Filtering Options - QMD (Query Markup Documents)

QMD provides flexible filtering options to narrow search results and retrieve exactly the documents you need.

Collection Filtering

Restrict searches to specific collections using -c or --collection.

Single Collection

Search only one collection:

qmd search "authentication" -c docs
qmd query "error handling" -c api-reference
qmd vsearch "deployment guide" -c operations

Example output:

qmd://docs/auth.md  #a3f2c1  [0.84]
Context: API security documentation
────────────────────────────────────────────────
...authentication uses JWT tokens...

Multiple Collections

Filter across multiple collections (combines results):

qmd search "config" -c docs -c examples
qmd query "database setup" --collection docs --collection guides

How it works:

Searches all specified collections
Combines and ranks results across collections
Equivalent to OR logic: collection=docs OR collection=examples

No Filter (Default)

Without -c, searches all collections:

qmd search "deployment"  # Searches everything

This includes collections with includeByDefault: true (default) and excludes those with includeByDefault: false in your ~/.config/qmd/index.yml.

Collection Name Matching

Collection names can be specified in several ways:

# Exact name match
qmd search "api" -c my-docs

# Works with collection:// virtual paths
qmd search "auth" -c docs  # Searches qmd://docs/*

# Case-sensitive exact match only
qmd search "config" -c Docs  # Won't match "docs"

When to Use Collection Filters

Use Case	Command
Search specific project docs	`qmd search "setup" -c myproject`
Exclude archived content	`qmd search "current" -c active-docs`
Search meeting notes only	`qmd search "Q1 planning" -c meetings`
Multiple related collections	`qmd search "api" -c docs -c examples`
Isolate personal vs work	`qmd search "todo" -c personal`

Result Limits

Control how many results are returned.

Default Limits

Different formats have different default limits optimized for their use case:

Format	Default	Override
CLI (default)	5	`-n 10`
`--json`	20	`-n 50`
`--files`	20	`-n 100`
`--csv`	5	`-n 30`
`--md`	5	`-n 15`
`--xml`	5	`-n 25`

Specify Exact Limit

Use -n <number> to set a custom limit:

qmd search "error" -n 10        # Return top 10 results
qmd search "config" -n 1        # Single best match
qmd query "auth" -n 50 --json   # 50 results in JSON
qmd vsearch "guide" -n 3        # Top 3 vector matches

Example:

$ qmd search "deployment" -n 2
qmd://ops/deploy.md  #a3f2c1  [0.84]
qmd://guides/ci.md   #b5d8e9  [0.72]
# Only 2 results shown

All Results

Use --all to return all matching documents (up to 100,000):

qmd search "TODO" --all
qmd search "error" --all --files
qmd query "config" --all --min-score 0.5

Important: --all returns ALL matches, which can be thousands of results. Use with --min-score to filter:

# Get all high-quality matches
qmd search "deployment" --all --min-score 0.7 --files

# Export everything for analysis
qmd search "api" --all --json > all-api-docs.json

Multi-Get Limits

For multi-get, -l limits lines per file, not number of files:

qmd multi-get "*.md" -l 50              # First 50 lines of each file
qmd multi-get "logs/*.md" -l 10 --json  # First 10 lines, JSON format

All matching files are returned (no -n limit), but content is truncated:

$ qmd multi-get "notes/*.md" -l 5
============================================================
File: qmd://notes/meeting.md
============================================================
1: # Team Meeting
2: 
3: Date: 2024-01-15
4: 
5: ## Attendees

[... truncated 45 more lines]

Score Thresholds

Filter results by minimum relevance score using --min-score.

Default Thresholds

Different search commands have different defaults:

Command	Default	Reason
`search` (BM25)	0	Accept all keyword matches
`vsearch` (vector)	0.3	Filter low-similarity vectors
`query` (hybrid)	0	Reranker provides quality signal

Set Custom Threshold

Specify minimum score (0-1 scale):

qmd search "config" --min-score 0.5   # Only scores ≥ 0.5
qmd vsearch "guide" --min-score 0.7   # High similarity only
qmd query "auth" --min-score 0.6      # Medium-high relevance

Example output:

$ qmd search "deployment" --min-score 0.7
qmd://ops/deploy.md  #a3f2c1  [0.84]  ✓
qmd://guides/ci.md   #b5d8e9  [0.72]  ✓
# Results with score < 0.7 filtered out

Understanding Scores

BM25 scores (search):

Unbounded, typically 0-10 range
Higher = more keyword matches and better TF-IDF
Normalized to 0-1 for display
No score → No threshold by default

Vector scores (vsearch):

0-1 range (cosine similarity)
1.0 = identical embeddings
0.0 = completely unrelated
Default threshold: 0.3

Hybrid scores (query):

0-1 range (reranker confidence)
Combines BM25 + vector + reranking
Higher = model is more confident
No default threshold (reranker already filters)

Best Practices

Use Case	Threshold
Find exact matches only	`--min-score 0.8`
High-quality results	`--min-score 0.6`
Exploratory search	`--min-score 0.3`
Recall (find everything)	`--min-score 0` (default)
Precision (high accuracy)	`--min-score 0.7`

Combine with —all

Get all high-scoring results:

# All docs scoring above 0.7
qmd search "security" --all --min-score 0.7

# All similar documents
qmd vsearch "authentication flow" --all --min-score 0.5

# Export high-confidence results
qmd query "error handling" --all --min-score 0.6 --json > errors.json

Without --min-score, --all could return thousands of low-quality matches.

Output Modifiers

Additional options that affect what content is returned.

Full Content

Show complete documents instead of snippets (search only):

qmd search "deploy" --full
qmd query "auth" --full --md
qmd vsearch "guide" --full --json

Without —full (default):

qmd://docs/deploy.md  #a3f2c1  [0.84]
────────────────────────────────────────────────
...deployment process uses GitHub Actions which
automatically builds and deploys to staging...

With —full:

qmd://docs/deploy.md  #a3f2c1  [0.84]
────────────────────────────────────────────────
# Deployment Guide

## Overview

Our deployment process uses GitHub Actions...

## Staging Environment

Staging deploys happen automatically...

[entire document shown]

Use cases:

Get complete context for LLM prompts
Export full documents for analysis
Review entire files matching search
Generate documentation from search results

Line Numbers

Add line numbers to all output:

qmd search "error" --line-numbers
qmd get readme.md --line-numbers
qmd multi-get "*.md" --line-numbers

Example output:

# Error Handling

All errors inherit from BaseError class.
Custom exceptions are defined in errors.py

## Exception Types

Line numbers are added to:

Search snippets
Full document content (--full)
Single document retrieval (get)
Multi-get file content
All output formats (JSON, CSV, Markdown, XML, CLI)

Example with JSON:

{
  "snippet": "42: def handle_error(e):\n43:     logger.error(e)\n44:     raise CustomError(e)"
}

Line Slicing (Get Command)

Retrieve specific line ranges from a document:

# Start from line 10, show 5 lines
qmd get readme.md --from 10 -l 5

# Shorthand: path:line syntax
qmd get readme.md:10 -l 5
qmd get qmd://docs/api.md:100 -l 20

Example:

$ qmd get readme.md:5 -l 3
## Installation

Run `npm install` to install dependencies.

Starts at line 5, shows 3 lines (lines 5-7).

Multi-Get Size Limits

Control maximum file size for batch retrieval:

qmd multi-get "*.md" --max-bytes 50000   # Skip files > 50KB
qmd multi-get "logs/*.md" --max-bytes 1000000  # 1MB limit

Default: 10,240 bytes (10KB) Skipped files are reported in output:

$ qmd multi-get "*.md" --max-bytes 5000
============================================================
File: qmd://docs/small.md
============================================================
[content shown]

============================================================
File: qmd://docs/large.md
============================================================
[SKIPPED: File too large (15KB > 5KB). Use 'qmd get qmd://docs/large.md' to retrieve.]

Formats handle skipped files differently:

CLI: [SKIPPED: reason]
JSON: {"skipped": true, "reason": "..."}
CSV: true in skipped column
Markdown: > reason (blockquote)
XML: <skipped>true</skipped><reason>...</reason>
Files: [SKIPPED] suffix

Why size limits?

Prevent overwhelming terminal output
Keep JSON payloads reasonable
Avoid processing huge log files
Make CSV exports manageable

Retrieve large files individually:

qmd get qmd://docs/huge-file.md  # No size limit

Combining Filters

Stack multiple filters for precise results:

Collection + Score + Limit

qmd search "authentication" -c docs --min-score 0.7 -n 10
# Docs collection, score ≥ 0.7, max 10 results

Multiple Collections + All + Score

qmd query "deploy" -c ops -c guides --all --min-score 0.6
# Both collections, all results, score ≥ 0.6

Collection + Full + Format

qmd search "config" -c backend --full --json -n 5
# Backend collection, full docs, JSON, 5 results

Line Numbers + Full + Limit

qmd search "error" --line-numbers --full -n 3
# Top 3 results, full content with line numbers

Common Filter Patterns

Find Best Match

qmd search "deployment guide" -n 1
qmd query "how to authenticate" -n 1 --min-score 0.8

High-Quality Results Only

qmd search "security" --min-score 0.7 -n 20
qmd vsearch "architecture" --min-score 0.6

Everything in Collection

qmd search "TODO" -c myproject --all
qmd search "*" -c notes --all --files

Export for Analysis

qmd search "error" --all --min-score 0.5 --json > errors.json
qmd query "api" -c docs --all --csv > api-docs.csv

Preview Multiple Docs

qmd multi-get "guides/*.md" -l 20
qmd multi-get "*.md" --max-bytes 5000 -l 10

Specific Collection Slice

qmd search "config" -c backend -n 5 --full --md
qmd query "auth flow" -c api-docs --min-score 0.7 -n 3

Performance Tips

Use collection filters when possible - searching 1 collection is faster than 10
Set reasonable limits - -n 10 is faster than --all for exploration
Add score thresholds - --min-score 0.5 filters before ranking
Limit line counts - multi-get with -l 50 loads less data
Use —files - minimal format for quick scans

Filter Precedence

Filters are applied in this order:

Collection filter (-c) - narrows search scope
Search/query - finds matching documents
Score threshold (--min-score) - filters by relevance
Limit (-n or --all) - caps result count
Content modifiers (--full, --line-numbers) - affects output format

Commands

Options

​Collection Filtering

​Single Collection

​Multiple Collections

​No Filter (Default)

​Collection Name Matching

​When to Use Collection Filters

​Result Limits

​Default Limits

​Specify Exact Limit

​All Results

​Multi-Get Limits

​Score Thresholds

​Default Thresholds

​Set Custom Threshold

​Understanding Scores

​Best Practices

​Combine with —all

​Output Modifiers

​Full Content

​Line Numbers

​Line Slicing (Get Command)

​Multi-Get Size Limits

​Combining Filters

​Collection + Score + Limit

​Multiple Collections + All + Score

​Collection + Full + Format

​Line Numbers + Full + Limit

​Common Filter Patterns

​Find Best Match

​High-Quality Results Only

​Everything in Collection

​Export for Analysis

​Preview Multiple Docs

​Specific Collection Slice

​Performance Tips

​Filter Precedence

​See Also

Build docs developers (and LLMs) love

Collection Filtering

Single Collection

Multiple Collections

No Filter (Default)

Collection Name Matching

When to Use Collection Filters

Result Limits

Default Limits

Specify Exact Limit

All Results

Multi-Get Limits

Score Thresholds

Default Thresholds

Set Custom Threshold

Understanding Scores

Best Practices

Combine with —all

Output Modifiers

Full Content

Line Numbers

Line Slicing (Get Command)

Multi-Get Size Limits

Combining Filters

Collection + Score + Limit

Multiple Collections + All + Score

Collection + Full + Format

Line Numbers + Full + Limit

Common Filter Patterns

Find Best Match

High-Quality Results Only

Everything in Collection

Export for Analysis

Preview Multiple Docs

Specific Collection Slice

Performance Tips

Filter Precedence

See Also