.NET: [Feature]: Add persistence support for Agent Plan and Todo state

[naruto1227]

Description

Feature Request

Currently, Agent Plan and Todo state are maintained in memory during an agent run.

This works well for short-lived agent executions, but it makes it difficult to support long-running agents, process restarts, or scenarios where an agent needs to pause and resume later.

It would be useful to provide a persistence mechanism for Plan and Todo state.

Use Cases

• Resume an agent run after the process restarts.
• Support long-running agent tasks.
• Persist Todo progress across agent sessions.
• Support human-in-the-loop scenarios where the agent waits for user approval or input.
• Allow external Agent Runtimes to store and restore execution state.

Proposed Direction

Expose a persistence/checkpoint mechanism for Plan and Todo state, without requiring the framework to dictate a specific storage implementation.

For example, the framework could provide:

• Serialization/deserialization of Plan and Todo state.
• Checkpoint hooks when Plan or Todo state changes.
• A pluggable persistence abstraction that can be implemented using different storage backends.

This would allow applications and Agent Runtimes to persist and restore the agent's execution state while keeping the existing in-memory behavior as the default.

Expected Benefit

This would make Plan and Todo more suitable for long-running, resumable, and human-in-the-loop agent scenarios.

Code Sample

Language/SDK

.NET