Python: [Feature]: Add Checkpointing Support to AgentFrameworkWorkflow.run() in agent-framework-ag-ui

[mjod]

Summary

The AgentFrameworkWorkflow implementation provided by agent-framework-ag-ui currently exposes a run() method that only accepts a RunAgentInput. This differs from the core agent-framework workflow implementations, which support checkpointing and workflow state persistence.

This limitation prevents AG-UI-based workflows from taking advantage of one of the framework's key capabilities: resumable execution through checkpointing.

Problem

Today, the AG-UI adapter provides its own AgentFrameworkWorkflow abstraction, but its run() API does not expose any mechanism for:

• Creating checkpoints during execution
• Restoring workflow state from a checkpoint
• Resuming interrupted executions
• Integrating with existing checkpoint storage providers

As a result, applications built on agent-framework-ag-ui lose functionality that is otherwise available when using standard workflows in agent-framework.

This creates an inconsistency in the developer experience and makes AG-UI workflows less suitable for long-running or fault-tolerant agent scenarios.

Proposed Solution

Extend the AgentFrameworkWorkflow.run() API in agent-framework-ag-ui to support checkpointing in a manner consistent with the core agent-framework workflow APIs.

Potential approaches could include:

Option 1: Support Workflow Context

Allow run() to accept the same workflow context or execution options used by the core framework, enabling checkpoint providers to be configured and utilized during execution.

Option 2: Expose Checkpoint Parameters

Add optional checkpoint-related arguments to run(), such as:

await workflow.run(
    input=run_agent_input,
    checkpoint_storage=checkpoint_storage,
    checkpoint_id=checkpoint_id,
)

Option 3: Align with Core Workflow API

Refactor the AG-UI workflow implementation to use the same execution model as the standard framework workflows so that checkpointing works without introducing AG-UI-specific APIs.

Benefits

• Feature parity between agent-framework and agent-framework-ag-ui
• Support for long-running agent executions
• Improved fault tolerance and recovery
• Easier integration with persistent storage backends (PostgreSQL, Redis, etc.)
• Consistent developer experience across workflow implementations

Use Case

Many production deployments require durable execution. For example, an AG-UI workflow may orchestrate multiple agents, tool calls, and user interactions over an extended period of time.

If the process crashes or is redeployed, the workflow currently must be restarted from the beginning. With checkpointing support, execution could resume from the most recent checkpoint, matching the capabilities available in the core framework.

Code Sample

Language/SDK

Python