codex-as-api 0.9.2__tar.gz → 0.9.3__tar.gz

{codex_as_api-0.9.2 → codex_as_api-0.9.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: codex-as-api
-Version: 0.9.2
+Version: 0.9.3
 Summary: Use ChatGPT/Codex OAuth as a local OpenAI-compatible API server.
 Project-URL: Homepage, https://github.com/Eunho-J/codex-as-api
 Project-URL: Repository, https://github.com/Eunho-J/codex-as-api

{codex_as_api-0.9.2 → codex_as_api-0.9.3}/RELEASES.md

@@ -1,5 +1,21 @@
 # Release Notes
 
+## v0.9.3
+
+### Prompt cache routing headers
+
+- Send the session headers the official Codex CLI sends on every Responses
+  request: `session-id`, `thread-id` and `x-client-request-id`, derived from
+  `client_metadata.session_id` / `thread_id` or, failing those, from
+  `prompt_cache_key`. Applies to streamed `/responses` and `/responses/compact`
+  requests in Python, TypeScript and Rust. Values containing CR or LF are not sent.
+- Measured against the ChatGPT Codex backend: an identical 7k-token prefix sent
+  with `prompt_cache_key` alone read 0 cached tokens on every request; with the
+  headers, repeated requests read 7,296 cached tokens. The official Codex CLI on
+  the same account read ~30k cached tokens per turn.
+- No change when a request carries neither `prompt_cache_key` nor
+  `client_metadata.session_id`.
+
 ## v0.9.2
 
 ### TypeScript version lookup deadline

{codex_as_api-0.9.2 → codex_as_api-0.9.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "codex-as-api"
-version = "0.9.2"
+version = "0.9.3"
 description = "Use ChatGPT/Codex OAuth as a local OpenAI-compatible API server."
 readme = "README.md"
 license = { text = "Apache-2.0" }

{codex_as_api-0.9.2 → codex_as_api-0.9.3}/src/codex_as_api/__init__.py

@@ -1,3 +1,3 @@
 """codex-as-api: Use ChatGPT/Codex OAuth as a local OpenAI-compatible API server."""
 
-__version__ = "0.9.2"
+__version__ = "0.9.3"

{codex_as_api-0.9.2 → codex_as_api-0.9.3}/src/codex_as_api/provider.py

@@ -409,6 +409,29 @@ def _redact_failure_event(
     return value
 
 
+def codex_session_headers(payload: dict[str, Any]) -> dict[str, str]:
+    """Session headers the official Codex CLI sends on every Responses request.
+
+    The ChatGPT Codex backend does not reuse its prompt cache from
+    ``prompt_cache_key`` alone: measured 2026-10-27, an identical 7k-token prefix
+    read 0 cached tokens on every request without these headers and 7,296 with them.
+    """
+    metadata = payload.get("client_metadata")
+    if not isinstance(metadata, dict):
+        metadata = {}
+
+    def non_empty(value: Any) -> str | None:
+        return value if isinstance(value, str) and value.strip() else None
+
+    session_id = non_empty(metadata.get("session_id")) or non_empty(payload.get("prompt_cache_key"))
+    if session_id is None:
+        return {}
+    thread_id = non_empty(metadata.get("thread_id")) or session_id
+    if any(ch in value for value in (session_id, thread_id) for ch in ("\r", "\n")):
+        return {}
+    return {"session-id": session_id, "thread-id": thread_id, "x-client-request-id": thread_id}
+
+
 class ChatGPTOAuthProvider:
     name: str = "chatgpt_oauth"
     provider_namespace: str = "agent.provider.chatgpt_oauth"
@@ -1558,6 +1581,7 @@ class ChatGPTOAuthProvider:
             _require_catalog_account(token.account_id, catalog_key)
             headers = self._headers(token, client_version=resolved_client_version)
             headers["Accept"] = "text/event-stream"
+            headers.update(codex_session_headers(payload))
             if extra_headers:
                 headers.update(extra_headers)
             token_values = (token.access_token, token.refresh_token, token.id_token, token.account_id)
@@ -1649,6 +1673,7 @@ class ChatGPTOAuthProvider:
             token = token_for_request(self.auth_json_path)
             _require_catalog_account(token.account_id, catalog_key)
             headers = self._headers(token, client_version=resolved_client_version)
+            headers.update(codex_session_headers(payload))
             if extra_headers:
                 headers.update(extra_headers)
             token_values = (token.access_token, token.refresh_token, token.id_token, token.account_id)

{codex_as_api-0.9.2 → codex_as_api-0.9.3}/src/codex_as_api/server.py

@@ -465,7 +465,7 @@ def _health_error_message(exc: BaseException) -> str:
 app = FastAPI(
     title="codex-as-api",
     description="Local OpenAI-compatible API server backed by ChatGPT/Codex OAuth.",
-    version="0.9.2",
+    version="0.9.3",
 )
 
 

{codex_as_api-0.9.2 → codex_as_api-0.9.3}/tests/test_provider.py

@@ -53,6 +53,7 @@ from codex_as_api.provider import (
     _validate_response_event,
     _web_search_event_from_response_item,
     codex_cli_headers_for_version,
+    codex_session_headers,
     resolve_codex_cli_version,
     resolve_model_reasoning_effort,
 )
@@ -2192,7 +2193,7 @@ def test_codex_cli_headers_include_official_originator_and_versioned_user_agent(
 
     assert headers["originator"] == "codex_cli_rs"
     assert headers["User-Agent"].startswith("codex_cli_rs/0.164.0 (")
-    assert headers["User-Agent"].endswith(") codex-as-api/0.9.2")
+    assert headers["User-Agent"].endswith(") codex-as-api/0.9.3")
 
 
 def test_codex_cli_headers_reject_unpinned_version():
@@ -2932,7 +2933,7 @@ def test_provider_headers_include_codex_cli_headers(auth_json_factory):
 
     assert headers["originator"] == "codex_cli_rs"
     assert headers["User-Agent"].startswith("codex_cli_rs/0.164.0 (")
-    assert headers["User-Agent"].endswith(") codex-as-api/0.9.2")
+    assert headers["User-Agent"].endswith(") codex-as-api/0.9.3")
     assert headers["Authorization"].startswith("Bearer ")
 
 
@@ -4533,7 +4534,7 @@ def test_provider_injects_dynamic_npm_version_across_catalog_responses_and_401_r
     for full_url, headers in observed_requests:
         assert full_url == "https://example.test/codex/models?client_version=0.175.2"
         assert headers["user-agent"].startswith("codex_cli_rs/0.175.2 (")
-        assert headers["user-agent"].endswith(") codex-as-api/0.9.2")
+        assert headers["user-agent"].endswith(") codex-as-api/0.9.3")
 
 
 def test_provider_cold_version_resolver_failure_raises_typed_503(auth_json_factory, monkeypatch):
@@ -4682,3 +4683,61 @@ def test_responses_payload_replays_all_25_compatible_model_pairs(
                 "encrypted_content": "enc_payload",
                 "summary": [{"type": "summary_text", "text": "summary"}],
             }
+
+
+def test_codex_session_headers_derive_from_cache_key_and_metadata():
+    assert codex_session_headers({"prompt_cache_key": "key-1"}) == {
+        "session-id": "key-1",
+        "thread-id": "key-1",
+        "x-client-request-id": "key-1",
+    }
+    assert codex_session_headers(
+        {"prompt_cache_key": "key-1", "client_metadata": {"session_id": "s-1", "thread_id": "t-1"}}
+    ) == {"session-id": "s-1", "thread-id": "t-1", "x-client-request-id": "t-1"}
+    assert codex_session_headers({"client_metadata": {"session_id": "s-2"}}) == {
+        "session-id": "s-2",
+        "thread-id": "s-2",
+        "x-client-request-id": "s-2",
+    }
+    assert codex_session_headers({}) == {}
+    assert codex_session_headers({"prompt_cache_key": "   "}) == {}
+    assert codex_session_headers({"prompt_cache_key": "a\r\nInjected: 1"}) == {}
+    assert codex_session_headers({"client_metadata": {"session_id": "s", "thread_id": "t\nx"}}) == {}
+
+
+def test_responses_requests_carry_codex_session_headers_only_with_a_key(monkeypatch):
+    seen: list[tuple[str, dict[str, str]]] = []
+
+    class Handler(BaseHTTPRequestHandler):
+        def do_POST(self) -> None:  # noqa: N802
+            seen.append((self.path, {k.lower(): v for k, v in self.headers.items()}))
+            self.send_response(400)
+            self.end_headers()
+
+        def log_message(self, _format, *args):  # noqa: ANN001
+            del args
+
+    server = ThreadingHTTPServer(("127.0.0.1", 0), Handler)
+    thread = threading.Thread(target=server.serve_forever, daemon=True)
+    thread.start()
+    try:
+        token = _test_token()
+        provider = ChatGPTOAuthProvider(base_url=f"http://127.0.0.1:{server.server_port}")
+        monkeypatch.setattr(provider_module, "token_for_request", lambda _path: token)
+        with pytest.raises(ChatGPTOAuthUpstreamError):
+            list(provider._post_sse("/responses", {"prompt_cache_key": "cache-key-7"}))  # noqa: SLF001
+        with pytest.raises(ChatGPTOAuthUpstreamError):
+            provider._request_json("/responses/compact", {"prompt_cache_key": "cache-key-7"})  # noqa: SLF001
+        with pytest.raises(ChatGPTOAuthUpstreamError):
+            list(provider._post_sse("/responses", {}))  # noqa: SLF001
+    finally:
+        server.shutdown()
+        server.server_close()
+        thread.join(timeout=2)
+    assert [path for path, _ in seen] == ["/responses", "/responses/compact", "/responses"]
+    for _, headers in seen[:2]:
+        assert headers["session-id"] == "cache-key-7"
+        assert headers["thread-id"] == "cache-key-7"
+        assert headers["x-client-request-id"] == "cache-key-7"
+    assert "session-id" not in seen[2][1]
+    assert "thread-id" not in seen[2][1]

{codex_as_api-0.9.2 → codex_as_api-0.9.3}/tests/test_server.py

@@ -6648,7 +6648,7 @@ def test_dynamic_npm_version_resolver_updates_models_query_user_agent_and_cold_f
     responses_req_1 = recording_backend.requests.get(timeout=1)
     assert catalog_req_1["path"] == "/models?client_version=0.175.2"
     assert catalog_req_1["headers"]["user-agent"].startswith("codex_cli_rs/0.175.2 (")
-    assert catalog_req_1["headers"]["user-agent"].endswith(") codex-as-api/0.9.2")
+    assert catalog_req_1["headers"]["user-agent"].endswith(") codex-as-api/0.9.3")
     assert responses_req_1["headers"]["user-agent"].startswith("codex_cli_rs/0.175.2 (")
 
     active_version = "0.176.0"