sqler-cli 0.1.2

LLM memory management CLI - persistent storage with FTS5 search

sqler-cli (mem)

Persistent memory for LLMs. Store, search, and retrieve information across sessions using SQLite FTS5 full-text search.

Why This Exists

LLMs have no persistent memory between sessions. This CLI provides a simple way to:

• Remember things: project conventions, user preferences, API endpoints, decisions made
• Recall them later: full-text search finds relevant memories instantly
• Organize with tags, sessions, and importance levels
• Link related memories together with supersedes/see-also chains
• Deduplicate similar memories to keep your knowledge base clean
• Export/Import: backup and transfer memory databases

Installation

# From PyPI
pip install sqler-cli

# From source
cd sqler-cli
uv sync
uv run mem --help

# Or install globally
uv pip install -e .

Quick Start

# Store memories
mem remember "The API key is stored in .env file"
mem remember "User prefers dark mode and vim keybindings" --tag preferences
mem remember "POST /api/v1/users creates a new user" --tag api --tag docs

# Search memories (full-text search)
mem recall "API key"              # Finds the .env memory
mem recall "user" --tag api       # Finds only API-related user memories

# List all memories
mem list
mem list --tag preferences        # Filter by tag

# Get JSON output (for scripts/LLMs)
mem recall "preferences" --json

# Check what's stored
mem stats

Commands Reference

remember - Store a memory

mem remember "content to remember"
mem remember "content" --tag TAG [--tag TAG2 ...]
mem remember "content" --context "why this was stored"
mem remember "content" --source claude  # default: user
mem remember --file notes.txt           # read from file
echo "content" | mem remember           # read from stdin

# Session isolation - group related memories
mem remember "Auth refactor notes" --session auth-refactor

# Auto-tagging - detect tags from content (api, database, auth, config, error, security)
mem remember "JWT authentication setup" --auto-tag
# Output: Remembered (id=1) [auto-tagged: auth, security]

# Memory linking
mem remember "New API URL is /v2" --supersedes 42        # Replaces old memory
mem remember "Login flow" --see-also 15 --see-also 23   # Links related memories

# Source attribution
mem remember "Rate limit is 100/min" --source-url "https://api.example.com/docs"
mem remember "Config format" --source-file "/etc/app/config.yaml:42"

# Importance (1-5, default 3)
mem remember "CRITICAL: Never commit secrets" --importance 5

# Output formats
mem remember "content" --json    # {"id": 1, "content": "...", "tags": [], "auto_tags": [...]}
mem remember "content" --quiet   # just prints: 1

recall - Search memories (FTS)

Full-text search using SQLite FTS5 with BM25 ranking.

mem recall "search query"
mem recall "query" --tag TAG           # filter by tag
mem recall "query" --limit 5           # max results (default: 10)
mem recall "query" --json              # JSON output for parsing

# Show relevance scores (useful for debugging search)
mem recall "query" --show-score        # adds score field to output

# Sort by date instead of relevance
mem recall "query" --recent-first      # newest first

# Session filtering
mem recall "query" --session auth-refactor  # only search this session

# Importance filtering
mem recall "query" --min-importance 4       # only important memories
mem recall "query" --boost-important        # prioritize high-importance in results

Search syntax:

• mem recall "API endpoint" - finds memories containing both words
• mem recall "sqlite OR postgres" - boolean OR
• mem recall "config*" - prefix matching

update - Modify a memory in-place

Update content, tags, or metadata without losing the memory ID or creation timestamp.

mem update 42 "New content"              # Replace content
mem update 42 --tag newtag               # Add a tag
mem update 42 --context "New context"    # Update context
mem update 42 --clear-tags               # Remove all tags
mem update 42 --importance 5             # Mark as critical
mem update 42 --see-also 15 --see-also 23  # Add related memories
mem update 42 --session new-session      # Change session

list - Browse memories

mem list                       # all memories
mem list --tag preferences     # filter by tag
mem list --limit 20            # limit results
mem list --since 2024-01-01    # filter by date
mem list --session auth-work   # filter by session
mem list --min-importance 4    # only important memories
mem list --json                # JSON output

forget - Delete memories

mem forget 42                         # delete by ID
mem forget --tag temporary --confirm  # bulk delete by tag

tags - Manage tags

mem tags list              # show all tags with counts
mem tags add 42 important  # add tag to memory
mem tags rm 42 draft       # remove tag from memory

dedupe - Find and merge duplicates

Find near-duplicate memories and optionally merge them.

mem dedupe              # Interactive: show groups, ask which to merge
mem dedupe --dry-run    # Just show duplicates without action
mem dedupe --auto       # Auto-merge keeping newest content

# Adjust similarity threshold (lower = more similar required)
mem dedupe --threshold -3.0

When merging:

• Tags from all duplicates are combined
• Newest content is kept
• Older memories are deleted

Database management

mem init                   # initialize DB (auto on first use)
mem stats                  # show counts and DB size
mem stats --json           # JSON format
mem export backup.json     # export all memories
mem import backup.json     # import from file
mem rebuild-index          # rebuild FTS index (maintenance/recovery)

Global Options

Option	Description
--db PATH	Override database path (default: .sqler-cli/memory.db)
--json	Output as JSON (for scripts/LLMs)
--quiet	Minimal output (IDs only)

Environment variable: SQLER_CLI_DB sets default database path.

Output Formats

Default - Human-readable tables:

┏━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┓
┃ ID   ┃ Content                     ┃ Tags         ┃ Created        ┃
┡━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━┩
│ 1    │ API key is in .env          │ config       │ 2024-01-15 10:30│
│ 2    │ User prefers dark mode      │ preferences  │ 2024-01-15 10:32│
└──────┴─────────────────────────────┴──────────────┴────────────────┘

JSON (--json) - For programmatic access:

[
  {
    "id": 1,
    "content": "API key is in .env",
    "tags": ["config"],
    "context": null,
    "source": "user",
    "session_id": null,
    "supersedes": null,
    "see_also": [],
    "source_url": null,
    "source_file": null,
    "importance": 3,
    "created_at": "2024-01-15T10:30:00+00:00",
    "updated_at": "2024-01-15T10:30:00+00:00"
  }
]

With --show-score:

[
  {
    "id": 1,
    "content": "API key is in .env",
    "score": -2.45,
    ...
  }
]

Quiet (--quiet) - Just IDs:

1
2
3

Data Model

Each memory has:

Field	Type	Description
id	int	Auto-assigned unique identifier
content	string	The thing to remember (required, searchable)
tags	string[]	Categories for filtering
context	string	Why/where this was stored (searchable)
source	string	Who created it: "user", "claude", "file", etc.
session_id	string	Session for grouping related memories
supersedes	int	ID of memory this replaces
see_also	int[]	IDs of related memories
source_url	string	URL where info came from
source_file	string	File path where info came from
importance	int	Priority 1-5 (default: 3)
created_at	datetime	When created
updated_at	datetime	When last modified

Example Workflows

Session handoff

# End of session - save context
mem remember "Working on auth module, JWT refresh logic incomplete" \
  --tag session --session 2024-01-15-auth --importance 4

# New session - recall context
mem recall "session" --session 2024-01-15-auth
mem list --session 2024-01-15-auth

Project conventions

# Store project rules with auto-tagging
mem remember "Use snake_case for Python, camelCase for TypeScript" --tag conventions
mem remember "All API responses use {data, error, meta} envelope" --auto-tag --tag conventions

# Later, recall them
mem recall "conventions" --tag api

Decision log with linking

# Record initial decision
mem remember "Chose PostgreSQL over MySQL for JSON support" \
  --tag decisions --context "Database selection meeting"
# Returns: Remembered (id=42)

# Later, update the decision
mem remember "Switched to SQLite for simpler deployment" \
  --tag decisions --supersedes 42

# View with context
mem list --tag decisions --json  # Shows supersedes chain

Knowledge cleanup

# Find and review duplicates
mem dedupe --dry-run

# Auto-merge obvious duplicates
mem dedupe --auto

# Or interactively choose
mem dedupe

Importance-based filtering

# Mark critical info
mem remember "NEVER commit .env files" --importance 5 --tag security

# Quick recall of important stuff only
mem recall "security" --min-importance 4
mem list --min-importance 4

Database Location

Priority order:

1. --db PATH flag (explicit override)
2. SQLER_CLI_DB environment variable
3. --global flag → forces ~/.sqler-cli/memory.db
4. Local .sqler-cli/ exists in CWD → uses local
5. Fallback → global ~/.sqler-cli/memory.db

Global by default: Run mem from anywhere and it uses ~/.sqler-cli/memory.db.

Project-local opt-in: Run mem init in a project to create a local .sqler-cli/ directory. After that, commands in that directory use the local DB.

Force global: Use --global (or -g) to skip local detection and use global DB.

# From anywhere - uses global ~/.sqler-cli/
mem remember "Shared across all sessions"

# Create project-local DB
cd ~/projects/my-app
mem init                    # Creates .sqler-cli/ here
mem remember "Project-specific note"  # Goes to local DB

# Force global even when local exists
mem remember "Still global" --global
mem list --global

Tips for LLMs

1. Always use --json for reliable parsing
2. Use --quiet when you only need IDs for chaining
3. Tag consistently - use the same tags across sessions
4. Use sessions - isolate memories by conversation/task with --session
5. Auto-tag - let the CLI detect common categories with --auto-tag
6. Link memories - use --supersedes when updating info, --see-also for related
7. Mark importance - use --importance 5 for critical info, filter with --min-importance
8. Search is fuzzy - FTS5 handles stemming (e.g., "running" matches "run")
9. Context helps recall - store why you remembered something
10. Source tracking - use --source claude to track AI-generated memories
11. Check for similar - after remember, similar memories are shown to help avoid duplicates
12. Periodic cleanup - run mem dedupe --dry-run to find duplicates

Auto-Tag Categories

When using --auto-tag, the CLI detects these categories from content:

Tag	Detected Keywords
api	api, endpoint, rest, graphql, http
database	database, db, sql, postgres, sqlite, mongo
config	config, configuration, settings, .env, environment
auth	auth, authentication, jwt, oauth, password, login
error	error, exception, bug, fix, issue
security	security, secret, key, credential, token