match is a MetadataMatch. Both are MetadataPredicate<Condition>, the one recursive grammar over the conditions each record admits, so a document-only condition (exists, all, or a metadata key as the field) is a type error in a match as well as a runtime one. An option outside its domain throws MetadataOptionError, and a filter and match that together bind more SQL parameters than one statement allows throw MetadataBindingBudgetError.

The MCP metadata tool takes the same options with collections spelled as on query, returns the CLI text plus the result as structuredContent, and the MCP status tool lists each collection's most covered keys so an agent's first call reveals that metadata exists. POST /metadata accepts the same body as the tool and returns the same result (400 on an invalid match, filter, or option).

Output Format

Default output is colorized CLI format (respects NO_COLOR env).

When stdout is a TTY, result paths are emitted as clickable terminal hyperlinks (OSC 8). Clicking a path opens the file in your editor using an editor URI template.

When stdout is not a TTY (for example piped to another command or redirected to a file), QMD emits plain text paths with no escape sequences.

TTY example:

docs/guide.md:42 #a1b2c3
Title: Software Craftsmanship
Context: Work documentation
Score: 93%

This section covers the **craftsmanship** of building
quality software with attention to detail.
See also: engineering principles

notes/meeting.md:15 #d4e5f6
Title: Q4 Planning
Context: Personal notes and ideas
Score: 67%

Discussion about code quality and craftsmanship
in the development process.

Configure the editor link target with QMD_EDITOR_URI (or editor_uri in config):

# VS Code (default)
export QMD_EDITOR_URI="vscode://file/{path}:{line}:{col}"

# Cursor
export QMD_EDITOR_URI="cursor://file/{path}:{line}:{col}"

# Zed
export QMD_EDITOR_URI="zed://file/{path}:{line}:{col}"

# Sublime Text
export QMD_EDITOR_URI="subl://open?url=file://{path}&line={line}"

Template placeholders:

{path} absolute filesystem path (URI-encoded)
{line} 1-based line number
{col} or {column} 1-based column number
Path: Collection-relative path (e.g., docs/guide.md)
Docid: Short hash identifier (e.g., #a1b2c3) - use with qmd get #a1b2c3
Title: Extracted from document (first heading or filename)
Context: Path context if configured via qmd context add
Score: Color-coded (green >70%, yellow >40%, dim otherwise)
Snippet: Context around match with query terms highlighted

Examples

# Get 10 results with minimum score 0.3
qmd query -n 10 --min-score 0.3 "API design patterns"

# Output as markdown for LLM context
qmd search --md --full "error handling"

# JSON output for scripting
qmd query --json "quarterly reports"

# Inspect how each result was scored (RRF + rerank blend)
qmd query --json --explain "quarterly reports"

# Use separate index for different knowledge base
qmd --index work search "quarterly reports"

The --explain flag attaches a score breakdown to each result: the FTS/vector backend scores plus the RRF fusion math (rank, weight, top-rank bonus) and every sub-query's contribution. Abbreviated:

{
  "docid": "#6c90f0",
  "score": 0.89,
  "file": "qmd://qmd/README.md",
  "explain": {
    "ftsScores": [0.892, 0.907],
    "vectorScores": [0.540, 0.484],
    "rrf": {
      "rank": 1,
      "weight": 0.75,
      "baseScore": 0.123,
      "topRankBonus": 0.05,
      "totalScore": 0.173,
      "contributions": [
        { "source": "fts", "queryType": "original", "query": "reranking",
          "rank": 1, "weight": 2, "backendScore": 0.892, "rrfContribution": 0.0328 }
      ]
    }
  }
}

Index Maintenance

# Show index status and collections with contexts
qmd status

# Re-index all collections. If a collection has a configured update command
# (e.g. `git pull`), it runs first — set one with `qmd collection update-cmd`.
qmd update

# Diagnose the install; can reconcile config and migrate an older index
qmd doctor

# Initialize a project-local index in the current directory
qmd init

# Get document by filepath (with fuzzy matching suggestions)
qmd get notes/meeting.md

# Get document by docid (from search results)
qmd get "#abc123"

# Get document starting at line 50, max 100 lines
qmd get notes/meeting.md:50 -l 100

# Read 40 lines starting at line 120 via the :from:count suffix (works with docids)
qmd get notes/meeting.md:120:40
qmd get "#abc123:120:40"

# get / multi-get are line-numbered by default; disable with --no-line-numbers
qmd get notes/meeting.md --no-line-numbers

# Get multiple documents by glob pattern
qmd multi-get "journals/2025-05*.md"

# Get multiple documents by comma-separated list (supports docids)
qmd multi-get "doc1.md, doc2.md, #abc123"

# Limit multi-get to files under 20KB
qmd multi-get "docs/*.md" --max-bytes 20480

# Output multi-get as JSON for agent processing
qmd multi-get "docs/*.md" --json

# Preview cleanup actions; writable initialization can still migrate an old index
qmd cleanup --dry-run

# Drop caches and orphans; repack sparse vector chunks within each partition
qmd cleanup

--dry-run skips cleanup deletion, repacking, FTS optimization, and vacuum; it does not make the CLI's store initialization read-only. Use the upgrade backup procedure before any writable open of an older index.

Cleanup repacks when the table's useful-chunk occupancy is below 90% and an older sparse chunk can move. Each chunk moves in its own transaction; rowids, partitions, vector bytes, chunk metadata, and embedding identity are preserved by the repack itself. Normal orphan cleanup still removes rows no active document needs. Live chunks with missing vectors stay pending for an explicit qmd embed; cleanup does not call a model or re-embed them.

SDK callers can use Maintenance.vectorLayout(), preview(), repackVectors(), and run() for the same layout and maintenance operations.

Benchmarking

Measure search quality across all four backends with qmd bench and a fixture file of queries with known-relevant documents.

From a git checkout, an example fixture and its test corpus ship in the repo:

# One-time setup (indexes the repo's test corpus into its own collection)
qmd collection add test/eval-docs --name eval-docs
qmd embed -c eval-docs

# Run the benchmark (table output)
qmd bench src/bench/fixtures/example.json

# JSON output for programmatic analysis
qmd bench src/bench/fixtures/example.json --json

The example fixture (src/bench/fixtures/example.json) and its test corpus (test/eval-docs/) exist only in a git checkout — they are not part of the published npm package. If you installed the package, write your own fixture (see below) against a collection you have already indexed:
qmd bench my-fixture.json -c my-collection

Each query runs against four backends, reporting precision@k, recall, MRR, and F1:

Backend	What it tests	LLM required
`bm25`	Keyword search only (FTS5)	No
`vector`	Semantic similarity only	Embedding model
`hybrid`	BM25 + vector fusion (no reranking)	Embedding model
`full`	Full pipeline with LLM reranking	All three models

Score interpretation: 1.00 = perfect (all expected docs in top results), 0.00 = complete miss. The example fixture typically shows bm25 ~0.50, vector ~0.70, and hybrid/full ~1.00 — a concrete demonstration of why hybrid search beats either backend alone.

Custom fixtures are JSON:

{
  "description": "My benchmark",
  "version": 1,
  "collection": "my-collection",
  "queries": [
    {
      "id": "find-auth",
      "query": "authentication flow",
      "type": "semantic",
      "expected_files": ["docs/auth-design.md"],
      "expected_in_top_k": 3
    }
  ]
}

expected_files are collection-relative paths as shown by qmd ls. The type field (exact, semantic, topical, cross-domain, alias) labels queries for grouping — it does not change search behavior.

Heads-up: if the fixture's collection isn't indexed, bench currently runs to completion and reports all zeros with no warning. Verify setup with qmd ls <collection> first.

Upgrading

Upgrading the vector layout

Vectors are stored in one index partitioned by collection, so a scoped query scans only that collection's vectors. An index written by an older build keeps its vectors in a single vectors_vec table; the first writable store open converts it in place. Read-only SDK/MCP stores and qmd status do not run this conversion. qmd doctor can open an existing index writable when its config is parsed, and qmd cleanup --dry-run opens a writable store before previewing cleanup; both can therefore trigger initialization and migration. Back up before using either on an older index. For a hash_seq-keyed legacy table with parsed dimensions, the conversion copies, byte for byte, every vector whose chunk metadata belongs to an active document. It sends no request to any provider, never re-embeds, and leaves embedding_index_state alone. Vectors with no chunk metadata or no active document are not copied. Chunk metadata with no vector is deleted so the next qmd embed recreates it; this differs from qmd update and qmd cleanup, which retain a live document's vectorless metadata and leave it pending. A readable legacy table that is not keyed by hash_seq, or whose dimensions cannot be parsed, is dropped without copying its vectors. For a supported table, the conversion prints its progress, resumes if interrupted, and drops the old table in the transaction that stamps the new schema version, after a verification pass that copies rows the walk missed. It then runs a VACUUM that can take minutes on a large index. Do not run an older build against the index while it converts. Without sqlite-vec loaded the conversion is deferred and keyword search keeps working.

Before the first writable open, stop all CLI writers and MCP clients that hold the index. For a systemd user service, stop its unit first. Keep those processes stopped while copying the main database and any WAL/SHM sidecars; copying a live WAL database file by file is not a consistent backup.

# Use your actual unit name, if managed by systemd
systemctl --user stop qmd-mcp-http.service

qmd_backup_dir="$HOME/.local/state/qmd/backups/$(date -u +%Y%m%dT%H%M%SZ)"
mkdir -p "$qmd_backup_dir"
chmod 700 "$qmd_backup_dir"
cp -p "$HOME"/.cache/qmd/index.sqlite* "$qmd_backup_dir/"
chmod 600 "$qmd_backup_dir"/index.sqlite*

Adjust paths and index selection for named or project-local indexes. Verify the backup before continuing. Update the installed code, check qmd --version, then run qmd ls against that same index to trigger initialization without embedding or update hooks. Restart MCP after the conversion finishes.

An older build cannot read the converted layout. To roll back, stop all index holders, move aside all converted database components, and restore the complete pre-upgrade backup before starting the older build. Never pair an old main file with a new WAL/SHM. Without a pre-upgrade backup, returning to the old vector layout requires a full paid qmd embed --force.

Breaking SDK changes

The supported package entry remains createStore() with the existing remote config.models fields. Local-model execution through embed, vector search, expansion, and reranking is retired: missing remote settings now fail explicitly. Model-free callers should use searchLex() or explicit lexical queries with rerank: false and expansion skipped; model-backed queries never silently skip required expansion or reranking.

Code importing implementation modules must migrate away from LlamaCpp, LlamaCppConfig, LocalEmbeddingProvider, LocalEmbeddingProviderOwner, withLLMSession, the default-Llama singleton helpers, pullModels, GGUF/cache inspection, GPU helpers, and local tokenization APIs. They are removed, as is store.internal.localLlm. Shared model types and pure formatting helpers remain separate from inference; scoped sessions require an explicit remote owner. Existing local identity labels remain diagnostic data, not executable providers.

MCP Server

Although the tool works perfectly fine when you just tell your agent to use it on the command line, it also exposes an MCP (Model Context Protocol) server for tighter integration.

Tools exposed:

query — Search with typed sub-queries (lex/vec/hyde), combined via RRF + reranking
get — Retrieve a document by path or docid (with fuzzy matching suggestions)
multi_get — Batch retrieve by glob pattern, comma-separated list, or docids
status — Index health and collection info, including each collection's top metadata keys
metadata — Discover metadata keys, types, and value counts to filter on

Stdio Transport (Default)

Add QMD to your MCP client configuration (e.g. Cursor, Claude Desktop, Zed, or other MCP-compatible clients):

{
  "mcpServers": {
    "qmd": {
      "command": "qmd",
      "args": ["mcp"]
    }
  }
}

HTTP Transport

By default, QMD's MCP server uses stdio (launched as a subprocess by each client). For a shared, long-lived server used by multiple clients, use the HTTP transport:

# Foreground (Ctrl-C to stop)
qmd mcp --http                    # localhost:8181
qmd mcp --http --port 8080        # custom port
qmd mcp --http --host 0.0.0.0     # bind all interfaces (e.g. container probes)

# Background daemon
qmd mcp --http --daemon           # start, writes PID to ~/.cache/qmd/mcp.pid
qmd mcp stop                      # stop via PID file
qmd status                        # shows "MCP: running (PID ...)" when active

The server binds to localhost by default. Pass --host (or set the QMD_HOST environment variable) to override — --host 0.0.0.0 is useful when the server runs in a container and a liveness probe connects from a non-loopback address.

The HTTP server exposes these endpoints:

POST /mcp — MCP Streamable HTTP (JSON responses, stateless)
POST /query (alias /search) — structured search without the MCP protocol. Accepts the same optional filter object as the query tool (invalid filters return 400); see Metadata Filtering
POST /metadata — metadata discovery without the MCP protocol. Same body as the metadata tool (invalid filters return 400); see Metadata Discovery
GET /health — liveness check with uptime

Origin and Host validation

Every request is screened before routing: a request carrying an Origin header that does not name a loopback address is rejected with 403, as is a Host header naming something other than the address the server is bound to. This is what stops a web page you visit from reading your index through DNS rebinding — loopback binding alone does not, since the browser makes the request from your own machine.

Requests without an Origin header — curl, MCP clients, editors — are unaffected, which covers every normal local client.

Variable	Effect
`QMD_ALLOWED_ORIGINS`	Comma-separated origins to accept in addition to loopback, e.g. `https://notes.internal`. Set to `*` to disable the check entirely.
`QMD_ALLOWED_HOSTS`	Comma-separated `Host` values to accept in addition to loopback and the bind address.

--host 0.0.0.0 cannot know which Host values are legitimate, so it skips the host check and warns at startup. Set QMD_ALLOWED_HOSTS to re-enable it, and remember the endpoints are unauthenticated — put your own auth in front of a server that is reachable off-host.

MCP uses remote API clients for model operations. It does not load models into local RAM or VRAM.

Point any MCP client at http://localhost:8181/mcp to connect.

MCP opens the index read-only: starting it does not migrate an older vector layout or rebuild embeddings. Complete the layout upgrade with a writable CLI command before starting MCP on an older index. Restart the server after changing its installed code, model settings, or collection config. After editing collection/context YAML manually, first run qmd ls to reconcile those settings into SQLite; read-only MCP does not synchronize the config itself. Path/pattern changes still require an explicit qmd update to re-index files.

HTTP request bodies for /mcp, /query, /search, and /metadata are limited to 1 MiB, including chunked uploads. Larger requests receive HTTP 413. MCP tool arguments allow at most 10,000 combined array elements and object members; calls over that limit return a tool error. Split larger requests into smaller calls.

For a model-free smoke test, call the query tool with these arguments:

{"searches":[{"type":"lex","query":"project"}],"rerank":false,"limit":1}

Typed lex skips expansion and vector search; rerank: false also disables reranking, which otherwise defaults to true. /health checks process liveness, not retrieval or remote-model availability.

Running with systemd (Linux)

An existing user service, for example qmd-mcp-http.service, should run qmd mcp --http --host 127.0.0.1 --port 8181 in the foreground, without --daemon. Set its ExecStart to the absolute CLI path returned by command -v qmd, and make Node.js available in the service's PATH. The service manager owns restarts; qmd mcp stop only manages QMD's own daemon.

# After installing or editing your user unit
systemctl --user daemon-reload
systemctl --user enable --now qmd-mcp-http.service
systemctl --user status qmd-mcp-http.service
curl --fail http://127.0.0.1:8181/health

When updating a linked CLI, stop the service, follow the backup procedure, update and build the checkout that qmd resolves to, and check qmd --version for its commit. Run qmd ls once to complete any writable layout upgrade, then use systemctl --user start qmd-mcp-http.service. For an already upgraded index, use systemctl --user restart qmd-mcp-http.service after code or config changes.

MCP Tool Parameters

Tool	Parameter	Type	Notes
`query`	`query`	string	Plain query evaluated by the shared expansion policy. Mutually exclusive with `searches`; exactly one is required.
`query`	`searches`	array	Typed sub-queries (`lex`/`vec`/`hyde`), 1–10. Mutually exclusive with `query`; exactly one is required. First gets 2x weight.
`query`	`expansion`	string	Plain-query policy: `auto` (default), `force`, or `skip`. Ignored when `searches` is used.
`query`	`collections`	string[]	Filter by collection names (OR). Array only — singular `collection` is silently ignored.
`query`	`filter`	object	Metadata filter (recursive `operator`-discriminated JSON AST; see Metadata Filtering)
`query`	`expansionContext`	string	Additional context used only to generate `lex` / `vec` / `hyde` query expansions.
`query`	`rerankContext`	string	Additional context used only for reranking and snippet/chunk selection.
`query`	`intent`	string	Disambiguation context (alias for expansionContext & rerankContext; does not search on its own)
`query`	`limit`	number	Max results (default 10)
`query`	`minScore`	number	Minimum relevance 0–1 (default 0)
`query`

Unknown parameters are silently ignored (not rejected) — double-check names if results seem unscoped. The HTTP /query and /search endpoints return qmd://collection/path URIs in the file field, matching the CLI and MCP output.

Using with AI Agents

QMD's --json and --files output formats are designed for agentic workflows:

# Get structured results for an LLM
qmd search "authentication" --json -n 10

# List all relevant files above a threshold
qmd query "error handling" --all --files --min-score 0.4

# Retrieve full document content
qmd get "docs/api-reference.md" --full

QMD also ships a runtime skill that tells an agent how to use it:

qmd skills list              # List the bundled skills (name and description)
qmd skills get qmd           # Print the QMD skill; add --full to include its references
qmd skills path              # Print the directory the bundled skills live in
qmd skill show               # Print the QMD skill
qmd skill install            # Install it into ./.agents/skills/qmd
qmd skill install --global   # ...or into ~/.agents/skills/qmd
                             # --yes also creates the .claude/skills/qmd symlink
                             # -f, --force replaces an existing install or symlink

SDK / Library Usage

Use QMD as a library in your own Node.js applications.

SDK Installation

pnpm add @wei840222/qmd

SDK Quick Start

import { createStore } from '@wei840222/qmd'

const store = await createStore({
  dbPath: './my-index.sqlite',
  config: {
    collections: {
      docs: { path: '/path/to/docs', pattern: '**/*.md' },
    },
  },
})

const results = await store.searchLex("authentication flow")
console.log(results.map(r => `${r.title} (${Math.round(r.score * 100)}%)`))

await store.close()

Store Creation

createStore() accepts three modes:

import { createStore } from '@wei840222/qmd'

// 1. Inline config — no files needed besides the DB
const store = await createStore({
  dbPath: './index.sqlite',
  config: {
    collections: {
      docs: { path: '/path/to/docs', pattern: '**/*.md' },
      notes: { path: '/path/to/notes' },
    },
  },
})

// 2. YAML config file — collections defined in a file
const store2 = await createStore({
  dbPath: './index.sqlite',
  configPath: './qmd.yml',
})

// 3. DB-only — reopen a previously configured store
const store3 = await createStore({ dbPath: './index.sqlite' })

Search

The unified search() method handles both simple queries and pre-expanded structured queries:

// Simple query — evaluated by the shared auto policy, then retrieved and reranked
const results = await store.search({ query: "authentication flow" })

// With options
const results2 = await store.search({
  query: "rate limiting",
  expansionContext: "API throttling and abuse prevention",
  rerankContext: "API throttling and abuse prevention",
  collection: "docs",
  limit: 5,
  minScore: 0.3,
  explain: true,
  expansion: "auto", // "auto" | "force" | "skip"
})

// Pre-expanded queries — bypass the policy and control each sub-query
const results3 = await store.search({
  queries: [
    { type: 'lex', query: '"connection pool" timeout -redis' },
    { type: 'vec', query: 'why do database connections time out under load' },
  ],
  collections: ["docs", "notes"],
})

// Skip reranking for faster results
const fast = await store.search({ query: "auth", rerank: false })

// Metadata filter — every returned result satisfies it (also available on
// searchLex() and searchVector()); results expose indexed metadata via
// r.metadata. See "Metadata Filtering" for the full grammar.
const published = await store.search({
  query: "authentication flow",
  filter: {
    operator: "and",
    operands: [
      { field: "topics", operator: "all", value: ["typescript"] },
      { field: "status", operator: "ne", value: "draft" },
    ],
  },
})

For simple queries, explicit force or skip overrides auto. The automatic shortcuts and remote requirements are described under Query Syntax. Supplying queries bypasses expansion policy evaluation entirely.

For direct backend access:

// BM25 keyword search (fast, no LLM)
const lexResults = await store.searchLex("auth middleware", { limit: 10 })

// Vector similarity search (embedding model, no reranking)
const vecResults = await store.searchVector("how users log in", { limit: 10 })

// Manual query expansion for full control
const expanded = await store.expandQuery("auth flow", { expansionContext: "user login" })
const results4 = await store.search({ queries: expanded })

Retrieval

// Get a document by path or docid
const doc = await store.get("docs/readme.md")
const byId = await store.get("#abc123")

if (!("error" in doc)) {
  console.log(doc.title, doc.displayPath, doc.context)
}

// Get document body with line range
const body = await store.getDocumentBody("docs/readme.md", {
  fromLine: 50,
  maxLines: 100,
})

// Batch retrieve by glob or comma-separated list
const { docs, errors } = await store.multiGet("docs/**/*.md", {
  maxBytes: 20480,
})

Collections

// Add a collection
await store.addCollection("myapp", {
  path: "/src/myapp",
  pattern: "**/*.ts",
  ignore: ["node_modules/**", "*.test.ts"],
})

// List collections with document stats
const collections = await store.listCollections()
// => [{ name, pwd, glob_pattern, doc_count, active_count, last_modified, includeByDefault }]

// Get names of collections included in queries by default
const defaults = await store.getDefaultCollectionNames()

// Remove / rename
await store.removeCollection("myapp")
await store.renameCollection("old-name", "new-name")

Context

Context adds descriptive metadata that improves search relevance and is returned alongside results:

// Add context for a path within a collection
await store.addContext("docs", "/api", "REST API reference documentation")

// Set global context (applies to all collections)
await store.setGlobalContext("Internal engineering documentation")

// List all contexts
const contexts = await store.listContexts()
// => [{ collection, path, context }]

// Remove context
await store.removeContext("docs", "/api")
await store.setGlobalContext(undefined)  // clear global

Indexing

// Re-index collections by scanning the filesystem
const result = await store.update({
  collections: ["docs"],  // optional — defaults to all
  onProgress: ({ collection, file, current, total }) => {
    console.log(`[${collection}] ${current}/${total} ${file}`)
  },
})
// => { collections, indexed, updated, unchanged, removed, needsEmbedding }

// Generate vector embeddings
const embedResult = await store.embed({
  force: false,           // true to re-embed everything
  chunkStrategy: "auto",  // "regex" (default) or "auto" (AST for code files)
  onProgress: ({ current, total, collection }) => {
    console.log(`Embedding ${current}/${total}`)
  },
})

Types

Key types exported for SDK consumers:

import type {
  QMDStore,            // The store interface
  SearchOptions,       // Options for search()
  LexSearchOptions,    // Options for searchLex()
  VectorSearchOptions, // Options for searchVector()
  HybridQueryResult,   // Search result with score, snippet, context
  SearchResult,        // Result from searchLex/searchVector
  ExpandedQuery,       // Typed sub-query { type: 'lex'|'vec'|'hyde', query }
  DocumentResult,      // Document metadata + body
  DocumentNotFound,    // Error with similarFiles suggestions
  MultiGetResult,      // Batch retrieval result
  UpdateProgress,      // Progress callback info for update()
  UpdateResult,        // Aggregated update result
  EmbedProgress,       // Progress callback info for embed()
  EmbedResult,         // Embedding result
  StoreOptions,        // createStore() options
  CollectionConfig,    // Inline config shape
  IndexStatus,         // From getStatus()
  IndexHealthInfo,     // From getIndexHealth()
} from '@wei840222/qmd'

Utility exports:

import {
  extractSnippet,              // Extract a relevant snippet from text
  addLineNumbers,              // Add line numbers to text
  DEFAULT_MULTI_GET_MAX_BYTES, // Default max file size for multiGet (64KB)
  Maintenance,                 // Database maintenance operations
} from '@wei840222/qmd'

Lifecycle

// Close the store — releases remote clients and the DB connection
await store.close()

The SDK requires explicit dbPath — no defaults are assumed. This makes it safe to embed in any application without side effects.

CJK Lexical Search, Expansion, and Dictionary Maintenance

CJK lexical search combines independent character, word, and bigram channels by rank. If Jieba or the published word/bigram index is unavailable, stale, or incompatible, QMD omits both analyzed channels and falls back to character-only search instead of failing the query or issuing an empty FTS match. qmd status reports the reason and remediation.

Each lexical channel ranks candidates independently before channel-aware weighted RRF combines them; raw channel scores are never summed. The original query runs through every available lexical and vector path at ×2 weight. Typed expansion variants are ×1: lex variants use the character, word, and bigram channels, while vec and hyde variants use vector search. Use --explain to inspect channel contributions and tie-breakers.

Query expansion uses one shared auto | force | skip policy across CLI, SDK, and MCP. Precedence is: explicit skip or lex: skips expansion; explicit force, expand:, or --expand forces it; under auto, eligible queries expand when a r

Thêm server này

Host a copy with our keys and logs.

Server facts

Tools

Questions