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
"""compress() must not report a previous request's result for a request it did not touch (#4340).
The per-worker ImageCompressor is a singleton (proxy/image_isolation.py reuses one
instance across requests) and _compress_messages_worker reports last_result whenever
it is not None. Before the fix, compress() left last_result set from the previous call
on its early-return paths, so a request whose images were never touched inherited the
previous request's technique and savings and was marked image_compression-mutated.
"""
from __future__ import annotations
from headroom.image.compressor import CompressionResult, ImageCompressor, Technique
def _stale() -> CompressionResult:
return CompressionResult(
technique=Technique.PRESERVE,
original_tokens=765,
compressed_tokens=2295,
confidence=1.0,
)
def test_compress_clears_last_result_when_there_are_no_images() -> None:
compressor = ImageCompressor()
compressor.last_result = _stale()
messages = [{"role": "user", "content": "plain text, no images"}]
out = compressor.compress(messages)
assert out == messages
assert compressor.last_result is None
def test_compress_clears_last_result_for_url_only_images() -> None:
# A URL image (no inline base64 data) reaches the "no image_data" early return;
# it must not report the previous request's result either.
compressor = ImageCompressor()
compressor.last_result = _stale()
messages = [
{
"role": "user",
"content": [
{"type": "text", "text": "describe this"},
{"type": "image_url", "image_url": {"url": "https://example.test/a.png"}},
],
}
]
compressor.compress(messages)
assert compressor.last_result is None