Skip to content

Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.

fix(proxy): strip Codex lite header on the HTTP /responses path (#1663)

gglucass9fbd47ba6bdf
diff --git a/headroom/proxy/handlers/openai.py b/headroom/proxy/handlers/openai.py
index 4e7379924..11de719ac 100644
--- a/headroom/proxy/handlers/openai.py
+++ b/headroom/proxy/handlers/openai.py
@@ -3071,6 +3071,14 @@ class OpenAIHandlerMixin:
_pre_strip_count_resp = sum(1 for k in headers if k.lower().startswith("x-headroom-"))
headers = _strip_internal_headers(headers)
+ # Mirror the WS handler: never forward Codex's client-only lite header
+ # upstream. OpenAI rejects newer Codex models when it leaks, and the HTTP
+ # POST path (unlike the WS path) otherwise forwards request headers verbatim.
+ headers = {
+ key: value
+ for key, value in headers.items()
+ if key.lower() != _CODEX_RESPONSES_LITE_HEADER
+ }
log_outbound_headers(
forwarder="openai_responses",
stripped_count=_pre_strip_count_resp,
diff --git a/tests/test_openai_codex_routing.py b/tests/test_openai_codex_routing.py
index d8be9e6e3..f08ab7eed 100644
--- a/tests/test_openai_codex_routing.py
+++ b/tests/test_openai_codex_routing.py
@@ -288,6 +288,39 @@ def test_handle_openai_responses_routes_chatgpt_auth_to_backend_api(monkeypatch)
assert response.status_code == 200
+def test_handle_openai_responses_strips_codex_lite_header_upstream(monkeypatch):
+ # OpenAI rejects newer Codex models when the client-only lite header leaks
+ # upstream. The HTTP POST path must drop it like the WS handler does, while
+ # leaving adjacent headers intact.
+ token = _jwt(
+ {
+ "https://api.openai.com/auth": {
+ "chatgpt_account_id": "acct-from-jwt",
+ }
+ }
+ )
+ request = _build_request(
+ {"model": "gpt-5.4", "input": "hello"},
+ {
+ "Authorization": f"Bearer {token}",
+ "X-OpenAI-Internal-Codex-Responses-Lite": "true",
+ "X-OpenAI-Debug": "keep-me",
+ },
+ )
+ handler = _DummyOpenAIHandler()
+
+ monkeypatch.setattr("headroom.tokenizers.get_tokenizer", lambda model: _DummyTokenizer())
+
+ response = anyio.run(handler.handle_openai_responses, request)
+
+ assert response.status_code == 200
+ assert handler.captured_request is not None
+ _method, _url, headers, _body = handler.captured_request
+ lowered = {k.lower(): v for k, v in headers.items()}
+ assert "x-openai-internal-codex-responses-lite" not in lowered
+ assert lowered.get("x-openai-debug") == "keep-me"
+
+
def test_handle_openai_responses_chatgpt_codex_timeout_fails_open(monkeypatch):
token = _jwt(
{