SQLsaber

SQLsaber is an open-source agentic SQL assistant. Think Claude Code but for SQL.

Stop fighting your database.

Ask your questions in natural language and sqlsaber will gather the right context automatically and answer your query by writing SQL and analyzing the results.

Table of Contents

• Features

• Installation

• Configuration

  • Database Connection
  • AI Model Configuration
  • Memory Management

• Usage

  • Interactive Mode
  • Single Query
  • Resume Past Conversation
  • Database Selection

• Examples

• How It Works

• Contributing

• License

Features

• Automatic database schema introspection

• Safe query execution (read-only by default)

• Memory management

• Interactive REPL mode

• Conversation threads (store, display, and resume conversations)

• Support for PostgreSQL, MySQL, SQLite, DuckDB, and CSVs

• Extended thinking mode for select models (Anthropic, OpenAI, Google, Groq)

• Beautiful formatted output

Installation

uv

uv tool install sqlsaber

pipx

pipx install sqlsaber

brew

brew install uv
uv tool install sqlsaber

Configuration

Database Connection

Set your database connection URL:

saber db add DB_NAME

This will ask you some questions about your database connection

AI Model Configuration

SQLSaber uses Sonnet-4 by default. You can change it using:

saber models set

# for more model settings run:
saber models --help

Memory Management

You can add specific context about your database to the model using the memory feature. This is similar to how you add memory/context in Claude Code.

saber memory add 'always convert dates to string for easier formating'

View all memories

saber memory list

You can also add memories in an interactive query session by starting with the # sign

Extended Thinking Mode

For complex queries that require deeper reasoning, sqlsaber supports extended thinking mode. When enabled, you will see the model's reasoning process as it generates SQL queries and arrives at conclusions.

Enable/disable via CLI flags:

# Enable thinking for a single query
saber --thinking "analyze sales trends across regions"

# Disable thinking for a single query
saber --no-thinking "show me all users"

Toggle in interactive mode:

# In interactive mode, use slash commands
/thinking on   # Enable thinking
/thinking off  # Disable thinking

Configure default setting:

Thinking is disabled by default. To change the default, edit your config file at ~/.config/sqlsaber/model_config.json:

{
  "model": "anthropic:claude-sonnet-4-20250514",
  "thinking_enabled": true
}

Usage

Interactive Mode

Start an interactive session:

saber

You can also add memories in an interactive session by starting your message with the # sign

Single Query

Execute a single natural language query:

saber "show me all users created this month"

You can also pipe queries from stdin:

echo "show me all users created this month" | saber
cat query.txt | saber

Resume Past Conversation

Continue a previous conversation thread:

saber threads resume THREAD_ID

Database Selection

Use a specific database connection:

# Interactive mode with specific database
saber -d mydb

# Single query with specific database
saber -d mydb "count all orders"

# You can also pass a connection string
saber -d "postgresql://user:password@localhost:5432/mydb" "count all orders"
saber -d "duckdb:///path/to/data.duckdb" "top customers"

Examples

# Start interactive mode
saber

# Non-interactive mode
saber "show me orders with customer details for this week"

saber "which products had the highest sales growth last quarter?"

How It Works

SQLsaber uses a multi-step agentic process to gather the right context and execute SQL queries to answer your questions:

🔍 Discovery Phase

1. List Tables Tool: Quickly discovers available tables with row counts
2. Pattern Matching: Identifies relevant tables based on your query

📋 Schema Analysis

3. Smart Schema Introspection: Analyzes only the specific table structures needed for your query

⚡ Execution Phase

4. SQL Generation: Creates optimized SQL queries based on natural language input
5. Safe Execution: Runs read-only queries with built-in protections against destructive operations
6. Result Formatting: Presents results with explanations in tables

Contributing

If you like the project, starring the repo is a great way to show your support!

Other contributions are welcome! Please feel free to open an issue to discuss your ideas or report bugs.

License

This project is licensed under Apache-2.0 License - see the LICENSE file for details.