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.
Python agentaianthropicclaude-codecompressioncontext-engineeringcontext-windowcursorfastapilangchainllmmcpopenaiprompt-engineeringproxypythonragtoken-optimizationtokenstypescript
fix(proxy): strip duplicated upstream server headers (#1828)
Rod Boevd2a86b590978
diff --git a/headroom/providers/proxy_routes.py b/headroom/providers/proxy_routes.pyindex 80abd2544..c81b76a4b 100644--- a/headroom/providers/proxy_routes.py+++ b/headroom/providers/proxy_routes.py@@ -14,6 +14,7 @@ from fastapi.responses import Responsefrom headroom.proxy.handlers.openai import (_custom_base_passthrough_telemetry,_resolve_codex_routing_headers,+ _sanitize_forwarded_response_headers,)logger = logging.getLogger("headroom.proxy.routes")@@ -447,9 +448,7 @@ async def _handle_chatgpt_codex_images(content=body,timeout=120.0,)- response_headers = dict(resp.headers)- response_headers.pop("content-encoding", None)- response_headers.pop("content-length", None)+ response_headers = _sanitize_forwarded_response_headers(resp.headers)return Response(content=resp.content,status_code=resp.status_code,diff --git a/headroom/proxy/handlers/openai.py b/headroom/proxy/handlers/openai.pyindex 7f33dc672..dbff4e627 100644--- a/headroom/proxy/handlers/openai.py+++ b/headroom/proxy/handlers/openai.py@@ -116,6 +116,16 @@ def _header_get(headers: dict[str, str], name: str) -> str | None:return None+def _sanitize_forwarded_response_headers(+ headers: httpx.Headers | dict[str, str],+ *extra_names: str,+) -> dict[str, str]:+ cleaned = dict(headers)+ for name in ("content-encoding", "content-length", "server", *extra_names):+ cleaned.pop(name, None)+ return cleaned++def _custom_base_passthrough_telemetry(method: str, path: str, base_url: str) -> tuple[str, str]:"""Return passthrough telemetry metadata for narrow custom-base exceptions."""# OpenCode Zen sends provider-prefixed OpenAI-compatible traffic through@@ -2285,9 +2295,7 @@ class OpenAIHandlerMixin:)# Remove compression headers from cached response- response_headers = dict(cached.response_headers)- response_headers.pop("content-encoding", None)- response_headers.pop("content-length", None)+ response_headers = _sanitize_forwarded_response_headers(cached.response_headers)return Response(content=cached.response_body, headers=response_headers)@@ -3478,9 +3486,7 @@ class OpenAIHandlerMixin:)# Remove compression headers since httpx already decompressed the response- response_headers = dict(response.headers)- response_headers.pop("content-encoding", None)- response_headers.pop("content-length", None) # Length changed after decompression+ response_headers = _sanitize_forwarded_response_headers(response.headers)# Inject Headroom compression metrics (for SaaS metering)response_headers["x-headroom-tokens-before"] = str(original_tokens)@@ -4481,9 +4487,7 @@ class OpenAIHandlerMixin:get_codex_rate_limit_state().update_from_headers(dict(response.headers))# Remove compression headers- response_headers = dict(response.headers)- response_headers.pop("content-encoding", None)- response_headers.pop("content-length", None)+ response_headers = _sanitize_forwarded_response_headers(response.headers)if buffered_stream_ccr and response.status_code == 200 and resp_json:sse_headers = {@@ -7378,9 +7382,7 @@ class OpenAIHandlerMixin:)# Remove compression headers since httpx already decompressed the response- response_headers = dict(response.headers)- response_headers.pop("content-encoding", None)- response_headers.pop("content-length", None) # Length changed after decompression+ response_headers = _sanitize_forwarded_response_headers(response.headers)response_content = response.contentif provider == "anthropic" and endpoint_name == "models":@@ -7524,11 +7526,11 @@ class OpenAIHandlerMixin:media_type="application/json",)- response_headers = dict(upstream_response.headers)- response_headers.pop("content-length", None)- response_headers.pop("transfer-encoding", None)- response_headers.pop("connection", None)- response_headers.pop("content-encoding", None)+ response_headers = _sanitize_forwarded_response_headers(+ upstream_response.headers,+ "transfer-encoding",+ "connection",+ )if upstream_response.status_code >= 400:try:diff --git a/tests/test_provider_proxy_routes.py b/tests/test_provider_proxy_routes.pyindex be588a3c1..0c10124a9 100644--- a/tests/test_provider_proxy_routes.py+++ b/tests/test_provider_proxy_routes.py@@ -644,6 +644,7 @@ def test_openai_image_codex_response_strips_stale_compression_headers(monkeypatc"content-encoding": "gzip","content-length": stale_content_length,"content-type": "application/json",+ "server": "upstream-edge","x-upstream": "kept",},)@@ -671,6 +672,7 @@ def test_openai_image_codex_response_strips_stale_compression_headers(monkeypatcassert response.status_code == 200assert response.content == upstream_bodyassert response.headers["x-upstream"] == "kept"+ assert response.headers.get("server") is Noneassert response.headers.get("content-encoding") is Noneassert response.headers.get("content-length") == str(len(upstream_body))