Architecture

Pygent now prioritizes a simple single-agent architecture.

Core components

• Agent: orchestrates messages, model calls, and tool execution.
• Runtime: executes commands and file operations in local or Docker workspace.
• Model (OpenAIModel by default): abstraction for OpenAI-compatible chat backends.
• tools registry: global registry used to expose callable actions to the model.

Default request flow

1. User message enters Agent.
2. Agent updates history and sends context + tool schemas to model.
3. Model returns text and/or tool calls.
4. Agent executes tool calls through Runtime.
5. Results are appended to history and displayed.

Optional legacy layer

A separate legacy layer still exists for delegated background tasks:

• pygent.task_manager.TaskManager
• pygent.task_tools.register_task_tools()

This layer is opt-in and not part of the default CLI flow.