Introduction

Encord Agents

Encord Agents allow you to automate and customize your annotation workflows with minimal code.

Agents can be hosted on your own infrastructure or in the cloud, including platforms like GCP.

They are divided into two main categories:

• Task Agents automate operations at the project and task level.
• Editor Agents interact directly with the annotation editor to assist with data labeling.

Use Agents to extend the Encord platform programmatically, enabling advanced automation, quality assurance, and model-assisted labeling.

Choosing the Right Type of Agent

Task Agents run as part of your Encord Workflow, executing custom logic on each task. They are ideal for:

• Pre-labeling using foundation models (for example: GPT-4o)
• Automated quality checks
• Custom task routing and orchestration

• Editor Agents are custom API endpoints that integrate with the Encord annotation editor. Use them to:

• Extend in-editor functionality
• Perform real-time inference or validation
• Enable context-aware labeling tools