Register and serve models, build agents that reason over multiple steps, call tools, and expose AIDB's tools to external MCP clients.
Model serving
Register a model once to power every AI operation in AIDB, including embeddings, text chunking, summarization, OCR, and completions, and reuse it across pipelines, agents, and SQL functions.
Install and configure AIDB before registering a model — see Installing and configuring AIDB.
Register a local model that runs on the Postgres host, or a remote model called over an API, and store its credentials securely — see Local models, Remote models, and Model credentials.
Agents
Define agents as named, reusable configurations, including instructions, a model, and a set of tools, that AIDB runs through a ReAct-style reasoning loop to answer multi-step questions, entirely from SQL.
Register the model an agent will use before creating it — see Model serving.
Create an agent, converse with it, and let agents delegate work to each other — see Managing agents, Invoking agents, and Delegation.
Tools
List every built-in AIDB operation, custom parameterized SQL query, and operation imported from an external MCP server as a tool an agent can invoke by name, in one unified catalog.
AIDB's tool catalog is available once AIDB is installed and configured — see Installing and configuring AIDB.
Call AIDB's built-in operations, register your own SQL queries as tools, or import tools from an external MCP server — see Native tools, Custom SQL tools, and MCP tools.
EDB endpoints
The edb-endpoints server, run by an AIDB background worker, serves two protocols to clients outside the database: an MCP server that exposes AIDB's tool catalog, and an OTLP receiver that accepts telemetry from external agents into aidb_otel.
Turn on a protocol and set up a Postgres role for each client before connecting — see Configuring the endpoint, Enabling the MCP server, and Enabling the OTLP receiver.
Connect an MCP client over the MCP wire protocol and call tools or run SQL directly — see Connecting a client. Send OpenTelemetry traces, logs, and metrics from an external agent — see Sending telemetry.