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: clear CI mypy + rust test failures introduced in eaf5980
Tejas Chopra17ffae0cd8e4
diff --git a/crates/headroom-core/tests/live_zone_thresholds.rs b/crates/headroom-core/tests/live_zone_thresholds.rsindex 946cb54a1..429db3f29 100644--- a/crates/headroom-core/tests/live_zone_thresholds.rs+++ b/crates/headroom-core/tests/live_zone_thresholds.rs@@ -51,14 +51,14 @@ fn first_tool_result_action(out: &LiveZoneOutcome) -> BlockAction {#[test]fn below_threshold_no_compression_attempted() {- // 200 bytes of homogeneous JSON dicts — well below the 1 KiB+ // 200 bytes of homogeneous JSON dicts — well below the 512 B// JsonArray threshold. The dispatcher must record// `BelowByteThreshold` and emit `NoChange`; no compressor runs.let small_array: Vec<Value> = (0..3).map(|i| json!({"id": i, "v": "x"})).collect();let payload = serde_json::to_string(&small_array).unwrap();assert!(- payload.len() < 1024,- "fixture must stay below the 1 KiB JsonArray threshold; got {}",+ payload.len() < 512,+ "fixture must stay below the 512 B JsonArray threshold; got {}",payload.len());@@ -80,7 +80,7 @@ fn below_threshold_no_compression_attempted() {} => {assert_eq!(content_type, "json_array");assert_eq!(byte_count, payload.len());- assert_eq!(threshold_bytes, 1024);+ assert_eq!(threshold_bytes, 512);}other => panic!("expected BelowByteThreshold for sub-threshold JSON, got {other:?}"),}diff --git a/headroom/transforms/compression_units.py b/headroom/transforms/compression_units.pyindex 16a424cd5..7253fbcf1 100644--- a/headroom/transforms/compression_units.py+++ b/headroom/transforms/compression_units.py@@ -9,7 +9,7 @@ replacements back into their native request shape.from __future__ import annotationsfrom collections.abc import Iterable-from dataclasses import dataclass, field+from dataclasses import dataclass, field, replacefrom typing import Protocolfrom .content_router import CompressionStrategy, ContentRouter, RouterCompressionResult@@ -63,11 +63,9 @@ def find_content_router(transforms: object) -> ContentRouter | None:"""Return the first ContentRouter in a pipeline or iterable."""candidates = getattr(transforms, "transforms", transforms)- try:- iterator = iter(candidates) # type: ignore[arg-type]- except TypeError:+ if not isinstance(candidates, Iterable):return None- for transform in iterator:+ for transform in candidates:if isinstance(transform, ContentRouter):return transformreturn None@@ -86,32 +84,32 @@ def compress_unit_with_router("""tokens_before = tokenizer.count_text(unit.text)- base = {- "original": unit.text,- "compressed": unit.text,- "modified": False,- "tokens_before": tokens_before,- "tokens_after": tokens_before,- "tokens_saved": 0,- "transforms_applied": [],- "strategy": CompressionStrategy.PASSTHROUGH.value,- "router_result": None,- }+ base = UnitCompressionResult(+ original=unit.text,+ compressed=unit.text,+ modified=False,+ tokens_before=tokens_before,+ tokens_after=tokens_before,+ tokens_saved=0,+ transforms_applied=[],+ strategy=CompressionStrategy.PASSTHROUGH.value,+ router_result=None,+ )if not unit.mutable:- return UnitCompressionResult(**base, reason="immutable")+ return replace(base, reason="immutable")if unit.role == "user":- return UnitCompressionResult(**base, reason="protected_user_message")+ return replace(base, reason="protected_user_message")if unit.role in {"system", "developer"}:- return UnitCompressionResult(**base, reason="protected_system_message")+ return replace(base, reason="protected_system_message")if unit.role == "assistant" and unit.metadata.get("compress_assistant") != "true":- return UnitCompressionResult(**base, reason="protected_assistant_message")+ return replace(base, reason="protected_assistant_message")if unit.cache_zone != "live":- return UnitCompressionResult(**base, reason=f"cache_zone_{unit.cache_zone}")+ return replace(base, reason=f"cache_zone_{unit.cache_zone}")if len(unit.text) < unit.min_bytes:- return UnitCompressionResult(**base, reason="below_unit_floor")+ return replace(base, reason="below_unit_floor")if "Retrieve more: hash=" in unit.text or "Retrieve original: hash=" in unit.text:- return UnitCompressionResult(**base, reason="already_compressed")+ return replace(base, reason="already_compressed")router_result = router.compress(unit.text,@@ -122,21 +120,21 @@ def compress_unit_with_router(replacement = router_result.compressedstrategy = router_result.strategy_used.valueif replacement == unit.text:- return UnitCompressionResult(- **{**base, "strategy": strategy, "router_result": router_result},+ return replace(+ base,+ strategy=strategy,+ router_result=router_result,reason="router_no_change",)tokens_after = tokenizer.count_text(replacement)if tokens_after >= tokens_before:- return UnitCompressionResult(- **{- **base,- "compressed": replacement,- "tokens_after": tokens_after,- "strategy": strategy,- "router_result": router_result,- },+ return replace(+ base,+ compressed=replacement,+ tokens_after=tokens_after,+ strategy=strategy,+ router_result=router_result,reason="rejected_not_smaller",)