Agents¶
An agent is a core concept in the MatrixOne Intelligence product model. This page explains the problems it solves, its relationships with other objects, and its usage boundaries.
Definition¶
An agent combines models, instructions, knowledge bases, skills, tools, memory, and channels into a runnable configuration.
Conversation tasks expose execution through plans, tool calls, and streamed events.
Versions freeze reproducible configuration; publishing and automations run validated versions.
Object relationships¶
Knowledge bases provide controlled sources of facts, tools provide external or platform actions, and skills package reusable workflows.
Channels connect external events and messaging systems, while automation tasks provide scheduled, event, or API triggers.
Each run should be associated with the agent version, conversation, messages, and artifacts for traceability and rollback.
Design principles¶
Use least privilege and keep credentials in managed configuration.
Start by validating with minimal data or test objects, then expand the scope.
Version executable configuration and retain each run as an execution record.
Do not equate a visible entry with backend availability or resource authorization.
Note
These pages follow the current MOI 5.0 product model. If an entry is missing, check the workspace, permissions, and feature flags.