v0.117.0

[0.117.0] - 2026-03-28

Added

• App-server helpers for newly exposed upstream RPCs:
• thread_shell_command, thread_background_terminals_clean,
• experimental_feature_enablement_set, fs_watch, and fs_unwatch.
• CodexModel(..., hooks=ThreadHooks(...)) now dispatches typed SDK thread hooks for
• app-server-backed PydanticAI requests, including thread start, turn start/completion,
• turn failure, and item lifecycle events.

Updated

• AppServerOptions now supports opt_out_notification_methods, wiring
• initialize.capabilities.optOutNotificationMethods for app-server clients that
• want to suppress selected notification streams.
• The PydanticAI model-provider path now runs through upstream
• Model.prepare_request(), keeping request customization, output-mode handling,
• builtin-tool validation, and thinking resolution aligned with PydanticAI 1.73.0.
• CodexModel now maps PydanticAI model_settings["thinking"] onto Codex
• model_reasoning_effort when no explicit thread-level reasoning effort is set.
• PydanticAI dependency support updated to >=1.73.0,<2, with the dev dependency
• pinned to 1.73.0 for reproducible test coverage against the latest release.
• thread_list now supports the upstream cwd and search_term filters.
• plugin_list now supports force_remote_sync, config_batch_write now supports
• reload_user_config, and windows_sandbox_setup_start now accepts an optional
• workspace cwd.
• README and PydanticAI examples updated to show typed output models, runtime hooks,
• and per-run thinking configuration.

Notes

• Codex 0.116.0 and 0.117.0 introduced app-server support for thread-scoped !
• shell commands, filesystem watch subscriptions, richer plugin sync flows, and
• per-connection notification suppression.