Azure/azure-sdk-for-python azure-ai-agentserver-core

2.0.0b8 (2026-07-22)

Added azure.ai.agentserver.core.storage, the protocol-neutral Foundry durable state-store layer. FoundryStateStore.get_or_create(name, ...) is the primary entry point -- an async classmethod that resolves (or creates, on first use) the store in one call. Store-level operations act on the bound store: get() (its descriptor), update(...) (its mutable description / tags), and delete() (the whole store, cascade-deleting every item). Item operations are explicit and consistently named: create_item, set_item, get_item, delete_item, list_keys. Store names are path-encoded with base64url on the wire, store-level item_ttl_seconds is configured once at create, optional user_isolation is declared at store create, and trusted callers may delegate end-user partitioning with user_id (x-ms-user-id). Response bodies (StateStore, StateStoreItem, StateStoreItemRef, DeletedStateStore, DeletedStateStoreItem, StateStoreItemKey) are typed model classes generated from a formal TypeSpec contract, not hand-written dataclasses. See the Durable State Store Guide.

Fixed span attribute enrichment under opentelemetry-sdk >= 1.43.0, where span._attributes became a BoundedAttributes (backed by ._dict) that no longer supports item assignment. Enrichment now resolves the backing store so agent identity attributes are written on both older and newer OpenTelemetry SDKs.