[Feature Request] Allow passing existing chat history or a runtime context for multi-turn orchestration

[hjerbii]

Hello!

Currently, the Magentic orchestration in the Agent Framework (MagenticBuilder / workflow.run*) only accepts a single string task input and internally manages its own MagenticContext.
There is no supported way to initialize or inject an existing chat history (e.g., a list of ChatMessageContent objects) into a new workflow run.

This limitation makes it impossible to:

• Replay or re-run a workflow using an existing conversation.
• Start a Magentic orchestration with prior user–assistant context (multi-turn setup).
• Integrate Magentic into existing conversational agents or chat interfaces that already maintain message history.

Checkpoints are not a solution here (please correct me if I am wrong), as they only allow resuming from saved states within a previous interrupted run, not injecting prior dialogue into a new orchestration.

I am aware that in Semantic Kernel passing a chat history was not supported as well, but is it possible to implement this feature?

Thanks in advance!

[APiccardo4Quantis]

i was looking forward same features,
i want my magentic for multipleturns, in a user/thread-like style .

[droideronline]

I’d really like to see this feature added as well. I’m coming from Autogen, and it would be great if we could have a similar capability — the ability to save the state of orchestrators (and even ChatAgents) so that we can resume from the middle of a ChatAgent execution.

I know ChatAgent already supports serialize and deserialize, but having a public API for saving and restoring state would make this much more convenient and accessible.

cc- @ekzhu

[ekzhu]

I know ChatAgent already supports serialize and deserialize, but having a public API for saving and restoring state would make this much more convenient and accessible.

ChatAgent uses thread for its state. The thread itself can be modified externally, and serialized and deserialized.

Thanks for the feedback. I think formalizing the concept of "prior history" is useful here. We should consider matching the thread concept in agent with a corresponding concept in workflows

[droideronline]

I know ChatAgent already supports serialize and deserialize, but having a public API for saving and restoring state would make this much more convenient and accessible.

ChatAgent uses thread for its state. The thread itself can be modified externally, and serialized and deserialized.

Thanks for the feedback. I think formalizing the concept of "prior history" is useful here. We should consider matching the thread concept in agent with a corresponding concept in workflows

Added #1698

[MattB-msft]

You can do this with the Agents SDK if your using it to wrap the Agent Framework Orchestration and tool calls.
this is demonstrated here: https://github.com/microsoft/Agents/tree/main/samples/dotnet/Agent%20Framework

We will add this sample to this repo at some point in the future,
and we are adding a variant of this sample as a template tin the Agents Toolkit and AI toolkit.

though I agree it would be nice to be able to access the thread content from inside a tool call.

[TaoChenOSU]

This has been addressed by this PR: #1538