# YAMS CLI Reference

A practical, task-focused reference for the `yams` command-line interface.

This document covers global flags, environment variables, and each subcommand with usage notes and examples. It is structured so tools can extract command-specific sections.

For help:
- `yams --help` - Show all available commands
- `yams <command> --help` - Show command-specific help

## Synopsis

```bash
yams [OPTIONS] <command> [command-options]
```

## Environment Variables

**Precedence:** explicit CLI flags > configuration file > environment variables > defaults

- **YAMS_STORAGE** - Storage root directory when `--data-dir` is not provided and config does not set `core.data_dir`
- **XDG_DATA_HOME** - Default storage location (`~/.local/share` if unset)
- **XDG_CONFIG_HOME** - Default config location (`~/.config` if unset)
- **HOME** - Fallback when XDG paths are not set

## Global Options

- `--data-dir, --storage <path>` - Override storage root (default: `$XDG_DATA_HOME/yams`)
- `--json` - Output in JSON format where supported
- `-v, --verbose` - Enable verbose output
- `--version` - Display program version
- `-h, --help` - Print help message

## Commands

### Core Operations
- `init` - Initialize YAMS storage and configuration
- `add` - Add document(s) or directory to content store
- `get` - Retrieve a document from content store
- `cat` - Display document content to stdout
- `delete, rm` - Delete documents by hash, name, or pattern
- `list, ls` - List stored documents
- `update` - Update metadata for existing document

### Search & Discovery
- `search` - Search documents by query (semantic + keyword)
- `grep` - Search for regex patterns within file contents
- `tree` - Inspect aggregated path tree statistics
- `graph` - Inspect knowledge graph relationships

### Collections & Snapshots
- `restore` - Restore documents from collections/snapshots to filesystem
- `diff` - Compare two snapshots and show file changes

### System Management
- `daemon` - Manage YAMS daemon process (start/stop/restart/status)
- `doctor` - Diagnose system health, connectivity, and repair issues
- `status` - Show quick system status and health overview
- `stats` - Show detailed system statistics
- `config` - Manage YAMS configuration settings
- `auth` - Manage authentication keys and tokens

### Advanced Features
- `model` - Download and manage ONNX embedding models
- `plugin, plugins` - Manage plugins (list/scan/load/unload/trust)
- `session` - Manage interactive session selectors and warming
- `watch` - Enable session-based auto-ingest for a project
- `download` - Download artifacts and store directly into YAMS
- `repair` - Repair and maintain storage integrity
- `migrate` - Migrate YAMS data and configuration
- `dr` - Disaster recovery operations (plugin-gated)

### User Interface
- `serve` - Start MCP (Model Context Protocol) server
- `completion` - Generate shell completion scripts (bash/zsh/fish/powershell)

### Utilities
- `uninstall` - Remove YAMS from your system

---

## init {#cmd-init}

Initialize YAMS storage and configuration (interactive or non-interactive).

Synopsis:
- yams init [OPTIONS]

Options:
- --non-interactive
  - Run without prompts, using defaults and passed flags.
- --auto
  - Run setup with defaults for headless environments.
- --force
  - Overwrite existing config/keys if already initialized.
- --no-keygen
  - Skip authentication key generation.
- --print
  - Print resulting configuration to stdout (secrets masked).
- --enable-plugins
  - Create and trust a local plugins directory (`~/.local/lib/yams/plugins`).

Notes:
- The storage directory can be set globally via --storage/--data-dir or YAMS_STORAGE.
- On first run, initialization will create the storage directory, database, and configuration.
- Interactive mode prompts for optional model, grammar, and agent-skill setup.
- Init also bootstraps a per-project session (scoping + watch).
- Use `yams watch` to enable auto-ingest for existing projects.

Examples:
```bash
# Interactive initialization
yams init

# Auto mode for headless environments
yams init --auto

# Non-interactive with explicit storage
yams --storage "$HOME/.local/share/yams" init --non-interactive

# Force re-initialization
yams init --force

# Print config without secrets
yams init --print
```

---

## watch {#cmd-watch}

Enable or disable session-based auto-ingest for a project.

Synopsis:
- yams watch [OPTIONS]
- yams watch --stop [OPTIONS]

Options:
- --start
  - Enable watch (default)
- --stop
  - Disable watch
- --interval <ms>
  - Set polling interval in milliseconds
- --root <path>
  - Project root to watch (default: git root or cwd)
- --session <name>
  - Session name (default: auto)
- --no-use
  - Do not set the session as current
- --no-selector
  - Do not add the project root as a session selector
- --daemon-ready-timeout-ms <ms>
  - Max time to wait for daemon readiness (0 to skip)

Notes:
- If the session does not exist, `yams watch` creates it and sets watch settings.
- If no root is provided, the git root is used when available.
- `yams watch` waits for daemon readiness by default to avoid ingesting while the daemon is still initializing.

Examples:
```bash
# Enable watch for the current project (auto session)
yams watch

# Use a 5s polling interval
yams watch --interval 5000

# Disable watch
yams watch --stop

# Explicit root and session
yams watch --root . --session my-project
```

---

## add {#cmd-add}

Manual indexing (when watch is disabled)
Use YAMS to index code updates if auto-ingest is not enabled.

Examples:
```bash
# Index entire source tree (initial import)
yams add src/ --recursive --include="*.cpp,*.hpp,*.h" --tags "code,source"
yams add include/ --recursive --include="*.hpp,*.h" --tags "code,headers"

# Re-index after local edits (quick pass)
yams add . --recursive --include="*.cpp,*.hpp,*.h,*.md" --tags "code,working"

# Add a single updated file
yams add path/to/file.cpp --tags "code,source"

# Update metadata for an existing document by name
yams update --name path/to/file.cpp --metadata "updated=$(date -Iseconds)"
```

Tips:
- Prefer comma-separated patterns with --include for multiple file types.
- Tag code consistently (e.g., code,source; code,headers) to improve later queries.

Add a document to the store from a file or stdin.

Synopsis:
- yams add <path> [options]
- yams add <paths...> [options]
- yams add - [options]    # read from stdin

Options:
- -n, --name <name>
  - Set the document name (especially useful for stdin input)
- -t, --tags <tags>
  - Comma-separated tags for the document (repeatable)
- -m, --metadata <key=value>
  - Custom metadata key-value pairs (repeatable)
- --mime-type <type>
  - Override MIME type detection
- --no-auto-mime
  - Disable automatic MIME type detection
- --no-embeddings
  - Disable automatic embedding generation for added documents
- -r, --recursive
  - Recursively add files from directories
- --include <patterns>
  - File patterns to include (e.g., '*.txt,*.md')
- --exclude <patterns>
  - File patterns to exclude (e.g., '*.tmp,*.log')
- -c, --collection <name>
  - Collection name for organizing documents
- --snapshot-id <id>
  - Unique snapshot identifier
- --snapshot-label <label>
  - Human-readable label for the automatic snapshot (optional)
- --verify
  - Verify stored content (hash + existence) after add; slower but safer
- --global, --no-session
  - Add to global memory (bypass active session)

Description:
- Ingests the specified file or standard input and stores it in the content-addressed store.
- Rich metadata support for tagging, naming, and custom properties.
- Content is automatically indexed for full-text and fuzzy search.
- **Automatic snapshots (snapshot improvements):** Every `yams add` operation automatically creates a snapshot with a timestamp-based ID (e.g., `2025-10-01T14:30:00.123Z`). A Merkle tree is built and stored for efficient diff operations.
- **Invocation-level snapshot consistency:** A single `yams add` command invocation reuses one snapshot ID across all provided inputs (multiple files, stdin, and directories) unless `--snapshot-id` is explicitly set.
- **Snapshot listing parity:** Non-directory adds (single file/stdin) are also persisted into `tree_snapshots`, so `yams list --snapshots` reflects both directory and non-directory snapshot operations.
- Trees enable fast `yams diff` operations using O(log n) subtree matching.
- **Git integration:** When run in a git repository, snapshot metadata includes the current commit hash and branch name.

Examples:
```bash
# Basic document addition (automatic snapshot created)
yams add ./README.md
# → Snapshot: 2025-10-01T14:30:00.123Z

# Add with human-friendly label
yams add . --recursive --snapshot-label "Release 1.0"
# → Snapshot: 2025-10-01T14:30:00.456Z (label: "Release 1.0")

# Organize by collection
yams add docs/ --recursive --collection=documentation
# → Snapshot: 2025-10-01T14:30:01.789Z (collection: "documentation")

# Standard operations (all create automatic snapshots)
cat notes.txt | yams add - --name "meeting-notes" --tags "work,meeting"
yams add document.pdf --tags "research,papers" --metadata "author=John Doe"
yams add src/ --recursive --include="*.cpp,*.h" --tags "code,source"
```

**Automatic Snapshot Workflows:**

1. **Initial import (automatic snapshot):**
   ```bash
   yams add . --recursive --include="*.cpp,*.h,*.md" --snapshot-label "Initial import"
   # → Snapshot: 2025-10-01T09:00:00.000Z
   ```

2. **After making changes (automatic snapshot):**
   ```bash
   yams add . --recursive --include="*.cpp,*.h,*.md" --snapshot-label "Added new feature"
   # → Snapshot: 2025-10-01T14:30:00.000Z
   ```

3. **Compare any two snapshots:**
   ```bash
   # List all snapshots to find IDs
   yams list --snapshots
   
   # Compare using timestamp IDs
   yams diff 2025-10-01T09:00:00.000Z 2025-10-01T14:30:00.000Z
   ```

4. **View snapshot timeline:**
   ```bash
   yams list --snapshots
   # Shows chronological list of all automatic snapshots
   ```

**Performance Notes:**
- Tree building adds ~5-10% overhead to `yams add` but enables O(log n) diff operations.
- Trees are stored once in CAS and deduplicated across snapshots.
- For large repositories (>10K files), tree-based diff is 10-100x faster than flat comparison.
- Automatic snapshots have zero mental overhead - just use `yams add` normally.

---

## get {#cmd-get}

Retrieve a document by hash or name for downloading.

Synopsis:
- yams get <hash> [options]
- yams get --name <name> [options]

Options:
- --name <name>
  - Retrieve document by name instead of hash
- -o, --output <path>
  - Write output to specified file instead of stdout
- -v, --verbose
  - Enable verbose output
- --graph
  - Show related documents from knowledge graph
- --depth <N>
  - Depth of graph traversal, 1-5 (default: 1)

Description:
- Downloads content by hash or name.
- For viewing content directly, use the `cat` command instead.
- Supports both stdout redirection and explicit output file specification.

Examples:
```
yams get abcd1234... -o output.txt
yams get --name "document.pdf" -o restored.pdf
yams get --name "config.json" > config.json
yams get abcd1234... --verbose
```

---

## graph {#cmd-graph}

Show related documents via the knowledge graph for a given document (read-only).

Synopsis:
- yams graph <hash> [--depth N]
- yams graph --name <path|name> [--depth N]

Options:
- --name <path|name>
  - Resolve the target by path/name instead of hash
- --depth <N>
  - Depth of graph traversal, 1-5 (default: 1)
- -v, --verbose
  - Verbose header details

Description:
- Mirrors the graph view available in `get --graph` without fetching content.
- Uses the same daemon request path; shows related documents and their relationship/distance.

Examples:
```
yams graph 0123abcd... --depth 2
yams graph --name "docs/readme.md" --depth 3
```

---

## cat {#cmd-cat}

Display document content to stdout.

Synopsis:
- yams cat <hash>
- yams cat --name <name>

Options:
- --name <name>
  - Display document by name instead of hash

Description:
- Outputs content directly to stdout for viewing or piping.
- Silent operation - no status messages (ideal for piping).
- Use `get` command for downloading files.

Examples:
```
yams cat abcd1234...
yams cat --name "notes.txt"
yams cat --name "config.json" | jq .
yams cat --name "script.sh" | bash
```

---

## delete {#cmd-delete}
## rm {#cmd-rm}

Delete documents by hash, name, or pattern. (Alias: `rm`)

Synopsis:
- yams delete <hash> [options]
- yams delete --name <name> [options]
- yams delete --names <name1,name2,...> [options]
- yams delete --pattern <pattern> [options]

Options:
- --name <name>
  - Delete a document by its name
- --names <names>
  - Delete multiple documents by names (comma-separated)
- --pattern <pattern>
  - Delete documents matching a glob pattern (e.g., *.log, temp_*.txt)
- --force, --no-confirm
  - Skip confirmation prompt
- --dry-run
  - Preview what would be deleted without actually deleting
- --keep-refs
  - Keep reference counts (don't decrement)
- -v, --verbose
  - Show detailed deletion progress

Description:
- Supports deletion by hash (original behavior), name, multiple names, or pattern matching.
- When deleting by name and multiple documents have the same name, you'll be prompted to confirm unless --force is used.
- Pattern matching supports standard glob patterns (* for any characters, ? for single character).
- Dry-run mode shows what would be deleted without making changes.
- Bulk deletions show progress and report both successes and failures.

Examples:
```
# Delete by hash (original)
yams delete abcd1234...
yams rm abcd1234...  # Using alias

# Delete by name
yams delete --name "meeting-notes.txt"
yams rm --name "old-file.txt" --force  # Using alias

# Delete multiple files
yams delete --names "old-report.pdf,draft.txt,temp.log"

# Delete by pattern
yams delete --pattern "*.tmp"
yams delete --pattern "backup_*.zip" --dry-run

# Force deletion without confirmation
yams delete --pattern "temp_*" --force

# Verbose output
yams delete --names "file1.txt,file2.txt" --verbose
```

---

## list {#cmd-list}
## ls {#cmd-ls}

List stored documents with rich metadata display. (Alias: `ls`)

Synopsis:
- yams list [options]

Options:
- --format <format>
  - Output format: table | json | csv | minimal (default: table)
- -l, --limit <number>
  - Maximum number of documents to list (default: 100)
- --offset <number>
  - Offset for pagination (default: 0)
- --recent <N>
  - Show N most recent documents
- --sort <field>
  - Sort by: name | size | date | hash (default: date)
- -r, --reverse
  - Reverse sort order
- -v, --verbose
  - Show detailed information
- --show-snippets
  - Show content previews (default: true)
- --show-metadata
  - Show all metadata for each document
- --show-tags
  - Show document tags (default: true)
- --snippet-length <N>
  - Length of content snippets (default: 50)
- --no-snippets
  - Disable content previews

Description:
- Displays documents with names, types, sizes, content snippets, and timestamps.
- Rich table format shows stable NAME, TYPE, SIZE, SNIPPET, and WHEN columns.
- By default, a compact detail line shows tags plus key metadata (`owner`, `source`) under each row when present.
- `--show-metadata` expands detail lines to include all metadata keys.
- The --recent flag filters to the N most recent documents before applying other sorting options.
- Content snippets provide quick previews of document contents.
- Multiple output formats support different use cases from human-readable to machine processing.
- JSON output separates `metadata` and `tags` (tags are an array, metadata excludes tag keys).
- `--show-metadata` prints metadata even without `--verbose` for table output.

Examples:
```
yams list
yams ls  # Using alias
yams list --recent 10  # Show 10 most recent documents
yams list --sort size --reverse  # Sort by size, largest first
yams list --format json
yams ls --format csv --limit 50  # Using alias
yams list --format minimal --offset 20
yams list --no-snippets --show-metadata  # No previews, full metadata
yams list --format json --show-metadata | jq '.documents[] | {name,metadata,tags}'
```

#### Tree-Based Query Optimization (v0.7.5+)

The `list` command uses tree-based path indexing for efficient queries on large repositories. Path patterns with wildcards (`*`, `**`) automatically leverage tree structures for O(log n) performance:

**Pattern-based queries** (tree-optimized):
```bash
# List all documents under docs/ directory (recursive)
yams list --name "docs/**"

# List all markdown files in src/
yams list --name "src/**/*.md"

# List all files in a specific directory (non-recursive)
yams list --name "project/bin/*"
```

**Combining patterns with filters** (tree query + filtering):
```bash
# Documents under docs/ with specific tag
yams list --name "docs/**" --tags "api"

# C++ source files under src/ with multiple tags
yams list --name "src/**/*.cpp" --tags "code,reviewed"

# All text files under project/ by MIME type
yams list --name "project/**" --mime "text/plain"

# Files by extension under specific path
yams list --name "examples/**" --extension ".json"

# Combined filters for precise results
yams list --name "src/**" --tags "test" --extension ".cpp" --mime "text/x-c++"
```

**Performance notes:**
- Tree queries with path prefixes: ~160 μs per query (6.25k queries/sec)
- Tree queries with filters: 100-126 μs per query (up to 10k queries/sec)
- Filters can improve performance by reducing result set size
- Use `/**` for recursive directory matching, `/*` for single-level
- Patterns are automatically canonicalized on macOS to handle `/var` → `/private/var` symlinks

**Best practices:**
- Prefer tree-based patterns (`path/**`) over substring matches for better performance
- Combine multiple filters to narrow results efficiently
- Use `--format json` with tree queries for programmatic processing

### Snapshot Operations (Enhanced in snapshot improvements)

The `list` command has been enhanced with smart snapshot and file history capabilities:

**List all snapshots:**
```bash
yams list --snapshots
yams list --snapshots --format json
yams list --snapshots --collection myproject
```

**Show file history across snapshots:**
```bash
# Shows timeline of changes (Added, Modified, Renamed) across all snapshots
yams list src/main.cpp

# Output example:
# File: src/main.cpp
# 
# Snapshot History:
# v1.2  2025-10-01 14:30  Modified  hash: abc123...  2.5 KB  (current)
# v1.1  2025-09-22 10:15  Modified  hash: def456...  2.3 KB  
# v1.0  2025-09-15 09:00  Added     hash: ghi789...  2.1 KB
```

**Show file at specific snapshot:**
```bash
yams list src/main.cpp --snapshot-id=v1.0

# Output shows metadata and suggests next actions:
# File: src/main.cpp (v1.0)
# Size: 2.1 KB
# Hash: ghi789...
# 
# Use 'yams cat src/main.cpp --snapshot-id=v1.0' to view content
# Use 'yams list src/main.cpp' to see history across all snapshots
```

**Compare file between snapshots (inline diff):**
```bash
yams list src/main.cpp --snapshot-id=v1.0 --compare-to=v1.1

# Shows inline diff with syntax highlighting:
# File: src/main.cpp
# Comparing: v1.0 → v1.1
# 
# @@ -10,6 +10,8 @@
#  int main(int argc, char** argv) {
# +    // Initialize logging
# +    initLogger();
#      processArgs(argc, argv);
#      return runApp();
#  }
```

**Smart behavior:**
- `yams list` → List all documents
- `yams list --snapshots` → List all available snapshots
- `yams list <directory>` → List indexed descendants under that directory (default `--limit 100`, use `--offset` to paginate)
- `yams list <pattern-with-*-or-?>` → Treat input as a list pattern, including absolute path globs
- `yams list <filepath>` → Show file history across snapshots when the input resolves to a file or historical path
- `yams list <filepath> --snapshot-id=X` → Show file at snapshot X
- `yams list <filepath> --snapshot-id=X --compare-to=Y` → Inline diff

The command classifies inputs by meaning instead of just checking for `/` or `.`.

---

## diff {#cmd-diff}

Compare two snapshots and show file changes with tree diff.

**New in snapshot improvements:** Efficient snapshot comparison using Merkle tree diffs with rename detection.

Synopsis:
- yams diff <snapshotA> <snapshotB> [options]

Options:
- --format <format>
  - Output format: tree (default) | flat | json
  - tree: Structured diff with rename detection (recommended)
  - flat: Legacy flat diff without rename tracking
  - json: Machine-readable output for tools
- --include <patterns>
  - Include only files matching patterns (comma-separated globs)
  - Example: --include="*.cpp,*.h"
- --exclude <patterns>
  - Exclude files matching patterns (comma-separated globs)
  - Example: --exclude="*.log,*.tmp,build/**"
- --type <change-type>
  - Filter by change type: added | deleted | modified | renamed
  - Example: --type=modified (show only modified files)
- --stats
  - Show only summary statistics (no file list)
- --no-renames
  - Disable rename detection (faster, but less accurate)
- -v, --verbose
  - Show detailed information including file hashes

Description:
- Compares two snapshots and displays changes using efficient tree diff algorithm.
- **Rename detection:** Automatically detects moved/renamed files by matching content hashes.
- **O(log n) subtree matching:** Unchanged subtrees are skipped for performance.
- Changes are grouped by type: Added, Deleted, Modified, Renamed, Moved.
- Tree format is the default and recommended for most use cases.

Examples:
```bash
# Basic diff
yams diff v1.0 v1.1

# Output example:
# Comparing v1.0 → v1.1
# 
# Added (6 files):
#   + src/new-feature.cpp
#   + src/new-feature.h
#   + tests/test_new_feature.cpp
#   + docs/new-feature.md
#   + lib/dependency.so
#   + config/feature-flags.json
# 
# Deleted (2 files):
#   - src/deprecated.cpp
#   - src/deprecated.h
# 
# Modified (3 files):
#   ~ src/main.cpp                 (2.3 KB → 2.5 KB)
#   ~ README.md                     (1.1 KB → 1.3 KB)
#   ~ Makefile                      (845 B → 920 B)
# 
# Renamed (1 file):
#   → src/utils.cpp → src/common/utils.cpp
# 
# Summary:
#   +6 files added
#   -2 files deleted
#   ~3 files modified
#   →1 file renamed
#   Total: 12 changes

# Filter by file type
yams diff v1.0 v1.1 --include="*.cpp,*.h"

# Show only modified files
yams diff v1.0 v1.1 --type=modified

# Show summary statistics only
yams diff v1.0 v1.1 --stats

# Output example (stats):
# Snapshot Comparison: v1.0 → v1.1
#   Files added:    6
#   Files deleted:  2
#   Files modified: 3
#   Files renamed:  1
#   Total changes:  12

# JSON output for tools
yams diff v1.0 v1.1 --format=json

# Legacy flat diff (convenience flag)
yams diff v1.0 v1.1 --flat-diff
# Or equivalently:
yams diff v1.0 v1.1 --format=flat

# Compare specific directory
yams diff v1.0 v1.1 --include="src/**"

# Exclude build artifacts
yams diff v1.0 v1.1 --exclude="build/**,*.o,*.so"

# Verbose output with hashes
yams diff v1.0 v1.1 -v
```

**Performance Notes:**
- Tree diff is computed in O(log n) time when trees are pre-built during `yams add`.
- If trees don't exist, they will be computed on-demand (may be slower for large snapshots).
- Use `--no-renames` to skip rename detection for faster results on very large diffs.

**See Also:**
- `yams list --snapshots` - List available snapshots
- `yams list <file>` - Show file history across snapshots
- `yams add --snapshot-id` - Create snapshots

---

## search {#cmd-search}
Note: Default search type is hybrid. When strict/hybrid returns zero results, the CLI auto-retries with fuzzy (similarity 0.7).

YAMS-first code search
Always use YAMS to search the indexed codebase (no external grep/find/rg).

Notes:
- If results are empty, ensure the repo is indexed (`yams add ...`) and the daemon is ready (`yams status`).
- Scope searches to the repo with `--cwd` or `--path` when multiple projects share a storage.

Examples:
```bash
# List only file paths for efficient context
yams search "IndexingPipeline" --paths-only

# Fuzzy search for broader discovery
yams search "vector database" --fuzzy --similarity 0.7 --paths-only

# Restrict by extension via keyword filters
yams search "class DocumentIndexer ext:.cpp" --paths-only

# Show JSON for scripting
yams search "SyncManager" --json

# Queries that start with '-' (option-looking) — pass safely
yams search -q "--start-group --whole-archive --end-group -force_load Darwin" --paths-only
yams search -- "--start-group --whole-archive --end-group -force_load Darwin" --paths-only
printf '%s\n' "--start-group --whole-archive --end-group -force_load Darwin" | yams search --stdin --paths-only
yams search --query-file /tmp/query.txt --paths-only
yams search --query-file - --paths-only < /tmp/query.txt
```

Hints:
- Prefer hybrid or fuzzy search for exploratory queries; narrow with exact keywords as you iterate.
- Combine with --paths-only to feed subsequent yams get calls.
- Session helpers (experimental): Use `yams session add|list|rm-path|warm` to manage hot data. See PROMPT docs for examples: ../PROMPT-eng.md and ../PROMPT.md

Search for documents with advanced query capabilities.

Synopsis:
- yams search "<query>" [options]
- yams search -q "<query>" [options]
- yams search --stdin [options]
- yams search --query-file <path> [options]

Options:
- -q, --query <text>
  - Provide the query as an option (useful when it starts with '-')
- --stdin
  - Read query from standard input when no query is provided
- --query-file <path>
  - Read query from a file path (use '-' to read from stdin)
- -l, --limit <number>
  - Maximum number of results to return (default: 20)
- -t, --type <type>
  - Search type: keyword | semantic | hybrid (default: hybrid)
- -f, --fuzzy
  - Enable fuzzy search for approximate matching
- --similarity <value>
  - Minimum similarity for fuzzy search, 0.0-1.0 (default: 0.7)
- --hash <hash>
  - Search by file hash (full or partial, minimum 8 characters)
- -n, --line-numbers
  - Show line numbers with matches
- -A, --after <N>
  - Show N lines after match (default: 0)
- -B, --before <N>
  - Show N lines before match (default: 0)
- -C, --context <N>
  - Show N lines before and after match (default: 0)

Description:
- Supports both exact keyword searches and fuzzy approximate matching.
- Fuzzy search uses BK-tree indexing for efficient similarity matching with configurable thresholds.
- Full-text search with FTS5 indexing provides fast document content queries with robust special character handling.
- Hash search allows finding documents by their SHA256 hash (full 64-character hash or partial prefix).
- Automatic fallback: if FTS5 query fails (due to special characters), automatically falls back to fuzzy search.
- Robust query sanitization handles special characters like hyphens, quotes, and operators (e.g., "task-6", "task 4-").
- JSON output includes relevance scores, execution times, and content snippets.
- Searches document names, content, tags, and metadata fields.
- Auto-detects hash format: if query looks like a hash (8-64 hex chars), automatically searches by hash.
- Verbosity control: concise output by default, detailed output with --verbose flag.

Examples:
```
# Text search
yams search "database performance"
yams search "config file" --fuzzy --similarity 0.6
yams search "meeting notes" --limit 10 --type keyword
yams search "project roadmap" --json

# Hash search
yams search --hash abcd1234ef567890  # Partial hash (minimum 8 chars)
yams search --hash abcd1234ef567890abcd1234ef567890abcd1234ef567890abcd1234ef567890  # Full hash
yams search abcd1234ef567890  # Auto-detected hash search

# Verbose output for detailed information
yams search "query" --verbose
yams search --hash abcd1234 --verbose
```

---

## grep {#cmd-grep}

Regex across indexed code (YAMS only, hybrid by default)
Use YAMS grep for project-wide regex; avoid system utilities for repository queries.

Examples:
```bash
# Find class definitions
yams grep "class\\s+IndexingPipeline" --include="**/*.hpp,**/*.cpp"

# Locate TODOs in headers and sources
yams grep "TODO\\b" --include="**/*.hpp,**/*.h,**/*.cpp"

# Search for CLI subcommand declarations
yams grep "##\\s+(watch|git|sync)\\b" --include="**/*.md"
```

Notes:
- Default mode is hybrid: regex plus semantic suggestions; use --regex-only to disable. Tune with --semantic-limit (default: 10).
- Semantic suggestions are also shown in -l/--files-with-matches, -L/--files-without-match, --paths-only, and -c/--count modes. Regex counts only reflect text matches.
- Hot/Cold modes: control behavior via environment variables. For grep, set YAMS_GREP_MODE=hot_only|cold_only|auto (hot uses extracted text cache; cold scans CAS bytes). For list, use YAMS_LIST_MODE, and for retrieval (cat/get), use YAMS_RETRIEVAL_MODE.
- --include accepts comma-separated globs or repeated usage; prefer quoting patterns.
- Pair with yams search --paths-only to scope subsequent grep runs.

Search for regex patterns within file contents.

Synopsis:
- yams grep <pattern> [paths...] [options]

Options:
- -A, --after <N>
  - Show N lines after match (default: 0)
- -B, --before <N>
  - Show N lines before match (default: 0)
- -C, --context <N>
  - Show N lines before and after match (default: 0)
- -i, --ignore-case
  - Case-insensitive search
- -w, --word
  - Match whole words only
- -v, --invert
  - Invert match (show non-matching lines)
- -n, --line-numbers
  - Show line numbers
- -H, --with-filename
  - Show filename with matches
- --no-filename
  - Never show filename
- -c, --count
  - Show only count of matching lines
- -l, --files-with-matches
  - Show only filenames with matches
- -L, --files-without-match
  - Show only filenames without matches
- --color <mode>
  - Color mode: always, never, auto (default: auto)
- --max-count <N>
  - Stop after N matches per file
- --limit <N>
  - Alias for --max-count

Description:
- Searches through the content of all indexed files using regular expressions.
- Supports standard grep-like options for context, case sensitivity, and output control.
- Automatically highlights matches when outputting to a terminal.
- Can search all indexed files or specific paths.
- Uses ECMAScript regex syntax for pattern matching.

Examples:
```
# Basic pattern search
yams grep "TODO"
yams grep "error.*failed" -i

# With context
yams grep "function" -C 2
yams grep "class.*Repository" -A 5

# File listing
yams grep "deprecated" -l
yams grep "test" src/ -c

# Complex patterns
yams grep "^import.*from" -n
yams grep "\bclass\s+\w+Command\b" --color=always
```

---

## config {#cmd-config}

Manage YAMS configuration settings.

Subcommands:
- get <key>
  - Print a configuration value.
- set <key> <value>
  - Update a configuration value.
- list
  - Print all keys/values.
- validate [--config-path <file>]
  - Validate configuration.
- export [--format toml|json]
  - Output current configuration.
- embeddings
  - Manage embedding configuration (model, dimensions).
- search
  - Manage search configuration.
- grammar
  - Manage tree-sitter grammars for symbol extraction.
- tuning
  - Manage daemon tuning settings.
- storage
  - Configure storage settings.
- migrate
  - Migrate configuration to v3.
- update
  - Add newly introduced v3 keys without overwriting existing values.
- check
  - Check if config needs migration.

Examples:
```bash
yams config list
yams config get core.data_dir
yams config set embeddings.preferred_model all-MiniLM-L6-v2
yams config validate
yams config export --format json
yams config grammar list              # List available grammars
yams config grammar download cpp      # Download C++ grammar
```

---

## auth {#cmd-auth}

Manage authentication material (keys, tokens, API keys). Availability and exact options depend on your build.

Examples:
```
yams auth --help
```

---

## stats {#cmd-stats}

Show storage statistics and health.

Synopsis:
- yams stats [options]

Description:
- Adaptive default view: when relevant, includes Recommendations and Service Status along with the compact footer; use -v for full System Health and detailed sections.
- Telemetry includes auto-repair counters (repair_queue_depth, repair_batches_attempted, repair_embeddings_generated, repair_failed_operations), latency percentiles (p50/p95), top_slowest components, and not_ready flags.
- Use --json for machine-readable output.

Examples:
```
yams stats
yams stats -v
yams stats --json
```

---

## uninstall {#cmd-uninstall}

Remove YAMS data/config from the system.

Synopsis:
- yams uninstall [--force]

Warning:
- This is destructive. Make sure you have backups.

Examples:
```
yams uninstall
yams uninstall --force
```

---

## migrate {#cmd-migrate}

Run metadata migrations if any are pending.

Synopsis:
- yams migrate

Examples:
```
yams migrate
```

---

## update {#cmd-update}

Update metadata for existing documents.

Synopsis:
- yams update <hash> --metadata <key=value>...
- yams update --name <name> --metadata <key=value>...

Options:
- hash
  - Document hash (full or partial) to update
- --name <name>
  - Document name to update (useful for stdin documents)
- -m, --metadata <key=value>
  - Metadata key-value pairs to set (can be specified multiple times)
- -v, --verbose
  - Show detailed update information

Notes:
- Either hash or --name must be specified, but not both
- Multiple metadata pairs can be updated in a single command
- Existing metadata values will be overwritten
- Works with both file-based and stdin documents
- If multiple documents have the same name, you'll be prompted to use the hash

Examples:
```
# Update by hash
yams update fc8fc5fa --metadata "status=completed" --metadata "reviewed=true"

# Update by name  
yams update --name "project-notes.md" --metadata "priority=high"

# Update stdin document
yams update --name "arxiv-2402.05391" --metadata "read=true" --metadata "rating=5"

# Task tracking example
yams update --name "task-001.md" --metadata "status=in_progress" --metadata "assignee=alice"
```

---

## model {#cmd-model}

Manage embedding models: list installed, download new, inspect, and verify runtime.

Synopsis:
- yams model list
- yams model download <name> [--url <onnx_url>]
- yams model info <name>
- yams model check

Description:
- list: Shows locally available models (autodiscovers ~/.yams/models/<name>/model.onnx and other configured roots)
- download: Fetches a model by name; use --url to override with a custom ONNX URL
- info: Prints details for a model (dimensionality, path, notes when available)
- check: Verifies ONNX runtime support, plugin directory status, and autodiscovery paths

Configuring the preferred model:
- yams config embeddings model <name>  # sets embeddings.preferred_model
- embeddings.model_path can point to the models root; name-based resolution order:
  configured root → ~/.yams/models → models/ → /usr/local/share/yams/models

Examples:
```
yams model list
yams model download all-MiniLM-L6-v2
yams model download my-custom --url https://example.com/models/custom.onnx
yams model info all-MiniLM-L6-v2
yams model check
yams config embeddings model all-MiniLM-L6-v2
```

---

## plugin {#cmd-plugin}

Manage plugins: scanning, trust policy, load/unload, install, and info.

Synopsis:
- yams plugin list
- yams plugin scan [--dir DIR] [TARGET]
- yams plugin info <name>
- yams plugin health
- yams plugin load <path|name>
- yams plugin unload <name>
- yams plugin trust add <path> | list [--details] | remove <path> | reset | status
- yams plugin install <name|url>
- yams plugin uninstall <name>
- yams plugin update [name]
- yams plugin repo [query]
- yams plugin search <query>

Subcommands:
- list: Show loaded plugins and their interfaces.
- scan: Inspect directories for plugins without initializing; prints name/version/ABI.
- info: Show manifest/health JSON for a loaded plugin.
- health: Show health status of loaded plugins.
- load: Load a plugin by absolute path or name; respects trust policy.
- unload: Unload a plugin by name.
- trust: Manage plugin trust and discovery status. Canonical trust file is `<data_dir>/plugins.trust` (default `~/.local/share/yams/plugins.trust`).
- install: Install plugin from repository or URL.
- uninstall: Uninstall a plugin.
- update: Update plugins to latest version.
- repo: Query plugin repository.
- search: Search plugin catalog (local index.json).
- enable/disable: Enable/disable plugins (stub).
- verify: Verify plugin signature/hash (stub).

Notes:
- Default discovery roots: persisted trusted paths plus built-in defaults (`$HOME/.local/lib/yams/plugins`, `/opt/homebrew/lib/yams/plugins` on macOS, `/usr/local/lib/yams/plugins`, `/usr/lib/yams/plugins`, `${CMAKE_INSTALL_PREFIX}/lib/yams/plugins`). Strict mode (`YAMS_PLUGIN_DIR_STRICT=1` or `daemon.plugin_dir_strict=true`) disables built-in defaults.
- The daemon prefers host-backed `model_provider_v1` when an ONNX plugin is trusted/loaded; otherwise it falls back to the legacy registry or mock/null provider.
- Disable plugin subsystem: start the daemon with `--no-plugins`.

Examples:
```bash
# Discover and trust a system plugins directory
yams plugin scan
yams plugin trust add /usr/local/lib/yams/plugins
yams plugin trust list
yams plugin trust status
yams plugin trust reset

# Load ONNX plugin by path (respects trust policy)
yams plugin load /usr/local/lib/yams/plugins/libyams_onnx_plugin.so

# Check plugin health
yams plugin health

# Inspect and unload
yams plugin info onnx
yams plugin unload onnx
```

---

## doctor {#cmd-doctor}

Diagnose daemon connectivity, model/provider readiness, vector DB dimensions, and plugins. Includes repair helpers.

Synopsis:
- yams doctor [OPTIONS]
- yams doctor daemon
- yams doctor plugin [<path|name>]
- yams doctor plugins
- yams doctor embeddings
- yams doctor repair [--embeddings] [--fts5] [--graph] [--all]
- yams doctor validate
- yams doctor dedupe
- yams doctor prune
- yams doctor tuning

Subcommands:
- daemon: Check daemon socket and status
- plugin: Check a specific plugin (.so/.wasm or by name)
- plugins: Show plugin summary (loaded + scan)
- embeddings: Embeddings diagnostics and actions
- repair: Repair common issues (embeddings, FTS5, graph)
- validate: Validate knowledge graph health
- dedupe: Detect (and optionally remove) duplicate documents
- prune: Remove build artifacts, logs, cache, temporary files, and git artifacts
- tuning: Auto-configure [tuning] based on system baseline

Options:
- --json: Output results in JSON format
- --fix: Fix everything (embeddings + FTS5)
- --fix-config-dims: Align config embedding dims to target (non-interactive)
- --recreate-vectors: Drop and recreate vector tables to target dim
- --dim <N>: Target dimension to use with --recreate-vectors
- --stop-daemon: Attempt to stop daemon before DB operations

Highlights:
- Summary shows daemon status, vector DB dimension vs model target, and loaded plugins.
- Knowledge Graph section appears when available; if empty, recommends:
  `yams doctor repair --graph` to build from tags/metadata.

Repair options:
- --embeddings: Generate missing document embeddings (daemon streaming when available; local fallback).
- --fts5: Rebuild the text index best-effort.
- --graph: Construct/repair the knowledge graph from tags/metadata.
- --all: Run all repair operations.

Examples:
```bash
yams doctor
yams doctor --json
yams doctor plugin onnx
yams doctor plugins
yams doctor repair --graph
yams doctor repair --all
yams doctor --recreate-vectors --dim 768 --stop-daemon
yams doctor dedupe --dry-run
yams doctor prune
yams doctor tuning
```

---

## repair {#cmd-repair}

Run storage/database maintenance and (optionally) embedding generation.

Synopsis:
- yams repair [OPTIONS]

Options:
- --orphans: Clean orphaned metadata entries
- --chunks: Remove orphaned chunk files and reclaim space
- --mime: Fix missing MIME types in documents
- --fts5: Rebuild FTS5 index for documents (best-effort)
- --embeddings: Generate missing vector embeddings
- --checksums: Verify and repair checksums
- --duplicates: Find and optionally merge duplicates
- --downloads: Repair download documents (add tags/metadata, normalize names)
- --path-tree: Rebuild path tree index for documents missing from the tree
- --optimize: Vacuum/optimize the database
- --all: Run all repair operations
- --dry-run: Preview operations without making changes
- --force: Skip confirmations
- --foreground: Run embeddings repair in foreground (stream progress)
- --include-mime <mime>: Additional MIME types to embed (e.g., application/pdf), repeatable
- --model <name>: Embedding model to use (overrides preferred)
- --stop-daemon: Attempt to stop daemon before vector DB operations
- -v, --verbose: Show detailed progress

Notes:
- By default, embedding repair targets text-like MIME types only; binaries (PDFs/images) are skipped unless explicitly included with --include-mime.
- PDF text extraction and embedding depend on build configuration (PDF support must be available).

Examples:
```bash
yams repair --all --dry-run
yams repair --orphans --chunks --optimize --force
yams repair --embeddings
yams repair --embeddings --include-mime application/pdf
yams repair --fts5 --verbose
yams repair --path-tree
```

---

## status {#cmd-status}

Show service readiness, subsystem health, and runtime stats (daemon-aware).

Synopsis:
- yams status

Description:
- Prints a concise services summary (e.g., ✓ Content | ✓ Repo | ✓ Search | ⚠ (N) Models)
- During initialization, shows a WAIT line with not-ready components and progress
- Exposes fields such as running/ready state, uptime, request counts, active connections, memory, CPU, and version when available
- Highlights top slowest components and actionable recommendations when relevant

Tips:
- Use yams stats -v for detailed system health, recommendations, and service status sections.
- Bootstrap status file: ~/.local/state/yams/yams-daemon.status.json

Examples:
```
yams status
```

---

## daemon {#cmd-daemon}

Control the background daemon (if included in your build).

Synopsis:
- yams daemon start
- yams daemon stop
- yams daemon status
- yams daemon restart

Description:
- start/stop/restart the daemon process
- status shows readiness and subsystem overview similar to yams status

Notes:
- When client/daemon protocol versions differ, a one-time warning may be shown; upgrade the daemon if features appear limited.
- Runtime mode selection supports socket and embedded in-process transports.
- Set `YAMS_EMBEDDED=1|true|on|yes` to force in-process mode, `0|false|off|no` to force socket mode, or `auto` to probe.
- Config fallback: `[daemon].mode = "embedded" | "socket" | "auto"`.

Examples:
```
yams daemon status
yams daemon restart
```

---

## serve {#cmd-serve}

Start the MCP (Model Context Protocol) server over stdio.

Synopsis:
- yams serve [OPTIONS]

Options:
- --daemon-socket <path>
  - Override daemon socket path (env: YAMS_DAEMON_SOCKET)
- --verbose
  - Show startup banner and enable info-level logging (default: quiet mode)

Description:
- Exposes YAMS functionality through the Model Context Protocol for AI tool integration.
- Uses stdio transport (JSON-RPC over stdin/stdout) for direct AI integration.
- Provides search, retrieval, and document management capabilities to AI systems.
- Graceful shutdown on SIGINT/SIGTERM signals.
- Server runs quietly by default; use --verbose for startup banner.

Examples:
```bash
# Start MCP server (default quiet mode)
yams serve

# With verbose logging
yams serve --verbose

# With custom daemon socket
yams serve --daemon-socket /tmp/yams.sock
```

---

## completion {#cmd-completion}

Generate shell completion scripts for popular shells.

Synopsis:
- yams completion bash
- yams completion zsh
- yams completion fish
- yams completion powershell

Notes:
- Bash: if bash-completion isn't installed, a minimal fallback is baked into the script.
- Zsh: the generated script auto-runs compinit if needed to prevent "_arguments: command not found".
- Fish: installs via standard fish completions.
- PowerShell: uses `Register-ArgumentCompleter` for native tab completion.

### Installation Instructions

**Bash (Linux/macOS):**
```bash
# Quick use without installing
source <(yams completion bash)

# Install for current user
mkdir -p ~/.local/share/bash-completion/completions
yams completion bash > ~/.local/share/bash-completion/completions/yams

# Or install system-wide (requires sudo)
yams completion bash | sudo tee /etc/bash_completion.d/yams > /dev/null
```

Note: Homebrew installs the bash completion file, but your shell still needs bash-completion
startup support enabled.

**Zsh (Linux/macOS):**
```bash
# Quick use without installing
autoload -U compinit && compinit
source <(yams completion zsh)

# Install for current user (Oh My Zsh)
yams completion zsh > ~/.oh-my-zsh/completions/_yams

# Or standard location
mkdir -p ~/.local/share/zsh/site-functions
yams completion zsh > ~/.local/share/zsh/site-functions/_yams

# Ensure the install directory is on fpath before compinit runs
fpath=(~/.local/share/zsh/site-functions $fpath)
autoload -U compinit && compinit
```

The generated completions now support nested subcommands and nested leaf values such as:
- `yams config embeddings ...`
- `yams doctor embeddings ...`
- `yams plugin trust ...`
- `yams plugins trust ...` (alias of `plugin`)
- `yams daemon start --log-level <Tab>`
- `yams config search path-tree enable --mode <Tab>`

**Fish:**
```bash
yams completion fish > ~/.config/fish/completions/yams.fish
```

If you are not using Homebrew's fish package, make sure your `fish_complete_path`
includes the directory where the completion file is installed.

**PowerShell (Windows/Linux/macOS):**
```powershell
# Quick use for current session
Invoke-Expression (yams completion powershell | Out-String)

# Install permanently (add to profile)
yams completion powershell | Out-File -Encoding utf8 $PROFILE.CurrentUserAllHosts -Append

# Or create a separate completion file
$completionPath = "$HOME\.config\yams\completions"
New-Item -ItemType Directory -Path $completionPath -Force
yams completion powershell | Out-File -Encoding utf8 "$completionPath\yams.ps1"
# Then add to your profile: . "$HOME\.config\yams\completions\yams.ps1"
```

### Verifying Installation

After installing, restart your shell or reload your profile:

```bash
# Bash
source ~/.bashrc   # or ~/.zshrc

# PowerShell
. $PROFILE
```

For zsh, make sure `compinit` runs after your `fpath` updates.

Then test by typing one of these and pressing Tab:

```bash
yams plugin trust 
yams plugins trust 
yams daemon start --log-level 
yams config search path-tree enable --mode 
```


## Symbol Extraction {#symbol-extraction}

YAMS extracts functions, classes, imports, and calls from source files via tree-sitter grammars managed under `yams config grammar`.

```bash
yams config grammar list                        # available + installed
yams config grammar download cpp python rust    # fetch specific grammars
yams add src/ --recursive --include="*.cpp,*.py,*.rs"   # symbols extracted on add
```

**Supported languages (18):** C, C++, C#, Dart, Go, Java, JavaScript, Kotlin, P4, Perl, PHP, Python, R, Rust, Solidity, SQL, TypeScript, Zig.

Source of truth: [`plugins/symbol_extractor_treesitter/symbol_extractor.cpp`](../../plugins/symbol_extractor_treesitter/symbol_extractor.cpp) (`language_configs`).

Pin an existing grammar manually with `YAMS_TS_<LANG>_LIB=/path/to/libtree-sitter-<lang>.so` (e.g. `YAMS_TS_CPP_LIB`). Requires `git` + a C compiler when grammars are built locally.

---

## Exit Codes

- 0  Success.
- 1  General error (unexpected exception or failure).
- Non-zero values may indicate specific errors depending on the subcommand.

---

## Snapshot Workflows

**New in snapshot improvements:** Efficient snapshot management with Merkle tree diffs and rename detection.

### Overview

Snapshots provide point-in-time captures of your repository content. YAMS automatically builds Merkle trees during `yams add --snapshot-id` operations, enabling fast O(log n) diff comparisons between snapshots.

### Basic Workflow

**1. Initial import (automatic snapshot):**
```bash
# Capture current state of source code
yams add . --recursive \
  --include="*.cpp,*.hpp,*.h,*.md" \
  --snapshot-label "Initial release" \
  --tags "code,release"

# Snapshot created automatically: 2025-10-01T09:00:00.000Z
```

**2. Make changes to your codebase:**
```bash
# Edit files, add features, fix bugs...
vim src/main.cpp
git commit -am "Added new feature"
```

**3. Capture updated state (automatic snapshot):**
```bash
# Just add again - snapshot created automatically
yams add . --recursive \
  --include="*.cpp,*.hpp,*.h,*.md" \
  --snapshot-label "Feature release" \
  --tags "code,release"

# New snapshot: 2025-10-01T14:30:00.000Z
```

**4. Compare snapshots:**
```bash
# List snapshots to find timestamp IDs
yams list --snapshots

# Compare using timestamp IDs (fast O(log n) tree diff)
yams diff 2025-10-01T09:00:00.000Z 2025-10-01T14:30:00.000Z

# Output example:
# Comparing v1.0 → v1.1
# 
# Added (3 files):
#   + src/new-feature.cpp
#   + src/new-feature.h
#   + tests/test_new_feature.cpp
# 
# Modified (2 files):
#   ~ src/main.cpp                 (2.3 KB → 2.5 KB)
#   ~ README.md                     (1.1 KB → 1.3 KB)
# 
# Renamed (1 file):
#   → src/utils.cpp → src/common/utils.cpp
# 
# Summary: 6 changes (+3, ~2, →1)
```

### Snapshot Discovery

**List all snapshots:**
```bash
yams list --snapshots

# Output example (chronological, newest first):
# Snapshot ID                  Label              Collection      Files  Date
# 2025-10-01T14:30:00.000Z    Feature release    -               125    2025-10-01 14:30
# 2025-09-20T10:15:00.000Z    Updated docs       documentation    45    2025-09-20 10:15
# 2025-09-15T09:00:00.000Z    Initial release    -               122    2025-09-15 09:00
```

**Filter by collection:**
```bash
yams list --snapshots --collection=releases
```

**JSON output for tools:**
```bash
yams list --snapshots --format=json | jq '.[] | select(.label | contains("release"))'
```

### File History Tracking

**View file history across snapshots:**
```bash
yams list src/main.cpp

# Output example:
# File: src/main.cpp
# 
# Snapshot History:
# v1.1  2025-10-01 14:30  Modified  hash: abc123...  2.5 KB  (current)
# v1.0  2025-09-15 09:00  Added     hash: def456...  2.3 KB
```

**View file at specific snapshot:**
```bash
# See metadata
yams list src/main.cpp --snapshot-id=v1.0

# View content
yams cat src/main.cpp --snapshot-id=v1.0

# Restore to disk
yams get --name src/main.cpp --snapshot-id=v1.0 -o main-v1.0.cpp
```

**Inline diff between snapshots:**
```bash
yams list src/main.cpp --snapshot-id=v1.0 --compare-to=v1.1

# Shows inline diff with syntax highlighting
```

### Advanced Diff Operations

**Filter by file type:**
```bash
# Show only C++ changes
yams diff v1.0 v1.1 --include="*.cpp,*.hpp"

# Exclude build artifacts
yams diff v1.0 v1.1 --exclude="build/**,*.o"
```

**Filter by change type:**
```bash
# Show only added files
yams diff v1.0 v1.1 --type=added

# Show only renames (useful for tracking file moves)
yams diff v1.0 v1.1 --type=renamed
```

**Summary statistics:**
```bash
yams diff v1.0 v1.1 --stats

# Output:
# Snapshot Comparison: v1.0 → v1.1
#   Files added:    3
#   Files deleted:  0
#   Files modified: 2
#   Files renamed:  1
#   Total changes:  6
```

**JSON output for tools:**
```bash
yams diff v1.0 v1.1 --format=json > diff.json

# Analyze with jq:
jq '.changes[] | select(.type == "modified")' diff.json
```

**Verbose output with hashes:**
```bash
yams diff v1.0 v1.1 -v

# Shows full file hashes for each change
```

### Collections and Organization

**Organize snapshots by collection:**
```bash
# Development snapshots
yams add src/ --recursive --collection=dev --snapshot-id=dev-2025-10-01

# Documentation snapshots
yams add docs/ --recursive --collection=documentation --snapshot-id=docs-v3

# Release snapshots
yams add . --recursive --collection=releases --snapshot-id=v2.0 --snapshot-label="Major release"
```

**Compare within collections:**
```bash
yams list --snapshots --collection=releases
yams diff v1.0 v2.0  # Compare releases
```

### Restoration Workflows

**Restore entire snapshot:**
```bash
# Restore to specific directory
yams restore --snapshot-id=v1.0 --output=./restore-v1.0

# Preview without writing files
yams restore --snapshot-id=v1.0 --output=./test --dry-run

# Overwrite existing files
yams restore --snapshot-id=v1.0 --output=. --overwrite
```

**Selective restoration:**
```bash
# Restore only specific file types
yams restore --snapshot-id=v1.0 \
  --output=./restore \
  --include="*.cpp,*.h"

# Exclude directories
yams restore --snapshot-id=v1.0 \
  --output=./restore \
  --exclude="build/**,third_party/**"
```

**Restore by collection:**
```bash
yams restore --collection=releases --output=./releases
```

### Performance Characteristics

**Tree-based diff performance:**
- **Small changes (< 1%)**: O(log n) subtree matching, ~100x faster than flat diff
- **Large changes (> 50%)**: Approaches O(n) but still faster due to hash-based comparison
- **Rename detection**: Hash-based matching is O(n) but only runs on Added/Deleted pairs

**Storage efficiency:**
- Merkle trees are stored once in CAS and deduplicated across snapshots
- Unchanged subtrees share the same tree hash (deduplication)
- Tree overhead: ~5-10% additional storage for large repositories
- Diff computation: ~5-10% slower during `yams add --snapshot-id` (tree building)

**Optimization tips:**
- Use `--exclude` patterns to skip build artifacts, caches, and temporary files
- Use `--no-renames` for faster diff when rename tracking isn't needed
- Pre-build trees during snapshot creation for instant diff operations

### Integration with Git

**Automatic git integration:**
When `yams add` runs in a git repository, snapshot metadata automatically includes:
- `git_commit`: Current commit hash (e.g., `abc123def456`)
- `git_branch`: Current branch name (e.g., `main`, `develop`)
- `git_remote`: Remote repository URL (if configured)

This creates a natural bidirectional link between YAMS snapshots and git history.

**Snapshot on git commit (hook):**
```bash
#!/bin/bash
# Hook: .git/hooks/post-commit
# Automatically snapshot on every commit
yams add . --recursive \
  --include="*.cpp,*.hpp,*.h,*.md" \
  --snapshot-label "$(git log -1 --pretty=%B | head -1)" \
  --tags "code,git,auto"
```

**Snapshot on git tag:**
```bash
#!/bin/bash
# Hook: .git/hooks/post-tag
TAG=$(git describe --tags --abbrev=0)
yams add . --recursive \
  --include="*.cpp,*.hpp,*.h,*.md" \
  --snapshot-label="Git tag: $TAG" \
  --tags "code,git,release"
```

**Daily snapshots:**
```bash
#!/bin/bash
# Cron: daily snapshot at midnight
yams add . --recursive \
  --include="*.cpp,*.hpp,*.h,*.md" \
  --snapshot-label="Daily backup" \
  --tags "code,daily"
# Snapshot ID is auto-generated with current timestamp
```

### Troubleshooting

**Missing trees:**
```bash
# If diff is slow, trees may not exist
# Rebuild trees for existing snapshots (not yet exposed via CLI).
# Re-run yams add to regenerate trees as needed.
```

**Large diff output:**
```bash
# Use filters to reduce output
yams diff v1.0 v2.0 --type=modified --include="src/**"
yams diff v1.0 v2.0 --stats  # Summary only
```

**Restore conflicts:**
```bash
# Preview first
yams restore --snapshot-id=v1.0 --output=./test --dry-run

# Use overwrite cautiously
yams restore --snapshot-id=v1.0 --output=. --overwrite
```

### Best Practices

1. **Use labels liberally:** Labels make snapshot timelines human-readable (`--snapshot-label "Before refactor"`)
2. **Leverage automatic IDs:** No need to manually manage snapshot IDs - timestamps provide natural ordering
3. **Exclude build artifacts:** Use `--exclude="build/**,*.o,*.so"` to reduce noise and improve performance
4. **Tag meaningful snapshots:** Use `--tags` for filtering (e.g., `release`, `milestone`, `backup`)
5. **Collections for organization:** Group related content (`--collection=releases`, `--collection=docs`)
6. **Automate with git hooks:** Snapshot on every commit or tag for complete history tracking
7. **Test restores:** Verify snapshots with `--dry-run` before overwriting files
8. **Trust the timestamps:** Snapshot IDs are ISO 8601 timestamps - sort naturally and are globally unique

---

## download {#cmd-download}

Download an artifact and store it into YAMS, or manage daemon-tracked download jobs.

Synopsis:
- `yams download <url> [options]`
- `yams download --status <job-id> [--json]`
- `yams download --list-jobs [--json]`
- `yams download --cancel <job-id> [--json]`

Options:
- `<url>`
  - Artifact URL to download.
- `--checksum <algo:hex>`
  - Expected checksum. This is strongly recommended for direct downloads and may be required by daemon policy.
- `--status <job-id>`
  - Show the current state of a daemon-managed download job.
- `--list-jobs`
  - List tracked daemon download jobs.
- `--cancel <job-id>`
  - Cancel a daemon-managed download job.
- `--json`
  - Emit structured output for direct download results or job/status queries.

Description:
- Default behavior is store-only: downloaded content is verified and finalized into YAMS CAS.
- The daemon path is policy-gated and disabled by default.
- When daemon download is enabled, accepted requests are backgrounded as tracked jobs.
- Current daemon job control supports:
  - `start` via `yams download <url>`
  - `status`
  - `list-jobs`
  - `cancel`
- The daemon path is intentionally narrow:
  - HTTPS-style allowlist policy is expected
  - store-only behavior is enforced
  - checksum policy may reject requests before network activity
  - arbitrary export paths, custom headers, proxy overrides, and broader transport tuning are not part of the daemon job surface

Notes:
- The daemon feature gate is `YAMS_ENABLE_DAEMON_DOWNLOAD`.
- A successful daemon-backed start returns a `jobId`; use that with `--status` or `--cancel`.
- `cancel` is cooperative: the job transitions to `canceled`, and active downloader work is asked to stop.
- Resume-by-job is not implemented yet; current downloader resume behavior is still internal to the downloader layer.

Examples:
```bash
# Direct store-only download
yams download "https://example.com/archive.tar.gz" --checksum sha256:<hex>

# Start a daemon-managed download when the feature is enabled
YAMS_ENABLE_DAEMON_DOWNLOAD=1 yams download "https://example.com/archive.tar.gz" --checksum sha256:<hex> --json

# Check daemon job status
yams download --status job-123 --json

# List tracked daemon download jobs
yams download --list-jobs

# Cancel a running daemon download job
yams download --cancel job-123
```

## Tips

YAMS-first workflow
- Always search the codebase with YAMS (search/grep). Do not use system grep/find/rg for repository queries.
- Enable auto-ingest with `yams watch` when possible; otherwise re-index with `yams add`.
- Use --paths-only for path lists you can pipe into further commands or editors.
- Prefer --include with comma-separated patterns (e.g., "*.cpp,*.hpp,*.h") to bound searches and indexing.
- Retrieve exact files for review with yams get --name <path> -o <dest>.
- Keep tags consistent (e.g., code,source; code,headers; code,working) and update metadata with yams update as needed.

- Use --json where supported to integrate with scripts and tools.
- Specify a storage directory explicitly with --storage or via YAMS_STORAGE to keep data separate for testing vs production.
- When available, use --help --verbose for detailed per-command help, or yams --help-all for the full reference in the terminal.

---

## Links

- **Docs:** [yamsmemory.ai](https://yamsmemory.ai)
- **GitHub:** [github.com/trvon/yams](https://github.com/trvon/yams)
- **Discord:** [discord.gg/rTBmRHdTEc](https://discord.gg/rTBmRHdTEc)
docs/user_guide/cli.md