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): preserve Unicode separators in SSE JSON (#4337)
lorenzozane3afaa97ffc06
diff --git a/headroom/proxy/memory_tool_stream.py b/headroom/proxy/memory_tool_stream.pyindex 7f6a791bf..35410ae60 100644--- a/headroom/proxy/memory_tool_stream.py+++ b/headroom/proxy/memory_tool_stream.py@@ -57,7 +57,8 @@ def _parse_frame(frame: bytes) -> tuple[str | None, dict[str, Any] | None]:"""Return ``(event_name, payload)`` for one complete SSE frame."""event_name: str | None = Nonedata_lines: list[str] = []- for line in frame.decode("utf-8").splitlines():+ for raw_line in frame.splitlines():+ line = raw_line.decode("utf-8")if line.startswith(SSE_EVENT_LINE_PREFIX):event_name = line[len(SSE_EVENT_LINE_PREFIX) :].strip()elif line.startswith(SSE_DATA_LINE_PREFIX):diff --git a/headroom/proxy/sse_byte_buffer_policy.py b/headroom/proxy/sse_byte_buffer_policy.pyindex 42f5d9fcf..b043fd347 100644--- a/headroom/proxy/sse_byte_buffer_policy.py+++ b/headroom/proxy/sse_byte_buffer_policy.py@@ -48,10 +48,10 @@ def parse_sse_events_from_byte_buffer(event_bytes = bytes(buf[:idx])del buf[: idx + terminator_len]- event_text = event_bytes.decode("utf-8")event_name: str | None = Nonedata_lines: list[str] = []- for line in event_text.splitlines():+ for raw_line in event_bytes.splitlines():+ line = raw_line.decode("utf-8")if not line:continueif line.startswith(":"):diff --git a/tests/test_memory_tool_stream.py b/tests/test_memory_tool_stream.pyindex 93af46805..51e070951 100644--- a/tests/test_memory_tool_stream.py+++ b/tests/test_memory_tool_stream.py@@ -143,6 +143,29 @@ class TestMemoryToolStreamFilter:assert forwarded[0]["type"] == "content_block_start"assert {e["index"] for e in forwarded if "index" in e} == {1}+ @pytest.mark.parametrize("separator", ["\u0085", "\u2028", "\u2029"])+ def test_continuation_reindexes_text_with_unicode_line_separator(self, separator: str) -> None:+ flt = MemoryToolStreamFilter(MEMORY_TOOLS, index_offset=2, forward_message_start=False)+ start = {+ "type": "content_block_start",+ "index": 0,+ "content_block": {"type": "text", "text": ""},+ }+ delta = {+ "type": "content_block_delta",+ "index": 0,+ "delta": {"type": "text_delta", "text": f"before{separator}after"},+ }+ raw = b"".join(+ f"event: {payload['type']}\r\ndata: {json.dumps(payload, ensure_ascii=False)}\r\n\r\n".encode()+ for payload in (start, delta)+ )++ forwarded = _events(b"".join(flt.feed(raw)))++ assert [event["index"] for event in forwarded] == [2, 2]+ assert forwarded[1]["delta"]["text"] == f"before{separator}after"+def test_client_tool_alongside_memory_call_keeps_tool_use_stop(self) -> None:flt = MemoryToolStreamFilter(MEMORY_TOOLS)forwarded = _events(b"".join(flt.feed(_sse([SAVE, BASH], "tool_use"))))diff --git a/tests/test_sse_byte_buffer_policy.py b/tests/test_sse_byte_buffer_policy.pyindex 839cab622..716fa9aec 100644--- a/tests/test_sse_byte_buffer_policy.py+++ b/tests/test_sse_byte_buffer_policy.py@@ -1,5 +1,7 @@from __future__ import annotations+import json+import pytestfrom headroom.proxy.sse_byte_buffer_policy import (@@ -28,11 +30,31 @@ def test_parse_sse_events_preserves_split_utf8_tail() -> None:assert parse_sse_events_from_byte_buffer(buf) == [(None, "hello \U0001f642")]+def test_parse_sse_events_splits_crlf_lines() -> None:+ buf = bytearray(b"event: delta\r\ndata: one\r\ndata: two\r\n\r\n")++ assert parse_sse_events_from_byte_buffer(buf) == [("delta", "one\ntwo")]+ assert not buf++def test_parse_sse_events_raises_on_complete_invalid_utf8_event() -> None:with pytest.raises(UnicodeDecodeError):parse_sse_events_from_byte_buffer(bytearray(b"data: \xff\n\n"))[email protected]("separator", ["\u0085", "\u2028", "\u2029"])+def test_parse_sse_events_preserves_unicode_line_separators_in_json(+ separator: str,+) -> None:+ payload = {"text": f"before{separator}after"}+ wire = f"data: {json.dumps(payload, ensure_ascii=False)}\n\n".encode()++ events = parse_sse_events_from_byte_buffer(bytearray(wire))++ assert len(events) == 1+ assert json.loads(events[0][1]) == payload++def test_parse_sse_events_strips_only_one_leading_space() -> None:# SSE spec: "If value starts with a U+0020 SPACE character, remove it".# Further leading whitespace is part of the payload.