Modifying the Foundry Sample Workflow HITL with Checkpoints to use the FoundryCheckpointRepository

[colinrippeyciendos]

Hi,

I've been tweaking the sample for hitl checkpointing:

https://github.com/microsoft-foundry/foundry-samples/blob/main/samples/python/hosted-agents/agent-framework/human-in-the-loop/workflow-agent-with-checkpoint-and-hitl/README.md

The sample uses the FileCheckpointRepository to save workflow checkpoints to allow for HITL responses.

In my head if I were to deploy this sample as a hosted agent the checkpoints would be written to the local file system of the container, and if I have more that 1 replica (or my replicas scale to 0) I would guess the checkpoints would not be found if the hitl response is handled by a different replica.

I see in the package codebase there is a FoundryCheckpointRepository class and I've tried to shoe-horn this into the sample. I am of course "guessing" that this class is designed to use a Foundry project backend to handle the checkpointing (the guess is it will use the cosmos capability host or the storage capability host...or something that means my hitl response will "find" the checkpoint).

When I run the code what I see is that a call is made to the Foundry project endpoint:

PUT https://.services.ai.azure.com/api/projects//checkpoints/sessions/bc6b0a8e-ac38-493f-b993-16899ec65d3c?api-version=2025-11-15-preview HTTP/1.1

But this results in a 404.

Should I be using the FoundryCheckpointRepository class, and if so what are the pre-requisites I need to have configured on the Foundry Prpject.

TIA

Colin

[colinrippeyciendos]

Don't know if this will help but when I deploy the workflow as a hosted agent into a foundry project and then start a workflow I see the attached log which has a stack trace of all the various package calls that eventually result in the 404 error above being surfaced.

container-logs-2026-03-13T17-32-00-660Z.log