## Why The Python SDK previously protected the stdio transport with a single active turn-consumer guard. That avoided competing reads from stdout, but it also meant one `Codex`/`AsyncCodex` client could not stream multiple active turns at the same time. Notifications could also arrive before the caller received a `TurnHandle` and registered for streaming, so the SDK needed an explicit routing layer instead of letting individual API calls read directly from the shared transport. ## What Changed - Added a private `MessageRouter` that owns per-request response queues, per-turn notification queues, pending turn-notification replay, and global notification delivery behind a single stdout reader thread. - Generated typed notification routing metadata so turn IDs come from known payload shapes instead of router-side attribute guessing, with explicit fallback handling for unknown notification payloads. - Updated sync and async turn streaming so `TurnHandle.stream()`/`run()` and `stream_text()` consume only notifications for their own turn ID, while `AsyncAppServerClient` no longer serializes all transport calls behind one async lock. - Cleared pending turn-notification buffers when unregistered turns complete so never-consumed turn handles do not leave stale queues behind. - Removed the internal stream-until helper now that turn completion waiting can register directly with routed turn notifications. - Updated Python SDK docs and focused tests for concurrent transport calls, interleaved turn routing, buffered early notifications, unknown notification routing, async delegation, and routed turn completion behavior. ## Validation - `uv run --extra dev ruff format scripts/update_sdk_artifacts.py src/codex_app_server/_message_router.py src/codex_app_server/client.py src/codex_app_server/generated/notification_registry.py tests/test_client_rpc_methods.py tests/test_public_api_runtime_behavior.py tests/test_async_client_behavior.py` - `uv run --extra dev ruff check scripts/update_sdk_artifacts.py src/codex_app_server/_message_router.py src/codex_app_server/client.py src/codex_app_server/generated/notification_registry.py tests/test_client_rpc_methods.py tests/test_public_api_runtime_behavior.py tests/test_async_client_behavior.py` - `uv run --extra dev pytest tests/test_client_rpc_methods.py tests/test_public_api_runtime_behavior.py tests/test_async_client_behavior.py` - `git diff --check` --------- Co-authored-by: Codex <noreply@openai.com>
Codex App Server Python SDK (Experimental)
Experimental Python SDK for codex app-server JSON-RPC v2 over stdio, with a small default surface optimized for real scripts and apps.
The generated wire-model layer is currently sourced from the bundled v2 schema and exposed as Pydantic models with snake_case Python fields that serialize back to the app-server’s camelCase wire format.
Install
cd sdk/python
uv sync
source .venv/bin/activate
Published SDK builds pin an exact openai-codex-cli-bin runtime dependency
with the same version as the SDK. For local repo development, either pass
AppServerConfig(codex_bin=...) to point at a local build explicitly, or use
the repo examples/notebook bootstrap which installs the pinned runtime package
automatically.
Quickstart
from codex_app_server import Codex
with Codex() as codex:
thread = codex.thread_start(model="gpt-5")
result = thread.run("Say hello in one sentence.")
print(result.final_response)
print(len(result.items))
result.final_response is None when the turn completes without a final-answer
or phase-less assistant message item.
Docs map
- Golden path tutorial:
docs/getting-started.md - API reference (signatures + behavior):
docs/api-reference.md - Common decisions and pitfalls:
docs/faq.md - Runnable examples index:
examples/README.md - Jupyter walkthrough notebook:
notebooks/sdk_walkthrough.ipynb
Examples
Start here:
cd sdk/python
python examples/01_quickstart_constructor/sync.py
python examples/01_quickstart_constructor/async.py
Runtime packaging
The repo no longer checks codex binaries into sdk/python.
Published SDK builds are pinned to an exact openai-codex-cli-bin package
version, and that runtime package carries the platform-specific binary for the
target wheel. The SDK package version and runtime package version must match.
For local repo development, the checked-in sdk/python-runtime package is only
a template for staged release artifacts. Editable installs should use an
explicit codex_bin override for manual SDK usage; the repo examples and
notebook bootstrap the pinned runtime package automatically.
Maintainer workflow
cd sdk/python
python scripts/update_sdk_artifacts.py generate-types
python scripts/update_sdk_artifacts.py \
stage-sdk \
/tmp/codex-python-release/openai-codex-app-server-sdk \
--codex-version <codex-release-tag-or-pep440-version>
python scripts/update_sdk_artifacts.py \
stage-runtime \
/tmp/codex-python-release/openai-codex-cli-bin \
/path/to/codex \
--codex-version <codex-release-tag-or-pep440-version>
Pass --platform-tag ... to stage-runtime when the wheel should be tagged for
a Rust target that differs from the Python build host. The intended one-off
matrix is macosx_11_0_arm64, macosx_10_9_x86_64,
musllinux_1_1_aarch64, musllinux_1_1_x86_64, win_arm64, and
win_amd64.
This supports the CI release flow:
- run
generate-typesbefore packaging - stage
openai-codex-app-server-sdkonce with an exactopenai-codex-cli-bin==...dependency - stage
openai-codex-cli-binon each supported platform runner with the same pinned runtime version - build and publish
openai-codex-cli-binas platform wheels only; do not publish an sdist
Compatibility and versioning
- Package:
openai-codex-app-server-sdk - Runtime package:
openai-codex-cli-bin - Python:
>=3.10 - Target protocol: Codex
app-serverJSON-RPC v2 - Versioning rule: the SDK package version is the underlying Codex runtime version
Notes
Codex()is eager and performs startup +initializein the constructor.- Use context managers (
with Codex() as codex:) to ensure shutdown. - Prefer
thread.run("...")for the common case. Usethread.turn(...)when you need streaming, steering, or interrupt control. - For transient overload, use
codex_app_server.retry.retry_on_overload.