Command-Line Interface

The LocalVectorDB CLI provides comprehensive tools for database management, document operations, and server administration.

Exit Codes

Commands exit 0 on success and a non-zero code on failure. The semantic codes deliberately skip 2, which Click reserves for its own usage errors (unknown option, bad argument, missing value), so a configuration problem can never be confused with a usage mistake.

Code	Meaning
`0`	Success
`1`	Generic runtime error (operation failed, not found, invalid input)
`2`	Usage error — emitted by Click for unknown options, bad arguments, or missing values (not raised by the commands themselves)
`3`	Ollama check failed (`lvdb serve` could not find or reach Ollama)
`4`	Permission error (e.g. filesystem permission denied writing a config file or deleting a database)
`5`	Configuration error (config file missing, unreadable, invalid, or a bad `config`/`tuning` target)

Installation

For the CLI alone (everything except lvdb serve), install the light cli extra:

pip install "localvectordb[cli]"

# Verify installation
lvdb --help

To also run the HTTP server (lvdb serve), install the server extra, which includes the CLI:

pip install "localvectordb[server]"

Server Management

Starting the Server

# Basic server start
lvdb serve

# Custom configuration (--config is a global option: it goes BEFORE the subcommand)
lvdb --config production.toml serve --host 0.0.0.0 --port 8080

# Development mode
lvdb serve --debug --log-level DEBUG

# Specific database folder (--db-folder is also a global option)
lvdb --db-folder /data/vector_databases serve

# Disable Ollama check
lvdb serve --disable-ollama-check

Options:

--host, -H: Interface to bind to (default: 127.0.0.1)
--port, -p: Port to bind to (default: 8000; falls back to server.port from config)
--debug: Enable debug mode
--log-level, -l: Set logging level (DEBUG, INFO, WARNING, ERROR, CRITICAL)
--disable-ollama-check, -x: Skip Ollama availability check

Note

The Ollama availability check runs on startup only when the configured embedding provider is Ollama. A server backed by OpenAI, Jina, OpenRouter, or any other provider does not require a local Ollama install. --disable-ollama-check still overrides the probe when it would otherwise run.

Note

--config, -c, --db-folder, -d, --verbose, -v and --quiet, -q are global options on the top-level lvdb group, not on serve. They must be given before the subcommand, e.g. lvdb --config production.toml serve or lvdb --verbose db mydb add file.txt. (-h is reserved for --help; the host short flag is -H.)

Global options (given before the subcommand):

--config, -c: Path to config file (env: LVDB_SERVER_CONFIG)
--db-folder, -d: Directory containing vector databases (env: LVDB_DATABASE_ROOT_DIR)
--verbose, -v: Enable verbose (DEBUG) logging
--quiet, -q: Only log errors (suppress warnings/info)
--version, -V: Print the installed localvectordb version

Note

A configuration file is not required for database operations. When neither a config file nor --db-folder is given, the current working directory is used as the database folder, so lvdb db <command> --db <name> works in any folder that contains a database.

Configuration Management

For a complete overview of the configuration settings, see the Configuration Parameters Documentation.

Initialize Configuration

# Create default TOML configuration
lvdb config init

# Create with predefined schema
lvdb config init --schema research_papers --output research.toml

View Configuration

# Show current configuration
lvdb config show

# Show specific section
lvdb config show --section database

# Show in different format
lvdb config show --format json

# Show from specific file (--config is global; it precedes the subcommand)
lvdb --config /path/to/config.toml config show

Example Output:

[database]
root_dir = "./.lvdb"
timeout = 300
connection_pool_size = 10
enable_gpu = false
enable_fts = true
chunk_size = 500
chunk_overlap = 1
chunking_method = "lines"

[embedding]
provider = "ollama"
model = "embeddinggemma"
batch_size = 64
timeout = 30
max_retries = 3

[server]
host = "127.0.0.1"
port = 8000
log_level = "INFO"
log_format = "%(asctime)s - %(name)s - %(levelname)s - %(message)s"
max_request_size = 104857600

[server.security]
require_api_key = false
api_key_header = "Authorization"
auto_prune_expired_keys = false
key_audit_logging = true
auth_log_level = "INFO"
warn_expiring_days = 7
cors_enabled = true
cors_allowed_origins = "*"
cors_allowed_methods = ["GET", "POST", "PUT", "DELETE", "OPTIONS"]
cors_allowed_headers = ["Content-Type", "Authorization"]
cors_max_age = 86400

Get Configuration Values

Get individual configuration values using dot notation:

# Get basic configuration values
lvdb config get server.host
lvdb config get server.port
lvdb config get database.chunk_size
lvdb config get embedding.model

# Get security settings
lvdb config get server.security.require_api_key
lvdb config get server.security.cors_enabled
lvdb config get server.security.cors_allowed_origins

# Get metadata schema fields
lvdb config get database.metadata_schema.title
lvdb config get database.metadata_schema.author

# Different output formats
lvdb config get server.port --format raw         # Just the value
lvdb config get server.port --format json        # JSON format
lvdb config get server.port --format pretty      # Formatted (default)

Options:

--format, -f: Output format (raw, json, pretty)

Example Output:

$ lvdb config get server.security.require_api_key
Configuration: server.security.require_api_key
===============================================
false

$ lvdb config get server.security.cors_allowed_origins --format json
["*"]

$ lvdb config get database.chunk_size --format raw
500

Set Configuration Values

Set configuration values using dot notation with intelligent type conversion:

# Set basic values
lvdb config set server.port 8080
lvdb config set server.host "0.0.0.0"
lvdb config set database.chunk_size 1000

# Set boolean values
lvdb config set server.security.require_api_key true
lvdb config set database.enable_fts false

# Set list values (JSON format)
lvdb config set server.security.cors_allowed_origins '["http://localhost:3000", "https://app.example.com"]'

# Set list values (comma-separated)
lvdb config set server.security.cors_allowed_methods "GET,POST,PUT,DELETE"

# Set null values
lvdb config set server.security.key_database_path null

# Set metadata schema fields (JSON format)
lvdb config set database.metadata_schema.category '{"type": "text", "indexed": true, "required": false}'

# Preview changes without saving
lvdb config set server.port 9000 --dry-run

# Skip confirmation prompt
lvdb config set embedding.model "all-minilm-l6-v2" --force

Options:

--dry-run, -n: Show what would be changed without saving
--force, -y: Skip confirmation prompt

Type Conversion:

The set command automatically converts string inputs to the appropriate type:

Booleans: true, false, yes, no, 1, 0, on, off
Integers: 8080, 500, -1
Floats: 0.8, 3.14
Lists: JSON format ["item1", "item2"] or comma-separated item1,item2
Objects: JSON format {"key": "value"}
Null values: null, none, or empty string

Example Session:

$ lvdb config set server.security.require_api_key true
Configuration Change:
=====================
Key: server.security.require_api_key
Old value: false
New value: true

Apply this change? [y/N]: y

✓ Configuration updated and saved to server-cfg.toml

$ lvdb config set server.security.cors_allowed_origins '["http://localhost:3000"]' --dry-run
Configuration Change:
=====================
Key: server.security.cors_allowed_origins
Old value: ["*"]
New value: ["http://localhost:3000"]

[DRY RUN] No changes made.

Common Configuration Tasks:

# Enable API key authentication
lvdb config set server.security.require_api_key true

# Change server binding
lvdb config set server.host "0.0.0.0"
lvdb config set server.port 8080

# Configure CORS for web applications
lvdb config set server.security.cors_allowed_origins '["http://localhost:3000", "https://app.company.com"]'

# Adjust chunking parameters
lvdb config set database.chunk_size 750
lvdb config set database.chunk_overlap 2

# Change embedding model
lvdb config set embedding.model "all-minilm-l6-v2"
lvdb config set embedding.provider "openai"

# Enable GPU acceleration
lvdb config set database.enable_gpu true

# Configure logging
lvdb config set server.log_level "DEBUG"

Dot Notation Reference:

Configuration values can be accessed using dot notation:

Database Settings (database.*):

database.root_dir - Database storage directory
database.chunk_size - Maximum tokens per chunk
database.chunk_overlap - Overlap between chunks
database.chunking_method - Chunking strategy
database.enable_gpu - GPU acceleration
database.enable_fts - Full-text search
database.connection_pool_size - Connection pool size
database.timeout - Database timeout

Embedding Settings (embedding.*):

embedding.provider - Provider (ollama, openai)
embedding.model - Model name
embedding.base_url - Provider base URL
embedding.api_key - API key for providers
embedding.batch_size - Batch processing size
embedding.timeout - Request timeout
embedding.max_retries - Maximum retry attempts

Server Settings (server.*):

server.host - Server host address
server.port - Server port
server.log_level - Logging level
server.log_format - Log message format
server.max_request_size - Maximum request size

Security Settings (server.security.*):

server.security.require_api_key - Enable API key auth
server.security.api_key_header - Auth header name
server.security.key_database_path - API key database path
server.security.auto_prune_expired_keys - Auto cleanup
server.security.key_audit_logging - Audit key usage
server.security.cors_enabled - Enable CORS
server.security.cors_allowed_origins - Allowed origins
server.security.cors_allowed_methods - Allowed methods
server.security.cors_allowed_headers - Allowed headers

Metadata Schema (database.metadata_schema.*):

database.metadata_schema.<field_name> - Individual schema fields

Example Metadata Schema Configuration:

# Add a new metadata field
lvdb config set database.metadata_schema.priority '{"type": "integer", "indexed": true, "default_value": 1}'

# Add a simple text field
lvdb config set database.metadata_schema.department '{"type": "text", "indexed": true}'

# Add a JSON field for tags
lvdb config set database.metadata_schema.tags '{"type": "json"}'

Security Configuration Example:

# Complete security setup
lvdb config set server.security.require_api_key true
lvdb config set server.security.cors_enabled true
lvdb config set server.security.cors_allowed_origins '["https://app.company.com"]'
lvdb config set server.security.key_audit_logging true
lvdb config set server.security.auto_prune_expired_keys true

# Verify security settings
lvdb config get server.security.require_api_key
lvdb config get server.security.cors_allowed_origins

Authentication and API Key Management

The LocalVectorDB CLI provides comprehensive API key management with permission-based access control for secure server access.

Permission Levels

API keys support two permission levels:

read_only: Grants access to query, search, and read operations only. Cannot create, update, or delete any resources.
read_write: Full access to all operations including database creation, document management, and administrative functions.

API keys are stored in {root_dir}/api_keys.db by default. To change the location of this file, set the configuration setting key_database_path = "path/to/your/api_keys.db" in the [server.security] section.

Check Authentication Status

# Show current authentication configuration
lvdb auth status

# Show in JSON format
lvdb auth status --format json

Example Output:

Authentication Status
Configuration file: /home/user/server-cfg.toml
API Authentication: Enabled

Database-managed Keys:
  Status: Available
  Total keys: 5
  Active keys: 4
  Expired keys: 1
  Expiring soon (7 days): 1
  Recently used (24h): 2

⚠️  1 key(s) expiring within 7 days

Create API Keys

API keys support permission levels for fine-grained access control:

read_only: Can query and search, but cannot create, update, or delete resources
read_write: Full access to all operations (default)

# Create a basic read-write API key (default)
lvdb auth create-key

# Create a read-only key for monitoring/analytics
lvdb auth create-key --description "Monitoring Dashboard" --permission-level read_only

# Create a read-write key with expiration
lvdb auth create-key --description "Admin Access" --permission-level read_write --expires-days 30

# Create a read-only key for public API
lvdb auth create-key --description "Public Search API" \
                     --permission-level read_only \
                     --expires-days 365 \
                     --created-by "api-team"

# JSON output for scripting
lvdb auth create-key --description "Script Access" --permission-level read_only --format json

# Key-only output for automation
lvdb auth create-key --permission-level read_write --format key-only

Options:

--description, -d: Human-readable description of the key’s purpose
--permission-level, -p: Permission level (read_only, read_write) - defaults to read_write
--expires-days: Number of days until key expires (omit for no expiration)
--created-by: Identifier of who is creating the key
--format, -f: Output format (table, json, key-only)

Example Output:

✓ API Key Created Successfully

Key Details:
  Key ID: key_20241201_abc123
  Description: Production API Access
  Permission Level: read_write
  Created: 2024-12-01 10:30:00 UTC
  Expires: Never

API Key (save this now - it won't be shown again):
  lvdb_XyZ9k2mN7qP4wR8tL3vB5nM9kJ7hF2dS6xW1qE4yT8rL9pN3mK5

⚠️  Store this key securely - it cannot be retrieved again!

List API Keys

# List all API keys
lvdb auth list-keys

# List only active keys
lvdb auth list-keys --active-only

# Exclude expired keys
lvdb auth list-keys --no-expired

# Show key management statistics
lvdb auth list-keys --show-stats

# JSON output
lvdb auth list-keys --format json

Example Output:

Key Management Statistics:
  Total keys: 5
  Active keys: 4
  Expired keys: 1
  Expiring soon (7 days): 1
  Recently used (24h): 2

API Keys:

ID                   Description              Permission   Status     Created      Expires      Last Used
-----------------------------------------------------------------------------------------------------------------------------------
key_20241201_abc123  Production API Access    read_write   ACTIVE     2024-12-01   Never        2024-12-01
key_20241125_def456  CI/CD Pipeline          read_write   ACTIVE     2024-11-25   2024-12-25   Never
key_20241120_ghi789  Monitoring Dashboard    read_only    ACTIVE     2024-11-20   Never        2024-11-26
key_20241115_jkl012  Development Access      read_write   EXPIRED    2024-11-15   2024-11-22   2024-11-21

Get Key Information

# Show detailed key information
lvdb auth key-info key_20241201_abc123

# JSON output
lvdb auth key-info key_20241201_abc123 --format json

Example Output:

API Key Information: key_20241201_abc123

Basic Information:
  ID: key_20241201_abc123
  Description: Production API Access
  Permission Level: read_write
  Created by: admin@company.com

Status:
  Status: ACTIVE

Dates:
  Created: 2024-12-01 10:30:00 UTC
  Expires: Never
  Last used: 2024-12-01 15:45:00 UTC

Revoke API Keys

# Revoke a key (with confirmation)
lvdb auth revoke-key key_20241201_abc123

# Force revoke without confirmation
lvdb auth revoke-key key_20241201_abc123 --confirm

Example Output:

Key Details:
  ID: key_20241201_abc123
  Description: Production API Access
  Created: 2024-12-01 10:30:00 UTC

Are you sure you want to revoke key 'key_20241201_abc123'? [y/N]: y
✓ Key 'key_20241201_abc123' has been revoked.

Rotate API Keys

# Rotate a key (creates new, revokes old)
lvdb auth rotate-key key_20241201_abc123

# JSON output for automation
lvdb auth rotate-key key_20241201_abc123 --format json

# Key-only output
lvdb auth rotate-key key_20241201_abc123 --format key-only

Example Output:

✓ API Key Rotated Successfully

Original Key:
  ID: key_20241201_abc123 (now revoked)

New Key Details:
  Key ID: key_20241201_xyz789
  Description: Rotated from key_20241201_abc123: Production API Access
  Created: 2024-12-01 16:00:00 UTC
  Expires: Never

New API Key (save this now):
  lvdb_A1b2C3d4E5f6G7h8I9j0K1l2M3n4O5p6Q7r8S9t0U1v2W3x4Y5z6

⚠️  Update your applications with the new key!

Clean Up Expired Keys

# Preview what would be cleaned up
lvdb auth prune-expired --dry-run

# Soft delete (deactivate) expired keys
lvdb auth prune-expired

# Hard delete (permanently remove) expired keys
lvdb auth prune-expired --hard-delete

# Skip confirmation
lvdb auth prune-expired --confirm

Example Output:

Found 2 expired key(s):

  key_20241120_ghi789: Temporary Access (expired 4 days ago)
  key_20241110_mno345: Test Key (expired 21 days ago)

Are you sure you want to deactivate these 2 expired keys? [y/N]: y
✓ 2 expired key(s) deactivated.

Options:

--soft-delete/--hard-delete: Deactivate vs permanently remove (default: soft-delete)
--dry-run, -n: Show what would be pruned without doing it
--confirm, -y: Skip confirmation prompt

API Key Usage in Applications

Once you have created an API key, use it in your applications:

cURL Example:

# Using the API key in requests
curl -H "Authorization: Bearer lvdb_XyZ9k2mN7qP4wR8tL3vB5nM9kJ7hF2dS6xW1qE4yT8rL9pN3mK5" \
     -H "Content-Type: application/json" \
     -X POST http://localhost:8000/api/v1/databases/my_database/query \
     -d '{"query": "machine learning", "search_type": "vector", "k": 5}'

Python Example:

import requests

headers = {
    "Authorization": "Bearer lvdb_XyZ9k2mN7qP4wR8tL3vB5nM9kJ7hF2dS6xW1qE4yT8rL9pN3mK5",
    "Content-Type": "application/json"
}

response = requests.post(
    "http://localhost:8000/api/v1/databases/my_database/query",
    headers=headers,
    json={"query": "neural networks", "search_type": "vector", "k": 10}
)

Environment Variable:

# Set as environment variable
export LVDB_API_KEY="lvdb_XyZ9k2mN7qP4wR8tL3vB5nM9kJ7hF2dS6xW1qE4yT8rL9pN3mK5"

# Use in application
curl -H "Authorization: Bearer $LVDB_API_KEY" \
     -X GET http://localhost:8000/api/v1/databases/my_database/info

Security Best Practices

Key Management:

Rotate keys regularly: Use lvdb auth rotate-key for regular key rotation
Set expiration dates: Use --expires-days for temporary access
Use descriptive names: Always provide meaningful descriptions
Monitor usage: Regularly check lvdb auth list-keys --show-stats
Clean up expired keys: Use lvdb auth prune-expired to maintain hygiene

Key Storage:

Store keys securely (password managers, secrets management systems)
Never commit keys to version control
Use environment variables for applications
Rotate immediately if a key is compromised

Access Control:

Create separate keys for different applications/users
Use short-lived keys for temporary access
Revoke keys immediately when no longer needed
Monitor the “Last Used” column to identify unused keys

Example Security Workflow:

# Weekly security routine

# 1. Check for keys expiring soon
lvdb auth list-keys --show-stats

# 2. Clean up expired keys
lvdb auth prune-expired --dry-run
lvdb auth prune-expired

# 3. Rotate long-lived keys (quarterly)
lvdb auth rotate-key old_production_key

# 4. Check for unused keys
lvdb auth list-keys | grep "Never" | awk '{print $1}'

# 5. Review key descriptions and purposes
lvdb auth list-keys

Database Operations

List Databases

# List all databases
lvdb list

# List with details
lvdb list --details

# Specify database folder (--db-folder is a global option, before the subcommand)
lvdb --db-folder /path/to/databases list

Options:

--details: Show document/chunk count, model, and chunk method for each database

The database folder and config file are selected with the global options --db-folder, -d and --config, -c (given before list), e.g. lvdb --config server.toml list --details.

Example Output:

lvdb list --details

Databases in /home/user/.lvdb
Name                     Documents Chunks    Model                   Method
====================================================================================
research_papers         1250      8500      embeddinggemma        sentences
customer_support        845       3200      all-minilm              sentences
code_documentation      324       2100      embeddinggemma        code-blocks

Create Database

# Basic database creation
lvdb create my_database

# With custom configuration
lvdb create research_db \
  --embedding-model embeddinggemma \
  --chunk-size 600 \
  --chunking-method sections \
  --metadata-schema research_papers

# With specific provider
lvdb create openai_db \
  --embedding-provider openai \
  --embedding-model text-embedding-3-small

Options:

--embedding-model: Embedding model to use
--embedding-provider: Provider (ollama, openai)
--chunk-size: Maximum tokens per chunk
--chunking-method: Chunking strategy. One of sentences, tokens, words, lines, characters, paragraphs, sections, structure, code-blocks, delimiter (see Chunking)
--chunk-overlap: Overlap between chunks
--chunk-delimiter: Delimiter for --chunking-method delimiter (default: a blank line). The escapes \n, \t, \r are interpreted, e.g. --chunk-delimiter '\n---\n'; the value is persisted with the database
--metadata-schema: Predefined schema (files, documents, research_papers, code_repository, customer_support)
--reranker-provider: Persist a default reranker for the database (e.g. sentence_transformers, jina). Every search — CLI, library, MCP, server — then reranks with it automatically; --no-rerank on db <name> search disables it per call
--reranker-model: Model for --reranker-provider (provider default when omitted)

Delete Database

# Delete database (with confirmation)
lvdb delete old_database

# Force delete without confirmation
lvdb delete old_database --confirm

Rename Database

# Rename a database
lvdb rename old_database new_database

rename OLD NEW moves the database’s on-disk files rather than rewriting anything internally: the SQLite database (old.sqlite and its -wal/-shm companions), the FAISS index (old.faiss), and any hierarchical sidecar indexes (old_sections.faiss, old_documents.faiss) are renamed to the new prefix. The command aborts if OLD does not exist or if a database named NEW already exists.

Database-Specific Operations

All database-specific operations use the lvdb db <command> --db <database_name> form (or set LVDB_DB / database.default_database once and omit --db).

The database is resolved in this order: the --db flag (accepted both before and after the subcommand), the LVDB_DB environment variable, then database.default_database in the config file – so a script or agent can set the database once and every call is just lvdb db search "...". The pre-v0.2.0 positional form lvdb db <database_name> <command> still works but is deprecated (a warning is printed) and will be removed in a future release.

Database Information

# Show database info
lvdb db info --db research_papers

# Show detailed statistics
lvdb db stats --db research_papers

Example Output:

Database Info
-------------
  Database: research_papers
  Path: /home/user/.lvdb
  Embedding model: embeddinggemma
  Embedding provider: ollama
  Chunk size: 500
  Chunking method: sentences
  FTS search available: True
  Total Documents: 1250
  Total Chunks: 8500
  Metadata fields: 4

Corpus Diagnosis (doctor)

Several retrieval knobs (chunk_size, search_level, section_weight, vector_weight, …) have no defensible global default — their best values are properties of your corpus, and they disagree across corpora by more than any tuning effect. doctor measures which regime your corpus is in, from the built index, in seconds:

lvdb db doctor --db research_papers

It reports:

Encoder coverage of chunk text — the share of ingested text that actually fits the embedding model’s context window. Text past it never enters any vector, silently; at a measured 65% coverage this cost 0.109 nDCG@10. Token counts use the encoder’s own tokenizer where one is importable (sentence-transformers, local HuggingFace, OpenAI); otherwise they are labelled estimated rather than reported as confident percentages.
Section length against the encoder window — long sections are windowed and mean-pooled (never truncated), but pooled rawspan vectors degrade as spans grow; a mostly-long corpus wants section_vector_strategy="centroid".
Section reachability — the share of sections reachable by return_type="sections" roll-up. Chunks credit every section they overlap, so this is ~100% on a healthy database; a shortfall signals an incomplete chunk_sections backfill or a hierarchy needing rebuild_hierarchical_embeddings().
Fanout — chunks per document/section; at fanout ~1 every document-scoring aggregator is a no-op.
Keyword-leg health — whether each FTS table is present and in step with its base table. The keyword leg is worth +0.08 to +0.13 nDCG@10 where present.

The same report is available in the library as db.diagnose() (returning a structured DiagnoseReport with a summary string), and a related warning fires automatically at ingest when measured coverage falls substantially — mild truncation is measurably free, so the warning is reserved for real loss.

Document Management

Add Documents

# Add single file
lvdb db add document.txt --db my_database

# Add multiple files
lvdb db add file1.txt file2.txt --db my_database
# Or with globs
lvdb db add *.txt --db my_database

# Add with metadata
lvdb db add paper.txt --metadata '{"title": "Research Paper", "author": "Dr. Smith"}' --db my_database

# Add from stdin
cat document.txt | lvdb db add - --db my_database

# Add with custom IDs
lvdb db add file1.txt file2.txt --id "doc_1,doc_2" --db my_database

Note

Rich document formats (PDF, DOCX, HTML, CSV, XLSX, …) are automatically extracted to Markdown; plain text, Markdown, and source files are read directly as UTF-8. Use --extract/--no-extract to force or disable extraction. See File Extraction System.

Note

Unless --id is given, documents added from files use the filename without extension as their id (matching db.upsert_from_file(...)), so adding the same file again updates the existing document. Text and stdin input gets an auto-generated id.

Get Documents

# Get document by ID
lvdb db get doc_1 --db my_database

# Get with metadata
lvdb db get doc_1 --metadata --db my_database

# Pretty formatted output
lvdb db get doc_1 --pretty --db my_database

# Save to file
lvdb db get doc_1 --output retrieved_doc.txt --db my_database

# JSON output
lvdb db get doc_1 --format json --db my_database

By default get returns the whole document. The following flags return a part of the document instead and are mutually exclusive (only one may be given per invocation). They compose with --format, -f, --pretty, --metadata and --output.

# A single chunk (0-based index) or an inclusive chunk range, as indexed
lvdb db get doc_1 --chunk 3 --db my_database
lvdb db get doc_1 --chunk 2:5 --db my_database

# A character slice (0-based, end-exclusive — like Python's content[M:N])
lvdb db get doc_1 --range 0:200 --db my_database

# A line range (1-based, inclusive)
lvdb db get doc_1 --lines 10:20 --db my_database

# A section by its Markdown heading (case-insensitive)
lvdb db get doc_1 --section "Installation" --db my_database

# The document's section outline (headings, levels, start lines)
lvdb db get doc_1 --outline --db my_database

Selection options:

--chunk M[:N]: Return the persisted chunk at 0-based index M (as it was stored at ingest), or the inclusive range M:N. Open-ended forms M:/:N/: are accepted. In --format, -f mode each chunk is emitted with its index, content and full position.
--range M:N: Return the character slice content[M:N] (0-based, end-exclusive). Open-ended forms are accepted.
--lines M:N: Return lines M through N (1-based, inclusive). Open-ended forms are accepted.
--section NAME: Return the body of the section whose Markdown heading matches NAME (case-insensitive). Sections are detected on the fly from the document content (code-fence aware), so this works regardless of whether hierarchical embeddings were enabled at ingest. If no heading matches, the available headings are listed.
--outline: Print the document’s section outline as an indented tree (or a JSON list of {index, heading, level, start_line, end_line} with --format, -f).

Update Documents

# Update content from file
lvdb db update doc_1 new_content.txt --db my_database

# Update from stdin
echo "New content" | lvdb db update doc_1 - --db my_database

# Update metadata only
lvdb db update doc_1 --metadata '{"status": "revised"}' --db my_database

Patch Documents (in-place edit)

Edit a document without re-sending its whole content. --find must match exactly --count times (default 1) or the patch fails; --append and --prepend cover the no-anchor cases. Combine ops as long as they touch disjoint spans.

# Exact find/replace
lvdb db patch doc_1 --find "brown" --replace "red" --db my_database

# Replace all N occurrences (fails unless exactly N match)
lvdb db patch doc_1 --find "TODO" --replace "DONE" --count 3 --db my_database

# Append / prepend
lvdb db patch doc_1 --append " (revised)" --db my_database
lvdb db patch doc_1 --prepend "DRAFT: " --db my_database

# Optimistic concurrency: fail if the document changed since you read it
lvdb db patch doc_1 --find "v1" --replace "v2" --expect-hash 9f2b... --db my_database

Delete Documents

# Delete document
lvdb db delete doc_1 --db my_database

List Documents

# List document IDs
lvdb db list --db my_database

# List with pagination
lvdb db list --limit 20 --offset 40 --db my_database

# Save to file
lvdb db list --output doc_ids.txt --db my_database

# JSON format
lvdb db list --format json --db my_database

Browse Document IDs

ls navigates “filesystem-like” document ids S3-style: it treats a delimiter (/ by default) as a virtual path separator and lists the immediate children of a prefix. Virtual folders are shown with a trailing delimiter and a document count; leaf documents at the level are shown as-is. Omit the prefix to list the top level.

# List the top-level "folders" and documents
lvdb db ls --db my_database

# List the children of a prefix
lvdb db ls docs/ --db my_database

# Use a custom delimiter
lvdb db ls "a::" --delimiter "::" --db my_database

# JSON output
lvdb db ls docs/ --format json --db my_database

Options:

--delimiter, -d: Virtual path separator (default: /)
--format, -f: Output format (table or json, default table); -j is a shortcut for --format json

Search Operations

# Basic search (hybrid vector + keyword, the default)
lvdb db search "machine learning" --db my_database

# Vector-only search
lvdb db search "machine learning" --search-type vector --db my_database

# Keyword search
lvdb db search "neural networks" --search-type keyword --db my_database

# Hybrid search with a custom vector/keyword blend
lvdb db search "AI algorithms" --search-type hybrid --vector-weight 0.8 --db my_database

# Limit results
lvdb db search "deep learning" --limit 3 --db my_database

# Return chunks instead of documents
lvdb db search "optimization" --return-type chunks --db my_database

# Return matching chunks with surrounding context (2 neighbouring chunks each side)
lvdb db search "optimization" --return-type context --context-window 2 --db my_database

# Size the context by a token budget instead of a chunk count, hard-truncating to fit
lvdb db my_database search "optimization" \
  --return-type enriched --context-unit tokens --context-window 400 --context-truncate

# Search with metadata filter
lvdb db search "research" --metadata-filter '{"journal": "Science"}' --db my_database

# Minimum score threshold
lvdb db search "neural" --score-threshold 0.8 --db my_database

# Pretty output
lvdb db search "machine learning" --pretty --db my_database

# Save results
lvdb db search "AI" --output search_results.txt --db my_database

# JSON output
lvdb db search "algorithms" --format json --metadata --db my_database

Options:

--limit, -n: Maximum number of results (default: 5)
--search-type, -t: Search method (vector, keyword, hybrid) - defaults to hybrid (matching the API/MCP default)
--return-type, -r: What to return (documents, chunks, context, enriched, sections) - defaults to documents
--search-level: Which index to search (chunks, sections, documents, fused) - defaults to chunks
--score-threshold: Minimum score threshold (default: 0.0)
--vector-weight: Weight for the vector component in hybrid search (default: 0.5)
--context-window: Context size for --return-type context/enriched, measured in --context-unit (default: 2)
--context-unit: Unit for --context-window (chunks, tokens, words, characters) - defaults to chunks
--context-truncate: Hard-truncate assembled context to exactly the budget (non-chunk --context-unit only)
--metadata-filter: Metadata filter in JSON format
--rerank/--no-rerank: --rerank asserts reranking happens (via the database’s persisted default or --rerank-provider) and fails if neither exists; --no-rerank disables the persisted default for this search. Omitted: the persisted default applies when configured
--rerank-provider: One-off reranker for this search (e.g. sentence_transformers, jina), overriding any persisted default
--rerank-model: Model for --rerank-provider (provider default when omitted)
--rerank-k: Candidate-pool size handed to the reranker before truncating to --limit (default 5 * limit, clamped to 200)
--metadata/--no-metadata: Include metadata in output
--pretty, -p: Human-readable, titled output
--format, -f: Output format (table or json, default table); -j is a shortcut for --format json
--output, -o: Write results to a file instead of stdout

Context Results:

Use --return-type context or --return-type enriched to return matching chunks together with their neighbouring chunks. --context-window sets how much context to include, interpreted according to --context-unit: a count of neighbouring chunks (the default), or a tokens/words/characters budget. When a token/word/character budget is used, --context-truncate trims the assembled context to exactly that budget.

Hierarchical Search:

# Search the section-level index and return whole sections
lvdb db search "training loop" --search-level sections --return-type sections --db my_database

# Search the document-level index
lvdb db search "training loop" --search-level documents --db my_database

# Blend chunk and section rankings
lvdb db search "training loop" --search-level fused --db my_database

The --search-level sections/documents/fused options and the sections value for --return-type require a database created with hierarchical_embeddings=True. On a standard (chunk-only) database, only --search-level chunks is available. See Hierarchical Embeddings for details on hierarchical embeddings.

Lexical Search (grep)

grep performs exact/regex, line-oriented search over document content — command-line grep semantics, distinct from search (which does ranked semantic/keyword retrieval). It reports the document id, line number, and matching line, with optional surrounding context. Results are in document-id then line order, not by relevance.

# Literal substring match
lvdb db grep "TODO" --db my_database

# Case-insensitive, with two lines of context each side
lvdb db grep error -i -C 2 --db my_database

# Regex, scoped to a document-id prefix
lvdb db grep "^def " --regex --prefix code/ --db my_database

# Whole-word match, capped at 100 total results
lvdb db grep error --word --limit 100 --db my_database

# JSON output
lvdb db grep "TODO" --format json --db my_database

Options:

--regex, -e: Treat the pattern as a regular expression
--ignore-case, -i: Case-insensitive matching
--word, -w: Match whole words only
--context, -C: Lines of context around each match
--after, -A / --before, -B: Trailing / leading context (override --context)
--prefix, -p: Only search documents whose id starts with this prefix
--max-count, -m: Stop after this many matches per document
--limit, -n: Stop after this many matches in total
--format, -f: Output format (grep or json, default grep); -j is a shortcut for --format json

Example Search Output:

Vector Search Results for `machine learning`: 3 Results
=========================================================

1. Document: doc_123 (Score: 0.8924)
----------------------------------------
Machine learning algorithms have revolutionized the field of artificial intelligence.
Recent advances in deep learning have enabled breakthroughs in computer vision,
natural language processing, and reinforcement learning...

~~~~~

Metadata: {
  "title": "ML Advances",
  "author": "Dr. Smith",
  "journal": "AI Quarterly"
}

-----

2. Document: doc_456 (Score: 0.8456)
----------------------------------------
The application of machine learning techniques to real-world problems requires
careful consideration of data quality, model selection, and evaluation metrics...

Schema Management

The LocalVectorDB CLI provides comprehensive metadata schema management capabilities, allowing you to view, update, and evolve database schemas with optional column remapping to rename existing columns while preserving data.

View Current Schema

# Display current schema (pretty format)
lvdb db schema show --db my_database

# Table format for overview
lvdb db schema show --format table --db my_database

# JSON format for programmatic use
lvdb db schema show --format json --db my_database

# Save schema to file
lvdb db schema show --format json --output current_schema.json --db my_database

Options:

--format, -f: Output format (pretty, json, table)
--output, -o: Output to file instead of stdout

Example Output (Table Format):

+------------------+----------+---------+----------+---------------+
| Field Name       | Type     | Indexed | Required | Default Value |
+------------------+----------+---------+----------+---------------+
| title            | TEXT     | ✓       | ✓        | None          |
| author           | TEXT     | ✓       | ✗        | Unknown       |
| publication_year | INTEGER  | ✓       | ✗        | None          |
| tags             | JSON     | ✗       | ✗        | []            |
+------------------+----------+---------+----------+---------------+

Update Schema

Update database metadata schema from files or JSON strings, with optional column remapping for renaming existing columns:

# Update schema from JSON file
lvdb db schema update --schema new_schema.json --db my_database

# Update with column remapping
lvdb db my_database schema update \
  --schema new_schema.json \
  --mapping '{"old_author": "author", "pub_year": "publication_year"}'

# Update from JSON string
lvdb db my_database schema update \
  --schema '{"title": "text", "author": "text", "year": "integer"}'

# Dry run to preview changes
lvdb db schema update --schema new_schema.json --dry-run --db my_database

# Skip confirmation prompts
lvdb db schema update --schema new_schema.json --force --db my_database

# Verbose output with detailed progress
lvdb db schema update --schema new_schema.json --verbose --db my_database

# Column mapping from file
lvdb db my_database schema update \
  --schema new_schema.json \
  --mapping column_mappings.json

Options:

--schema, -s: JSON string or path to JSON file containing new schema definition
--mapping: Column mapping as JSON string or path to JSON file (old_name: new_name)
--drop-columns, --drop: Actually drop removed columns (WARNING: data loss)
--dry-run, --dry: Show what would be changed without making changes
--force: Skip confirmation prompts
--verbose, -v: Show detailed output

Schema File Format:

{
  "title": {
    "type": "text",
    "indexed": true,
    "required": true
  },
  "author": {
    "type": "text",
    "indexed": true,
    "required": false,
    "default_value": "Unknown"
  },
  "tags": {
    "type": "json",
    "indexed": false,
    "required": false,
    "default_value": []
  },
  "rating": {
    "type": "real",
    "indexed": true,
    "required": false,
    "default_value": 0.0
  }
}

Column Mapping Format:

{
  "old_column_name": "new_column_name",
  "author_name": "author",
  "pub_year": "publication_year",
  "rating_text": "rating"
}

Example Update Session:

$ lvdb db schema update --schema new_schema.json --mapping '{"old_author": "author"}' --db papers

Planned Changes:
  New fields: author, genre, rating
  Removed fields: category
  Column remapping:
    old_author → author

Proceed with schema update? [y/N]: y

Applying schema update...

Schema Update Complete!
  Added fields: author, genre, rating
  Remapped columns:
    old_author → author (150 rows transferred)
  Populated defaults:
    genre: 150 rows updated
    rating: 150 rows updated

Supported Field Types:

text: String values
integer: Whole numbers
real: Floating-point numbers
boolean: True/false values
date: Date strings (ISO format)
json: JSON objects and arrays

Type Conversion During Remapping:

The system supports safe type conversions during column remapping:

TEXT → Any type (SQLite handles conversion)
INTEGER → REAL (safe numeric widening)
BOOLEAN → INTEGER/REAL (True=1, False=0)
JSON → TEXT (already stored as text)

Export Schema

Export current metadata schema to a file for backup or editing:

# Export to JSON file
lvdb db schema export --output current_schema.json --db my_database

# Export with sample data for reference
lvdb db schema export --output schema_with_samples.json --include-data --db my_database

# Export to TOML format (requires: pip install toml)
lvdb db schema export --output schema.toml --format toml --db my_database

Options:

--output, -o: Output file path (required)
--format, -f: Output format (json, toml)
--include-data, --with-data: Include sample data for each field type

Example Output:

$ lvdb db schema export --output backup_schema.json --db papers
Schema exported to backup_schema.json
Fields exported: 4

Common Schema Evolution Patterns

Simple Column Renaming:

# Rename columns while preserving data
lvdb db my_database schema update \
  --schema '{"title": "text", "author": "text", "year": "integer"}' \
  --mapping '{"doc_title": "title", "author_name": "author", "pub_year": "year"}'

Adding New Fields with Defaults:

# Add new fields with default values
lvdb db my_database schema update \
  --schema new_schema.json

# Where new_schema.json contains:
# {
#   "title": "text",
#   "author": "text",
#   "category": {"type": "text", "default_value": "general"},
#   "tags": {"type": "json", "default_value": []},
#   "rating": {"type": "real", "default_value": 0.0}
# }

Type Conversion During Migration:

# Convert text fields to appropriate types
lvdb db my_database schema update \
  --schema converted_schema.json \
  --mapping '{"rating_text": "rating", "is_published": "published_flag"}'

# This converts:
# rating_text (TEXT) → rating (REAL)
# is_published (TEXT) → published_flag (BOOLEAN)

Safe Schema Evolution Workflow:

# 1. Backup current schema
lvdb db schema export --output backup_$(date +%Y%m%d).json --db my_database

# 2. Test changes with dry run
lvdb db schema update --schema new_schema.json --mapping mappings.json --dry-run --db my_database

# 3. Apply changes if satisfied
lvdb db schema update --schema new_schema.json --mapping mappings.json --verbose --db my_database

# 4. Verify results
lvdb db schema show --format table --db my_database
lvdb db search "test query" --limit 3 --db my_database

Best Practices

Planning Schema Changes:

# Always export current schema before major changes
lvdb db schema export --output backup_schema.json --db my_database

# Use dry-run to preview all changes
lvdb db schema update --schema new_schema.json --dry-run --db my_database

# Test on a copy of your database first
cp -r my_database my_database_test
lvdb db schema update --schema new_schema.json --db my_database_test

Column Mapping Guidelines:

Plan mappings carefully and test with sample data
Use descriptive column names that follow consistent conventions
Consider type compatibility when mapping between different field types
Document schema changes for team collaboration

Error Recovery:

# If something goes wrong, restore from backup
lvdb db schema update --schema backup_schema.json --force --db my_database

# Check for data integrity after major changes
lvdb db stats --db my_database
lvdb db search "test query" --db my_database

Production Considerations:

Always backup schema before production changes
Use staged deployments (dev → staging → production)
Monitor application logs after schema updates
Have rollback procedures ready
Test with representative data volumes

Example Production Workflow:

#!/bin/bash
# Production schema update script

DB_NAME="production_database"
SCHEMA_FILE="new_schema_v2.json"
MAPPING_FILE="v1_to_v2_mappings.json"

# 1. Backup current schema
echo "Creating backup..."
lvdb db $DB_NAME schema export --output "backup_$(date +%Y%m%d_%H%M%S).json"

# 2. Validate changes on test database
echo "Testing on copy..."
cp -r $DB_NAME "${DB_NAME}_test"
lvdb db "${DB_NAME}_test" schema update --schema $SCHEMA_FILE --mapping $MAPPING_FILE

if [ $? -ne 0 ]; then
    echo "Schema update failed on test database. Aborting."
    rm -rf "${DB_NAME}_test"
    exit 1
fi

# 3. Test search functionality on copy
echo "Testing search functionality..."
lvdb db "${DB_NAME}_test" search "test query" --limit 5 > /dev/null

if [ $? -ne 0 ]; then
    echo "Search test failed. Aborting."
    rm -rf "${DB_NAME}_test"
    exit 1
fi

# 4. Apply to production
echo "Applying to production..."
lvdb db $DB_NAME schema update --schema $SCHEMA_FILE --mapping $MAPPING_FILE --force

# 5. Verify production update
echo "Verifying production update..."
lvdb db $DB_NAME schema show --format table
lvdb db $DB_NAME stats

# 6. Cleanup
rm -rf "${DB_NAME}_test"
echo "Schema update completed successfully!"

This schema management system provides a safe, powerful way to evolve your database structure while preserving data integrity and maintaining application compatibility.

Interactive Shell

Start an interactive shell for comprehensive database operations, including document management and schema evolution:

lvdb db shell --db my_database

The interactive shell provides a REPL (Read-Eval-Print Loop) environment for performing multiple operations without repeatedly connecting to the database. It includes full schema management capabilities alongside traditional document operations.

Shell Startup:

$ lvdb db shell --db research_papers
Connected to database: research_papers
Documents: 1250, Chunks: 8500
Type 'help' for available commands, 'exit' to quit

research_papers>

Available Commands

Document Operations:

search "<query>" [limit] [type]    - Search for documents
get <id>                          - Get document by ID
add <file or glob>                - Add file(s) to database
delete <id>                       - Delete document by ID
list [limit] [offset]             - List document IDs
count                             - Show document count
stats                             - Show database statistics
info                              - Show database information

Schema Management:

schema show [format]              - Show current schema (pretty|json|table)
schema update <file>              - Update schema from JSON file
schema update-str <json>          - Update schema from JSON string
schema export <file>              - Export current schema to file
schema map <old> <new>            - Add column mapping for next update
schema map-clear                  - Clear column mappings
schema map-show                   - Show current column mappings

General Commands:

clear                             - Clear the console
help                              - Show this help
exit/quit                         - Exit shell

Document Operations Examples

Search Operations:

research_papers> search "neural networks" 3 vector
Found 3 results:
========================================

1. Document: paper_001 (Score: 0.8745)
----------------------------------------
Neural networks are computational models inspired by biological neural networks...

2. Document: paper_127 (Score: 0.8234)
----------------------------------------
Deep learning approaches to natural language processing...

3. Document: paper_089 (Score: 0.7892)
----------------------------------------
Reinforcement learning in autonomous systems...

research_papers> search "machine learning" 5 hybrid
Hybrid search for `machine learning`...
Found 5 results combining vector and keyword search...

Document Management:

research_papers> add new_papers/*.md
Found 3 files. Adding to database...
Successfully added 3 documents
Created IDs: paper_1251, paper_1252, paper_1253

research_papers> get paper_001
Document: paper_001
----------------------------------------
Neural networks are computational models inspired by biological neural networks...

research_papers> delete paper_old_001
Delete document 'paper_old_001'? [y/N]: y
Successfully deleted document: paper_old_001

research_papers> list 5 10
paper_011
paper_012
paper_013
paper_014
paper_015

research_papers> count
Total documents: 1253

Database Information:

research_papers> stats
Database Statistics:
Documents: 1253, Chunks: 8521, Avg chunks/doc: 6.80
Embedding model: embeddinggemma
Provider: ollama
Chunk size: 500, Overlap: 50
FTS enabled: True

research_papers> info
Database Info
-------------
  Database: research_papers
  Embedding model: embeddinggemma
  Total Documents: 1253
  Total Chunks: 8521
  Schema fields: 6

Schema Management Examples

View Current Schema:

research_papers> schema show table
+------------------+----------+---------+----------+---------------+
| Field Name       | Type     | Indexed | Required | Default Value |
+------------------+----------+---------+----------+---------------+
| title            | TEXT     | ✓       | ✓        | None          |
| authors          | JSON     | ✗       | ✗        | []            |
| journal          | TEXT     | ✓       | ✗        | None          |
| publication_year | INTEGER  | ✓       | ✗        | None          |
| keywords         | JSON     | ✗       | ✗        | []            |
| citation_count   | INTEGER  | ✓       | ✗        | 0             |
+------------------+----------+---------+----------+---------------+

research_papers> schema show json
{
  "title": {
    "type": "text",
    "indexed": true,
    "required": true,
    "default_value": null
  },
  "authors": {
    "type": "json",
    "indexed": false,
    "required": false,
    "default_value": []
  }
}

Schema Export and Backup:

research_papers> schema export backup_schema.json
Schema exported to backup_schema.json
Fields exported: 6

research_papers> schema export schema_$(date +%Y%m%d).json
Schema exported to schema_20241201.json
Fields exported: 6

Column Mapping and Schema Updates:

research_papers> schema map paper_title title
Added column mapping: paper_title → title

research_papers> schema map author_list authors
Added column mapping: author_list → authors

research_papers> schema map pub_year publication_year
Added column mapping: pub_year → publication_year

research_papers> schema map-show
Current column mappings:
  paper_title → title
  author_list → authors
  pub_year → publication_year

research_papers> schema update new_research_schema.json

Planned Changes:
  New fields: keywords, citation_count, impact_factor
  Removed fields: old_category
  Column remapping:
    paper_title → title
    author_list → authors
    pub_year → publication_year

Proceed with schema update? [y/N]: y

Applying schema update...

Schema Update Complete!
  Added fields: keywords, citation_count, impact_factor
  Remapped columns:
    paper_title → title (1253 rows transferred)
    author_list → authors (1253 rows transferred)
    pub_year → publication_year (1253 rows transferred)
  Populated defaults:
    keywords: 1253 rows updated
    citation_count: 1253 rows updated

Quick Schema Updates:

research_papers> schema update-str '{"title": "text", "author": "text", "tags": {"type": "json", "default_value": []}}'
Apply schema update? [y/N]: y
Schema updated successfully!

research_papers> schema map-clear
Column mappings cleared

Complete Workflow Examples

Research Paper Database Migration:

papers> # Starting with legacy schema
papers> schema show table
+---------------+----------+---------+----------+---------------+
| Field Name    | Type     | Indexed | Required | Default Value |
+---------------+----------+---------+----------+---------------+
| paper_title   | TEXT     | ✓       | ✓        | None          |
| author_names  | TEXT     | ✗       | ✗        | None          |
| journal_name  | TEXT     | ✓       | ✗        | None          |
| year_pub      | INTEGER  | ✓       | ✗        | None          |
+---------------+----------+---------+----------+---------------+

papers> # Export backup before changes
papers> schema export legacy_backup.json
Schema exported to legacy_backup.json
Fields exported: 4

papers> # Set up column mappings for modernization
papers> schema map paper_title title
Added column mapping: paper_title → title

papers> schema map author_names authors
Added column mapping: author_names → authors

papers> schema map journal_name journal
Added column mapping: journal_name → journal

papers> schema map year_pub publication_year
Added column mapping: year_pub → publication_year

papers> # Review planned mappings
papers> schema map-show
Current column mappings:
  paper_title → title
  author_names → authors
  journal_name → journal
  year_pub → publication_year

papers> # Apply modern research schema
papers> schema update modern_research_schema.json

Planned Changes:
  New fields: title, authors, journal, publication_year, keywords, doi, abstract
  Removed fields: (none - all mapped)
  Column remapping:
    paper_title → title
    author_names → authors
    journal_name → journal
    year_pub → publication_year

Proceed with schema update? [y/N]: y

Applying schema update...

Schema Update Complete!
  Added fields: title, authors, journal, publication_year, keywords, doi, abstract
  Remapped columns:
    paper_title → title (892 rows transferred)
    author_names → authors (892 rows transferred)
    journal_name → journal (856 rows transferred)
    year_pub → publication_year (892 rows transferred)
  Populated defaults:
    keywords: 892 rows updated
    doi: 892 rows updated
    abstract: 892 rows updated

papers> # Verify the migration worked
papers> schema show table
+------------------+----------+---------+----------+---------------+
| Field Name       | Type     | Indexed | Required | Default Value |
+------------------+----------+---------+----------+---------------+
| title            | TEXT     | ✓       | ✓        | None          |
| authors          | JSON     | ✗       | ✗        | []            |
| journal          | TEXT     | ✓       | ✗        | None          |
| publication_year | INTEGER  | ✓       | ✗        | None          |
| keywords         | JSON     | ✗       | ✗        | []            |
| doi              | TEXT     | ✓       | ✗        | None          |
| abstract         | TEXT     | ✓       | ✗        | None          |
+------------------+----------+---------+----------+---------------+

papers> # Test search with new schema
papers> search "machine learning" 3
Found 3 results:
========================================

1. Document: paper_045 (Score: 0.9123)
----------------------------------------
Machine learning techniques for automated research discovery...

papers> # Test document retrieval
papers> get paper_045
Document: paper_045
----------------------------------------
Machine learning techniques for automated research discovery...

Metadata:
{
  "title": "ML for Research Discovery",
  "authors": ["Chen, L.", "Rodriguez, M."],
  "journal": "Nature Machine Intelligence",
  "publication_year": 2023,
  "keywords": ["machine learning", "research", "automation"],
  "doi": "10.1038/s42256-023-00123-4",
  "abstract": null
}

papers> # Schema migration successful!
papers> count
Total documents: 892

Development and Testing Workflow:

dev_db> # Quick schema prototyping
dev_db> schema update-str '{"name": "text", "category": "text", "priority": {"type": "integer", "default_value": 1}}'
Apply schema update? [y/N]: y
Schema updated successfully!

dev_db> # Add test data
dev_db> add test_documents/*.txt
Successfully added 5 documents

dev_db> # Test search functionality
dev_db> search "test query" 3
Found 3 results...

dev_db> # Export schema for production use
dev_db> schema export production_ready_schema.json
Schema exported to production_ready_schema.json
Fields exported: 3

dev_db> # Clear screen and continue working
dev_db> clear

dev_db> # Final verification
dev_db> stats
Database Statistics:
Documents: 5, Chunks: 23, Avg chunks/doc: 4.60

Shell vs CLI Command Comparison

Interactive Shell Advantages:

Persistent Connection: No reconnection overhead between commands
Command History: Easy to repeat and modify previous commands
Context Awareness: Schema mappings persist across commands
Real-time Feedback: Immediate results and error handling
Workflow Continuity: Schema updates → test search → verify data

When to Use Shell vs CLI:

Use Interactive Shell for:

# Data exploration and analysis
lvdb db shell --db research
# > search "topic A" 10
# > search "topic B" 10
# > search "topic C" 10

# Schema development and testing
# > schema show
# > schema update-str '{"test": "text"}'
# > add test.txt
# > search "test"

# Complex migrations
# > schema export backup.json
# > schema map old new
# > schema update new.json
# > search "verify"

Use CLI Commands for:

# Scripting and automation
lvdb db search "query" --format json > results.json --db prod

# One-time operations
lvdb db add important_doc.txt --db prod

# Production deployments
lvdb db schema update --schema prod_schema.json --force --db prod

This enhanced interactive shell provides a powerful environment for both day-to-day database operations and complex schema evolution tasks, making it easy to iterate, test, and deploy changes safely.

Standalone Chunking

lvdb chunk runs LocalVectorDB’s position-aware chunkers on their own — no database, embedding provider, or configuration required. It reads text from files, globs, direct arguments, or - (stdin), chunks each input, and writes one JSON object per chunk (JSONL) to stdout or --output. This is handy for inspecting how a document will be split before ingesting it, or for feeding chunks to another tool.

# Chunk a Markdown file to stdout (JSONL)
lvdb chunk notes.md

# Chunk a PDF (extracted to Markdown first), 300 tokens per chunk
lvdb chunk report.pdf --method sentences --max-tokens 300

# Chunk every Markdown file matching a glob into a file
lvdb chunk "docs/*.md" --method paragraphs -o chunks.jsonl

# Chunk text from stdin with word chunks and a 20-token overlap
echo "some long text..." | lvdb chunk - --method words --overlap 20

Rich file formats (PDF, DOCX, HTML, …) are extracted to Markdown first, exactly as lvdb db <name> add does. Use --extract/--no-extract to force or disable extraction (the default is auto-detection per file).

Options:

--method, -M: Chunking strategy. One of sentences, tokens, words, lines, characters, paragraphs, sections, structure, code-blocks, delimiter (default: sentences)
--max-tokens, --chunk-size, -s: Maximum tokens per chunk (default: 500)
--overlap: Token overlap between consecutive chunks; ignored by some strategies (default: 0)
--delimiter: Delimiter for --method delimiter (default: a blank line). The escapes \n, \t, \r are interpreted, e.g. --delimiter '\n---\n'
--output, -o: Write JSONL to this file instead of stdout
--extract/--no-extract: Force or disable text extraction for file inputs (default: auto)

Output Format:

Each output line is a JSON object with content, index (position within its source), tokens, and position (a dict with start, end, line, column, end_line, end_column). When more than one input is chunked, each record also includes a source key naming its file. A summary of how many chunks were written is printed to stderr, so it does not pollute the JSONL on stdout.

$ echo "First sentence. Second sentence." | lvdb chunk - --method sentences
{"content": "First sentence.", "index": 0, "tokens": 2, "position": {"start": 0, "end": 15, "line": 1, "column": 0, "end_line": 1, "end_column": 15}}
{"content": "Second sentence.", "index": 1, "tokens": 2, "position": {"start": 16, "end": 32, "line": 1, "column": 16, "end_line": 1, "end_column": 32}}
Wrote 2 chunk(s) from 1 input(s)

Version

# Print the installed localvectordb version
lvdb version

# The top-level --version / -V flag prints the same value
lvdb --version

Advanced Usage Examples

Bulk Operations

# Add all markdown files in a directory tree
lvdb db add "documents/**/*.md" --db papers

# Add with structured metadata from JSON
lvdb db add papers.txt --metadata metadata.json --id ids.txt --db papers

# Search and save results for further processing
lvdb db search "deep learning" --format json --output dl_papers.json --db papers

# Batch process search results
for query in "neural networks" "machine learning" "computer vision"; do
  lvdb db search "$query" --limit 10 --output "results_${query// /_}.txt" --db papers
done

Configuration and Deployment

# Create production configuration
lvdb config init --format toml --output production.toml

# Start production server (--config is global; it precedes the subcommand)
lvdb --config production.toml serve --host 0.0.0.0 --port 8080

# Create backup-friendly database (--db-folder is global; it precedes the subcommand)
lvdb --db-folder /backup/vector_dbs create backup_db

# Monitor database growth
watch 'lvdb list --details'

Pipeline Integration

#!/bin/bash
# Document processing pipeline

# Create database if not exists.
# NOTE: `lvdb create --metadata-schema` accepts: files, documents,
# research_papers, code_repository, customer_support. The file-oriented
# `files` schema (file_path, created_at, last_modified, file_size_bytes, ...)
# is used here because metadata not present in the schema is ignored on
# insert. (`lvdb config init --schema files` writes the same schema into a
# config file instead.)
lvdb create research_pipeline --embedding-model embeddinggemma --metadata-schema files

# Process incoming documents
find /incoming/documents -name "*.txt" -newer /tmp/last_processed | while IFS= read -r file; do
   echo "Processing: $file"

   # Extract file metadata using stat
   last_modified=$(stat -c %Y "$file")
   file_size_bytes=$(stat -c %s "$file")
   created_at=$(stat -c %w "$file")

   # If creation time is unavailable (returns '-'), use the last modification time
   if [ "$created_at" = "-" ]; then
     created_at="$last_modified"
   fi

   # Construct JSON metadata string
   metadata=$(printf '{"file_path": "%s", "last_modified": "%s", "file_size_bytes": "%s", "created_at": "%s"}' \
     "$file" "$last_modified" "$file_size_bytes" "$created_at")

   # Add record using lvdb add command with metadata
   lvdb db add "$file" --metadata "$metadata" --db research_pipeline
done

# Update timestamp
touch /tmp/last_processed

# Search and generate report
lvdb db search "quarterly report" --format json > quarterly_matches.json --db research_pipeline

Troubleshooting

Common Issues

Database not found:

# Check available databases
lvdb list

# Verify database folder
ls -la ~/.lvdb/

# Create database if missing
lvdb create missing_database

Ollama connection errors:

# Check Ollama status
ollama list

# Start Ollama if needed
ollama serve

# Test with different model
lvdb create test_db --embedding-model all-minilm

Permission errors:

# Check database folder permissions
ls -la ~/.lvdb/

# Fix permissions
chmod 755 ~/.lvdb/
chmod 644 ~/.lvdb/*.sqlite

Configuration issues:

# Validate configuration
lvdb config show

# Reset to defaults
mv server-cfg.toml server-cfg.toml.backup
lvdb config init

Debug Mode

# Run server in debug mode
lvdb serve --debug --log-level DEBUG

# Test with verbose output (--verbose is a global option, before the subcommand)
lvdb --verbose db test_database add test.txt

# Check server logs
tail -f server.log

Backup and Restore Operations

The LocalVectorDB CLI provides comprehensive backup and restore capabilities for database protection and disaster recovery.

Create Backups

# Create full backup
lvdb backup create my_database --type full

# Create incremental backup (requires the parent backup ID)
lvdb backup create my_database --type incremental --parent backup-abc123

# Create backup with compression
lvdb backup create my_database --type full --compression gzip

# Create backup with custom location
lvdb backup create my_database --type full --location /backups/localvectordb

# Create backup, skipping the FAISS index and integrity check
lvdb backup create my_database --type full --exclude-faiss --no-verify

Options:

--type, -t: Backup type (full, incremental; default full)
--parent, -p: Parent backup ID (required for incremental backups)
--location, -l: Backup storage location (default ./backups)
--compression, -c: Compression algorithm (none, gzip, lzma, bzip2; default gzip)
--no-verify: Skip backup integrity verification after creation
--exclude-faiss: Exclude the FAISS index from the backup
--format, -f: Output format (table (default) or json); -j is a shortcut for --format json

List Backups

# List all backups
lvdb backup list

# List backups for specific database
lvdb backup list --database my_database

# Filter by backup type and limit the number shown
lvdb backup list --type full --limit 10

# Scan a specific backup location, in JSON format
lvdb backup list --location /backups/localvectordb --format json

Options:

--database, -d: Filter backups for a specific database
--type, -t: Filter by backup type (full, incremental)
--limit, -n: Limit the number of backups shown
--location, -l: Backup storage location to scan (default ./backups)
--format, -f: Output format (table (default) or json); -j is a shortcut for --format json

Example Output:

Backups for database: my_database
=====================================

ID                    Type        Created              Size      Status    Description
-----------------------------------------------------------------------------------------
backup_20241201_001   FULL        2024-12-01 10:30:00  125.4 MB  VALID     Pre-migration backup
backup_20241201_002   INCREMENTAL 2024-12-01 15:45:00  8.2 MB    VALID     After document updates
backup_20241202_001   FULL        2024-12-02 09:00:00  127.1 MB  VALID     Daily backup

Restore Backups

# Restore a backup to its original location (partial backup IDs are accepted)
lvdb backup restore abc12345

# Restore into a specific directory
lvdb backup restore backup_20241201_001 --to-location ./restored_database

# Overwrite existing files without confirmation
lvdb backup restore backup_20241201_001 --to-location ./restored --overwrite

# Restore from a specific backup store, JSON output
lvdb backup restore abc12345 --location /backups/localvectordb --format json

Options:

--to-location, -t: Directory to restore to (default: the backup’s original location)
--overwrite: Overwrite existing files without confirmation
--location, -l: Backup storage location to read from (default ./backups)
--format, -f: Output format (table (default) or json); -j is a shortcut for --format json

The BACKUP_ID argument accepts a partial ID; the first matching backup file is used. For incremental backups the full backup chain is located and applied automatically.

Verify Backups

# Verify a specific backup (partial backup IDs are accepted)
lvdb backup verify abc12345

# Verify a backup in a specific store, JSON output
lvdb backup verify backup_20241201_001 --location /backups/localvectordb --format json

Options:

--location, -l: Backup storage location to read from (default ./backups)
--format, -f: Output format (table (default) or json); -j is a shortcut for --format json

Example Output:

Verifying backup: backup_20241201_001
=====================================

✓ Backup metadata valid
✓ File integrity checksums match
✓ Database schema valid
✓ FAISS index accessible
✓ SQLite database accessible

Backup verification: PASSED

Clean Up Backups

# Clean up backups older than 30 days (the default), keeping at least 3 full backups
lvdb backup cleanup

# Delete backups older than 7 days
lvdb backup cleanup --older-than 7

# Keep at least 5 full backups, previewing what would be deleted
lvdb backup cleanup --keep-full 5 --dry-run

# Clean up a specific backup store, JSON output
lvdb backup cleanup --location /backups/localvectordb --format json

Options:

--older-than: Delete backups older than N days (default 30)
--keep-full: Minimum number of full backups to keep (default 3)
--location, -l: Backup storage location (default ./backups)
--dry-run: Show what would be deleted without actually deleting
--format, -f: Output format (table (default) or json); -j is a shortcut for --format json

cleanup operates on all backups in the store (it takes no database argument) while maintaining backup-chain integrity and keeping a minimum number of full backups.

Point-in-Time Recovery

# Restore database to a specific point in time (--to-location is required)
lvdb backup pitr "2024-12-01 14:30:00" --to-location ./pitr_restored

# ISO-8601 timestamp with a wider search tolerance, validating without restoring
lvdb backup pitr "2024-12-01T14:30:00Z" --to-location ./restored --tolerance 120 --dry-run

# Read from a specific backup store, JSON output
lvdb backup pitr "2024-12-01 14:30:00" --to-location ./restored --location /backups/localvectordb --format json

Options:

--to-location, -t: Directory to restore to (required)
--tolerance: Tolerance in minutes for finding a recovery point (default 60)
--location, -l: Backup storage location (default ./backups)
--dry-run: Validate recovery without actually restoring
--format, -f: Output format (table (default) or json); -j is a shortcut for --format json

The TIMESTAMP argument accepts YYYY-MM-DD HH:MM:SS or ISO-8601 format. The database name is read from the backup store, so it is not passed on the command line.

Example PITR Session:

$ lvdb backup pitr "2024-12-01 14:30:00" --to-location ./recovered

Point-in-Time Recovery
======================
Target time: 2024-12-01 14:30:00 UTC
Database: research_papers
Restore location: ./recovered

Recovery plan:
  Base backup: backup_20241201_001 (2024-12-01 10:00:00)
  Incremental backups: 2
    - backup_20241201_inc_001 (2024-12-01 12:00:00)
    - backup_20241201_inc_002 (2024-12-01 14:15:00)

Proceed with recovery? [y/N]: y

Restoring base backup...
Applying incremental backup 1...
Applying incremental backup 2...
Stopping at target time: 2024-12-01 14:30:00

✓ Point-in-time recovery completed successfully!

Database Migration and Schema Evolution

The LocalVectorDB CLI provides powerful migration capabilities for evolving database schemas and managing version upgrades.

Migration Status

# Check migration status for database
lvdb migrate status my_database

# Use a custom migrations directory
lvdb migrate status my_database --migrations-dir ./custom_migrations

# JSON output for automation
lvdb migrate status my_database --format json

Options:

--migrations-dir, -m: Directory containing migration files (default ./migrations)
--format, -f: Output format (table (default) or json); -j is a shortcut for --format json

Example Output:

Migration Status: my_database
=============================

Current Version: 1.1.0
Target Version: 1.2.0
Pending Migrations: 2

Available Migrations:
---------------------
Version   Status    Description                    Created
---------------------------------------------------------------
1.1.1     PENDING   Add citation_count field       2024-11-15
1.2.0     PENDING   Add full-text search indices   2024-11-20

Migration Path:
  1.1.0 → 1.1.1 → 1.2.0

Apply Migrations

# Apply all pending migrations (a backup is created first by default)
lvdb migrate apply my_database

# Apply migrations up to a specific version
lvdb migrate apply my_database --to-version 1.2.0

# Apply without creating a pre-migration backup
lvdb migrate apply my_database --no-backup

# Dry run to preview changes without a backup
lvdb migrate apply my_database --dry-run --no-backup

# Store the pre-migration backup in a custom location
lvdb migrate apply my_database --backup-location /backups/premigration

Options:

--to-version: Target version to migrate to (default: latest)
--migrations-dir, -m: Directory containing migration files (default ./migrations)
--backup/--no-backup: Create a backup before migrating (default: enabled)
--backup-location, -b: Backup storage location (default ./backups)
--dry-run: Validate migrations without applying them
--format, -f: Output format (table (default) or json); -j is a shortcut for --format json

Example Migration Session:

$ lvdb migrate apply research_papers --backup

Migration Plan: research_papers
===============================

Current Version: 1.1.0
Target Version: 1.2.0

Migrations to Apply:
  1. v1.1.1: Add citation_count field
  2. v1.2.0: Add full-text search indices

Creating backup before migration...
✓ Backup created: backup_20241201_premigration

Apply these migrations? [y/N]: y

Applying migration v1.1.1...
✓ Added citation_count field (default: 0)
✓ Updated 1,247 existing records

Applying migration v1.2.0...
✓ Created full-text search indices
✓ Rebuilt search index

✓ Migration completed successfully!
Database version: 1.1.0 → 1.2.0

Rollback Migrations

# Rollback to a specific version (both database name and target version are required)
lvdb migrate rollback my_database 1.1.0

# Rollback without creating a pre-rollback backup
lvdb migrate rollback my_database 1.0.0 --no-backup

# Validate a rollback without applying it
lvdb migrate rollback my_database 1.1.0 --dry-run

Options:

--migrations-dir, -m: Directory containing migration files (default ./migrations)
--backup/--no-backup: Create a backup before rolling back (default: enabled)
--backup-location, -b: Backup storage location (default ./backups)
--dry-run: Validate the rollback without applying it
--format, -f: Output format (table (default) or json); -j is a shortcut for --format json

Both the DATABASE_NAME and TARGET_VERSION positional arguments are required.

Create Migration Templates

# Create a new migration template (the description is the positional argument)
lvdb migrate create "add priority field" --version 1.3.0

# Create a schema-change template
lvdb migrate create "add user fields" --version 1.3.1 --template schema

# Create a data-transformation template in a specific directory
lvdb migrate create "migrate old data format" --version 1.4.0 --template data --migrations-dir ./migrations

Options:

--version: Version number for the migration, e.g. 1.2.0 (required)
--migrations-dir, -m: Directory to create the migration in (default ./migrations)
--template, -t: Template type — basic (default), schema, or data
--format, -f: Output format (table (default) or json); -j is a shortcut for --format json

Example Generated Migration:

Migrations subclass Migration and describe schema changes by returning a new_schema mapping from get_schema_changes() / get_rollback_changes() (there are no up/down methods). The schema template produces:

# Migration: add user fields
# Version: 1.3.1
# Created: 2024-12-01 10:30:00

from typing import Dict, Any
from localvectordb.migration import Migration
from localvectordb.core import MetadataField, MetadataFieldType


class Migration_1_3_1(Migration):
    """
    add user fields
    """

    version = "1.3.1"
    description = "add user fields"
    dependencies = []  # Add version dependencies here

    def get_schema_changes(self) -> Dict[str, Any]:
        """Get schema changes to apply in forward migration."""

        # Define the new complete metadata schema after this migration
        new_schema = {
            'user_id': MetadataField(
                type=MetadataFieldType.TEXT,
                indexed=True,
                required=False,
                default_value=None,
            ),
            'priority': MetadataField(
                type=MetadataFieldType.INTEGER,
                indexed=True,
                required=False,
                default_value=0,
            ),
            # Add existing fields that should remain...
        }

        return {
            'new_schema': new_schema,
            'column_mapping': {},  # Optional: rename columns {'old_name': 'new_name'}
            'drop_columns': False,  # Whether to drop unused columns
        }

    def get_rollback_changes(self) -> Dict[str, Any]:
        """Get schema changes to apply for rollback."""

        rollback_schema = {
            # Define schema without the changes from this migration
        }

        return {
            'new_schema': rollback_schema,
            'column_mapping': {},
            'drop_columns': False,
        }

List Migrations

# List all discovered migration files
lvdb migrate list

# Show each migration's dependencies, JSON output
lvdb migrate list --show-dependencies --format json

# List migrations from a custom directory
lvdb migrate list --migrations-dir ./custom_migrations

Options:

--migrations-dir, -m: Directory containing migration files (default ./migrations)
--show-dependencies, -d: Show migration dependencies
--format, -f: Output format (table (default) or json); -j is a shortcut for --format json

list scans the migrations directory itself, so it takes no database argument.

Example Output:

Available Migrations in migrations:

Version   Description                     File                        Dependencies
-------------------------------------------------------------------------------------
1.1.1     Add citation_count field        v1_1_1_add_citation.py      -
1.2.0     Add full-text search indices    v1_2_0_add_fts.py           1.1.1
1.3.0     Add priority field              v1_3_0_add_priority.py      1.2.0

Total: 3 migration(s)

SQLite Performance Tuning

The lvdb tuning command group manages SQLite performance profiles and pragma settings for individual databases. Unlike most commands, tuning does not require a configuration file; point it at your databases with the global --db-folder, -d option (given before the subcommand), e.g. lvdb --db-folder ./data tuning get mydb.

List Tuning Profiles

# List available SQLite tuning profiles with descriptions
lvdb tuning list

list takes no arguments or options.

Show Tuning Configuration

# Show the current tuning profile, overrides, and active pragmas for a database
lvdb --db-folder ./data tuning get my_database

# JSON output
lvdb --db-folder ./data tuning get my_database --format json

Options:

--format, -f: Output format (table (default), json)

Apply a Tuning Profile

# Apply a named profile to a database
lvdb --db-folder ./data tuning set my_database read_optimized

# Apply a profile with pragma overrides (repeat --override per pragma)
lvdb --db-folder ./data tuning set my_database balanced \
  --override cache_size=-64000 --override mmap_size=268435456

# Apply without persisting the settings to the database
lvdb --db-folder ./data tuning set my_database read_optimized --no-persist

# Preview what would be applied without changing anything
lvdb --db-folder ./data tuning set my_database read_optimized --dry-run

Options:

--override, -o KEY=VALUE: Pragma override (may be given multiple times)
--no-persist: Do not persist the settings to the database
--dry-run: Show what would be applied without applying it

Set a Single Pragma

# Set a single pragma override, keeping the current profile
lvdb --db-folder ./data tuning set-pragma my_database cache_size -64000

Options:

--no-persist: Do not persist the setting to the database

set-pragma takes three positional arguments: DATABASE, PRAGMA_KEY and PRAGMA_VALUE.

Auto-Tuning Recommendations

# Get a recommendation from an interactive workload interview
lvdb --db-folder ./data tuning auto my_database --interactive

# Get a recommendation from explicit workload parameters
lvdb --db-folder ./data tuning auto my_database \
  --workload-type read_heavy --memory-constraint generous --durability normal

# Apply the recommended settings immediately, JSON output
lvdb --db-folder ./data tuning auto my_database --workload-type balanced --apply --format json

Options:

--interactive, -i: Run an interactive workload interview
--workload-type: Workload type (read_heavy, write_heavy, balanced, batch_ingest, real_time) — skips the interview
--memory-constraint: Memory availability (generous, moderate, limited)
--durability: Data durability importance (critical, high, normal, low)
--apply: Apply the recommended settings immediately
--format, -f: Output format (table (default), json)

Database Maintenance

The lvdb maintenance command group runs SQLite maintenance operations on a database. Like tuning, it does not require a configuration file; select the database directory with the global --db-folder, -d option before the subcommand.

WAL Checkpoint

# Run a WAL checkpoint (default mode PASSIVE)
lvdb --db-folder ./data maintenance checkpoint my_database

# Run a TRUNCATE checkpoint
lvdb --db-folder ./data maintenance checkpoint my_database --mode TRUNCATE

Options:

--mode, -m: Checkpoint mode (PASSIVE (default), FULL, RESTART, TRUNCATE)

Optimize

# Run PRAGMA optimize on a database
lvdb --db-folder ./data maintenance optimize my_database

optimize takes only the DATABASE argument.

Vacuum

# Full VACUUM (prompts for confirmation; requires exclusive access)
lvdb --db-folder ./data maintenance vacuum my_database

# Skip the confirmation prompt
lvdb --db-folder ./data maintenance vacuum my_database --confirm

# Incremental vacuum, reclaiming a given number of pages
lvdb --db-folder ./data maintenance vacuum my_database --incremental --pages 5000

Options:

--incremental, -i: Run an incremental vacuum instead of a full VACUUM
--pages: Pages to reclaim (incremental only; default 2000)
--confirm: Skip the confirmation prompt (full VACUUM only)

Analyze System Resources

# Analyze system RAM, CPU, and disk for tuning recommendations
lvdb maintenance analyze-system

# JSON output
lvdb maintenance analyze-system --format json

Options:

--format, -f: Output format (table (default), json)

analyze-system inspects the host machine and takes no database argument.

MCP Server Integration

The LocalVectorDB CLI provides Model Context Protocol (MCP) server capabilities for integration with AI assistants and tools.

Start MCP Server

# Start MCP server in read-only mode (default)
lvdb mcp serve

# Start in read-write mode
lvdb mcp serve --mode read-write

# Start with specific configuration
lvdb mcp serve --config /path/to/mcp-config.toml

# Start with custom database root
lvdb mcp serve --databases-root /data/vector_databases

# Start with database mapping
lvdb mcp serve --databases-map '{"papers": "/data/research", "docs": "http://remote-server:8000/docs"}'

# Start with debug logging
lvdb mcp serve --log-level DEBUG

Options:

--mode: Server mode (read-only, read-write)
--config: Path to MCP configuration file (TOML format)
--databases-root: Root directory for local databases
--databases-map: JSON mapping of database names to paths/URLs
--log-level: Logging level (DEBUG, INFO, WARNING, ERROR)

MCP Configuration File Example:

Server-level settings (including the read-only/read-write mode) live under [mcp]. Databases are configured under [databases] with a root directory and an optional [databases.map] table mapping database names to specific local paths or remote URLs. Run lvdb mcp config-example to print a fully-commented template. See MCP Server for the complete configuration reference.

# mcp-config.toml
[mcp]
mode = "read-write"   # or "read-only" (read-only is a whole-server setting)
log_level = "INFO"

[databases]
# Root directory searched for local databases
root = "/data/databases"

# Optional: map specific names to local paths or remote URLs
[databases.map]
research = "/data/research_papers"
documentation = "/data/docs"
shared_knowledge = "http://knowledge-server:8000/shared"

Claude Desktop Integration:

To use with Claude Desktop, add this to your Claude Desktop configuration:

{
  "mcpServers": {
    "localvectordb": {
      "command": "lvdb",
      "args": ["mcp", "serve", "--mode", "read-write"],
      "env": {
        "LVDB_MCP_CONFIG": "/path/to/your/mcp-config.toml"
      }
    }
  }
}

Available MCP Tools:

When running as an MCP server, LocalVectorDB provides these tools to AI assistants. Read-only tools are available in every mode; write tools require --mode read-write. See MCP Server for the full parameter reference.

Read-only tools:

list_databases: List available databases
get_database_info: Get database statistics and configuration
query_database: Search with vector, keyword, or hybrid search
find_related_documents: Find documents related to a given document (nearest neighbours)
filter_documents: Filter documents by metadata (with limit/offset)
get_document: Retrieve a document by ID, or a portion of it (chunk/range/lines/section/outline)
check_documents_exist: Check whether documents exist
get_metadata_schema: Get a database’s metadata schema
get_system_info: Get version, configuration, and enabled tools

Write tools (read-write mode only):

create_database: Create a new database
delete_database: Delete a database and its data
upsert_documents: Insert or update documents
update_document: Update a document’s content and/or metadata
patch_document: Edit a document in place with find/replace (old_string/new_string)
delete_document: Delete a document by ID
update_metadata_schema: Add or modify metadata schema fields

Command-Line Interface

Exit Codes

Installation

Server Management

Starting the Server

Configuration Management

Initialize Configuration

View Configuration

Get Configuration Values

Set Configuration Values

Authentication and API Key Management

Permission Levels

Check Authentication Status

Create API Keys

List API Keys

Get Key Information

Revoke API Keys

Rotate API Keys

Clean Up Expired Keys

API Key Usage in Applications

Security Best Practices

Database Operations

List Databases

Create Database

Delete Database

Rename Database

Database-Specific Operations

Database Information

Corpus Diagnosis (doctor)

Document Management

Add Documents

Get Documents

Update Documents

Patch Documents (in-place edit)

Delete Documents

List Documents

Browse Document IDs

Search Operations

Lexical Search (grep)

Related Documents

Schema Management

View Current Schema

Update Schema

Export Schema

Common Schema Evolution Patterns

Best Practices

Interactive Shell

Available Commands

Document Operations Examples

Schema Management Examples

Complete Workflow Examples

Shell vs CLI Command Comparison

Standalone Chunking

Version

Advanced Usage Examples

Bulk Operations

Configuration and Deployment

Pipeline Integration

Troubleshooting

Common Issues

Debug Mode

Backup and Restore Operations

Create Backups

List Backups

Restore Backups

Verify Backups

Clean Up Backups

Point-in-Time Recovery

Database Migration and Schema Evolution

Migration Status

Apply Migrations

Rollback Migrations

Create Migration Templates

List Migrations

SQLite Performance Tuning

List Tuning Profiles

Show Tuning Configuration

Apply a Tuning Profile

Set a Single Pragma

Auto-Tuning Recommendations