AgentDB allows AI agents (and developers) to instantly provision databases with just a unique ID, without setup or managing compute. The serverless architecture scales from zero to millions of requests. AgentDB is optimized for AI applications with features like templates, MCP, and vector search out of the box.

AgentDB is designed for AI agents, developers building AI-powered applications, and teams that need to create databases at scale without infrastructure management overhead.

Agents create 1000x more databases than humans during conversations and tool calls, and can't wait for IT provisioning cycles. With AgentDB, an agent can create a database simply by generating a unique ID and use templates to immediately make use of the tables it needs.

Most agent-created databases won't get reused, but some will be valuable for future tasks. AgentDB's pay only for storage and queries model is much more cost-effective than traditional databases that charge for idle compute time, especially for the many ephemeral databases that AI agents create.

Templates keep all databases of a certain type synchronized with a common schema, and this schema is made available to agents through MCP.

Without templates, agents need to spend time learning what's in a database and how to use it, creating multiple opportunities for mistakes. AgentDB templates provide immediate context, enabling agents to properly write data into tables from the start or use already populated databases without additional work.

Templates consist of two components: a description section that explains the database's purpose, business logic, and enumeration mappings, and migration scripts that define the complete schema structure following database management best practices.

Without templates, agents face significant discovery overhead, typically requiring 4-7 API calls before performing any meaningful work as they struggle to understand database structure, available operations, and data relationships.

AgentDB functions as an MCP server with raw SQL access plus templates so agents can effectively use databases. You can use AgentDB directly from any agent like Claude Code, Cursor, or ChatGPT by registering its remote MCP server.

AgentDB includes built-in vector search, allowing you to store embeddings, perform semantic search, and build RAG systems out of the box. This eliminates the need to configure separate vector databases for AI applications.

Yes, you can upload or sync existing databases and create templates from them to standardize schema across multiple database instances.

Yes. Each database token is a UUID that acts as a namespace, grouping databases together and isolating them from databases under other tokens. This lets you organize databases by project, agent, or environment. You can switch tokens on the databases page to view and manage different sets of databases. Each user has a default token, but you can create and use as many as you need.

AgentDB currently supports both SQLite for fast, reliable transactional operations and DuckDB for powerful analytical queries, giving AI applications flexibility to choose the right engine for their specific use case.

Traditional databases store all user records in one database, making data portability and compliance complicated. AgentDB provides fully isolated databases that can be easily downloaded, reused as context, or deleted for compliance.

The AgentDB JavaScript SDK provides a simple interface for listing databases and executing SQL statements against AgentDB databases.

API keys are your programmatic access credentials for the AgentDB API. They let you authenticate from code, the CLI, and MCP clients instead of using browser-based login. You can create multiple named keys with optional expiration dates on your account page, and revoke any of them individually. Your API key is used by the SDK, the CLI tool, and when registering AgentDB as an MCP server.

You pay only for data stored and queries executed, with no server or configuration costs. Traditional databases require forecasting usage and paying for idle compute, whereas AgentDB costs decrease for low-access databases.