Autonomous planning
Run multi-step agents that decide which tool to call next until the goal is complete.
Autonomous agents
Run autonomous AI agents directly inside SQLite.
SQLite-Agent lets SQLite databases execute multi-step, tool-using agents that can fetch external data, populate tables, and prepare results for AI, search, and applications.
SQLite-Agent
SQLite-native extension
Best for
Autonomous data agents
Repository
GitHubProject
What it is
SQLite-Agent is a cross-platform SQLite extension that brings autonomous AI agents into the database. Agents can plan multi-step tasks, call external tools through SQLite-MCP, use SQLite-AI for model inference, and populate SQLite tables with structured results.
It is designed for edge apps and local workflows that need agents to fetch web or API data, persist it in SQLite, generate embeddings, and make data immediately queryable without a separate orchestration service.
Why it matters
Agents that write back
to SQLite.
Agent output is most useful when it becomes structured data. SQLite-Agent turns goals into rows that can be queried, analyzed, embedded, synced, and displayed by the rest of the application.
By combining MCP tools, local inference, table extraction, and optional vector indexing, it makes SQLite the execution surface for data-gathering agents that need to run close to the user or device.
Capabilities
Features and characteristics
MCP tool access
Use SQLite-MCP to connect agents to external tools, web APIs, and live services.
Table extraction mode
Populate target SQLite tables automatically from agent results using the table schema.
AI-powered execution
Use SQLite-AI for local LLM inference, embeddings, and model-powered reasoning.
Auto-embeddings
Generate embeddings for matching BLOB columns and prepare vector search automatically.
Cross-platform packages
Use prebuilt binaries and packages across Linux, macOS, Windows, Android, iOS, Swift, and Flutter.
Sample code
Run an agent that populates a table
Load SQLite-Agent with its MCP, AI, and vector dependencies, connect a tool server, and let the agent extract structured rows.
-- Load required extensions
.load ./mcp
.load ./ai
.load ./vector
.load ./agent
-- Initialize AI model and tool access
SELECT llm_model_load('/path/to/model.gguf', 'gpu_layers=99');
SELECT mcp_connect('http://localhost:8000/mcp');
CREATE TABLE listings (
id INTEGER PRIMARY KEY,
title TEXT,
location TEXT,
price REAL,
rating REAL,
embedding BLOB
);
-- The agent uses MCP tools and fills the table.
-- Embeddings and vector indexing are created automatically.
SELECT agent_run(
'Find affordable apartments in Rome under 100 EUR per night',
'listings',
8
);
SELECT title, location, price
FROM listings
ORDER BY rating DESC;