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.

Merge origin/main into fix/codex-ws-dashboard-performance

Gili Tzabari28a0e19deba8
diff --cc headroom/proxy/handlers/openai.py
index 42f63d04e,fe12b7d9d..484fd68f7
--- a/headroom/proxy/handlers/openai.py
+++ b/headroom/proxy/handlers/openai.py
@@@ -3300,26 -3264,20 +3320,26 @@@ class OpenAIHandlerMixin
_inner = _send_body["response"] if _wrapped else _send_body
_model = (_inner.get("model") if isinstance(_inner, dict) else None) or ""
- (
- _new_inner,
- _modified,
- _ws_saved,
- _ws_transforms,
- _ws_reason,
- _bytes_before,
- _bytes_after,
- _ws_attempted_tokens,
- ) = await self._compress_openai_responses_payload_in_executor(
- _inner,
- model=_model,
- request_id=request_id,
- )
+ _compression_started = time.perf_counter()
+ try:
+ (
+ _new_inner,
+ _modified,
+ _ws_saved,
+ _ws_transforms,
+ _ws_reason,
+ _bytes_before,
+ _bytes_after,
+ _ws_attempted_tokens,
- ) = self._compress_openai_responses_payload(
++ ) = await self._compress_openai_responses_payload_in_executor(
+ _inner,
+ model=_model,
+ request_id=request_id,
+ )
+ finally:
+ _record_ws_compression_overhead(
+ (time.perf_counter() - _compression_started) * 1000.0
+ )
if _modified:
if isinstance(_new_inner, dict):
if _wrapped:
@@@ -3530,26 -3488,20 +3550,26 @@@
try:
model_for_frame = inner_payload.get("model") or ""
_frame_auth_mode = classify_auth_mode(ws_headers)
- (
- new_inner,
- modified,
- frame_saved,
- frame_transforms,
- frame_reason,
- bytes_before,
- bytes_after,
- frame_attempted_tokens,
- ) = await self._compress_openai_responses_payload_in_executor(
- inner_payload,
- model=model_for_frame,
- request_id=request_id,
- )
+ _compression_started = time.perf_counter()
+ try:
+ (
+ new_inner,
+ modified,
+ frame_saved,
+ frame_transforms,
+ frame_reason,
+ bytes_before,
+ bytes_after,
+ frame_attempted_tokens,
- ) = self._compress_openai_responses_payload(
++ ) = await self._compress_openai_responses_payload_in_executor(
+ inner_payload,
+ model=model_for_frame,
+ request_id=request_id,
+ )
+ finally:
+ _record_ws_compression_overhead(
+ (time.perf_counter() - _compression_started) * 1000.0
+ )
except Exception as _frame_err:
logger.warning(
"[%s] WS /v1/responses frame compression "
diff --cc tests/test_openai_codex_ws_lifecycle.py
index 80b872371,f62e68078..17916a878
--- a/tests/test_openai_codex_ws_lifecycle.py
+++ b/tests/test_openai_codex_ws_lifecycle.py
@@@ -15,7 -15,8 +15,8 @@@ from unittest.mock import MagicMock, pa
import pytest
-from headroom.proxy.helpers import COMPRESSION_TIMEOUT_SECONDS
from headroom.proxy.handlers.openai import OpenAIHandlerMixin
++from headroom.proxy.helpers import COMPRESSION_TIMEOUT_SECONDS
from headroom.proxy.ws_session_registry import WebSocketSessionRegistry
# ---------------------------------------------------------------------------