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.

Rename tool output compression parallelism env

jamesx0416cfcaf0e364e8
diff --git a/headroom/proxy/handlers/openai.py b/headroom/proxy/handlers/openai.py
index dcfe6f8a1..9835ad6a1 100644
--- a/headroom/proxy/handlers/openai.py
+++ b/headroom/proxy/handlers/openai.py
@@ -52,7 +52,7 @@ logger = logging.getLogger("headroom.proxy")
_OPENAI_RESPONSES_UNIT_CACHE_MAX_ENTRIES = 10_000
_OPENAI_RESPONSES_UNIT_CACHE_VERSION = "openai_responses_unit_v1"
-_OPENAI_RESPONSES_UNIT_PARALLELISM_ENV = "HEADROOM_OPENAI_RESPONSES_UNIT_PARALLELISM"
+_OPENAI_RESPONSES_UNIT_PARALLELISM_ENV = "HEADROOM_TOOL_OUTPUT_COMPRESSION_PARALLELISM"
_OPENAI_RESPONSES_UNIT_PARALLELISM_DEFAULT = 4
_OPENAI_RESPONSES_UNIT_PARALLELISM_MAX = 16
_OPENAI_RESPONSES_UNIT_CACHE_INIT_LOCK = threading.RLock()
diff --git a/tests/test_openai_responses_compression_units.py b/tests/test_openai_responses_compression_units.py
index c0b8329ac..aaa8b5212 100644
--- a/tests/test_openai_responses_compression_units.py
+++ b/tests/test_openai_responses_compression_units.py
@@ -28,16 +28,16 @@ def _handler_with_router(router: ContentRouter) -> OpenAIHandlerMixin:
def test_openai_responses_unit_parallelism_env_defaults_and_clamps(monkeypatch):
- monkeypatch.delenv("HEADROOM_OPENAI_RESPONSES_UNIT_PARALLELISM", raising=False)
+ monkeypatch.delenv("HEADROOM_TOOL_OUTPUT_COMPRESSION_PARALLELISM", raising=False)
assert openai_handler._openai_responses_unit_parallelism() == 4
- monkeypatch.setenv("HEADROOM_OPENAI_RESPONSES_UNIT_PARALLELISM", "bad")
+ monkeypatch.setenv("HEADROOM_TOOL_OUTPUT_COMPRESSION_PARALLELISM", "bad")
assert openai_handler._openai_responses_unit_parallelism() == 4
- monkeypatch.setenv("HEADROOM_OPENAI_RESPONSES_UNIT_PARALLELISM", "0")
+ monkeypatch.setenv("HEADROOM_TOOL_OUTPUT_COMPRESSION_PARALLELISM", "0")
assert openai_handler._openai_responses_unit_parallelism() == 1
- monkeypatch.setenv("HEADROOM_OPENAI_RESPONSES_UNIT_PARALLELISM", "999")
+ monkeypatch.setenv("HEADROOM_TOOL_OUTPUT_COMPRESSION_PARALLELISM", "999")
assert openai_handler._openai_responses_unit_parallelism() == 16
@@ -271,7 +271,7 @@ def test_openai_responses_adapter_reuses_identical_tool_output_in_same_request()
def test_openai_responses_adapter_parallelizes_cache_misses_preserving_order(monkeypatch):
- monkeypatch.setenv("HEADROOM_OPENAI_RESPONSES_UNIT_PARALLELISM", "4")
+ monkeypatch.setenv("HEADROOM_TOOL_OUTPUT_COMPRESSION_PARALLELISM", "4")
router = ContentRouter()
lock = threading.Lock()
release = threading.Event()
diff --git a/tests/test_openai_responses_t3_replay_regression.py b/tests/test_openai_responses_t3_replay_regression.py
index 494062536..a4374ee54 100644
--- a/tests/test_openai_responses_t3_replay_regression.py
+++ b/tests/test_openai_responses_t3_replay_regression.py
@@ -96,7 +96,7 @@ def _json_bytes(value: object) -> int:
def test_t3_failed_size_responses_payload_parallelizes_uncached_tool_outputs(monkeypatch):
- monkeypatch.setenv("HEADROOM_OPENAI_RESPONSES_UNIT_PARALLELISM", "4")
+ monkeypatch.setenv("HEADROOM_TOOL_OUTPUT_COMPRESSION_PARALLELISM", "4")
case = T3_FAILED_CASES[0]
router = ContentRouter()
lock = threading.Lock()