Repository navigation
.NET: Python: Support durable Foundry-hosted workflows #7136
Description
Foundry Hosted Agents need to preserve a single stored background response while an Agent Framework workflow advances through multiple internal supersteps and may be restarted on replacement hosts.
The hosting adapter should:
- synchronize response checkpoints with workflow checkpoints without exposing incomplete workflow state;
- restore the correct workflow checkpoint after a crash or graceful host replacement;
- persist hosted checkpoint and approval state beneath the session's durable home directory;
- preserve emitted outputs and protected reasoning state across response history round-trips;
- recover only eligible stored background responses during shutdown;
- provide local and deployable samples that demonstrate repeated crash recovery within one response.
This issue tracks the durable workflow integration and its regression coverage for Foundry Hosted Agents.
Metadata
Metadata
Assignees
Labels
Type
Projects
- Status
Activity
[-]Python: Support durable Foundry-hosted workflows[/-][+].NET: Python: Support durable Foundry-hosted workflows[/+]eavanvalkenburg commentedon Nov 9, 2026
Python hosting release-scope decision
Future checkpoint redesign, expanded durable recovery, and new checkpoint-based retry guarantees are excluded from the current Python Foundry Hosting RC/stable scope. They will likely need to coincide with a Foundry Hosting 2.0 release alongside the relevant Core work; that is not a delivery commitment.
Retain this issue for traceability. This is a Python release-planning boundary, not a new .NET decision or a claim that every existing recovery path is broken. Preserve current supported behavior and handle independently scoped compatible defects normally; do not require the future design merely to freeze the current hosting API.