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.

fix(proxy): hoist compression_policy outside is_token_mode branch (F2.1 c5 followup)

chopratejas708b87a19fbe
diff --git a/headroom/proxy/handlers/anthropic.py b/headroom/proxy/handlers/anthropic.py
index 45251393b..b1c2b5570 100644
--- a/headroom/proxy/handlers/anthropic.py
+++ b/headroom/proxy/handlers/anthropic.py
@@ -846,6 +846,18 @@ class AnthropicHandlerMixin:
else None
)
+ # F2.1 c5/5: derive the per-request CompressionPolicy
+ # from the auth_mode classified at request entry. The
+ # policy short-circuits CacheAligner for subscription
+ # users (closes the cache-instability complaints in
+ # #327/#388). When the enforcement env var is off, the
+ # policy collapses to PAYG so behaviour is unchanged.
+ # Hoisted here so all three pipeline.apply call sites
+ # (token / non-cache / cache-delta) see the same policy.
+ from headroom.transforms.compression_policy import resolve_policy
+
+ compression_policy = resolve_policy(getattr(request.state, "auth_mode", None))
+
if is_token_mode(self.config.mode):
comp_cache = self._get_compression_cache(session_id)
@@ -880,17 +892,6 @@ class AnthropicHandlerMixin:
# Record all tool_results in the verified frozen prefix as stable
comp_cache.mark_stable_from_messages(messages, frozen_message_count)
- # F2.1 c5/5: derive the per-request CompressionPolicy
- # from the auth_mode classified at request entry. The
- # policy short-circuits CacheAligner for subscription
- # users (closes the cache-instability complaints in
- # #327/#388). When the enforcement env var is off, the
- # policy collapses to PAYG so behaviour is unchanged.
- from headroom.transforms.compression_policy import resolve_policy
-
- compression_policy = resolve_policy(
- getattr(request.state, "auth_mode", None)
- )
async with stage_timer.measure("compression_first_stage"):
result = await self._run_compression_in_executor(
lambda: self.anthropic_pipeline.apply(
@@ -941,6 +942,7 @@ class AnthropicHandlerMixin:
frozen_message_count=frozen_message_count,
biases=biases,
request_id=request_id,
+ compression_policy=compression_policy,
),
timeout=COMPRESSION_TIMEOUT_SECONDS,
)
@@ -971,6 +973,7 @@ class AnthropicHandlerMixin:
frozen_message_count=0,
biases=biases,
request_id=request_id,
+ compression_policy=compression_policy,
),
timeout=COMPRESSION_TIMEOUT_SECONDS,
)
diff --git a/headroom/proxy/handlers/openai.py b/headroom/proxy/handlers/openai.py
index 2b2805b5b..ab77b9913 100644
--- a/headroom/proxy/handlers/openai.py
+++ b/headroom/proxy/handlers/openai.py
@@ -422,6 +422,14 @@ class OpenAIHandlerMixin:
try:
context_limit = self.openai_provider.get_context_limit(model)
+ # F2.1 c5/5: per-request CompressionPolicy. Hoisted out of
+ # the is_token_mode branch so the else (non-token) branch
+ # below can pass it through too. See the equivalent block
+ # in handlers/anthropic.py.
+ from headroom.transforms.compression_policy import resolve_policy
+
+ compression_policy = resolve_policy(getattr(request.state, "auth_mode", None))
+
if is_token_mode(self.config.mode):
comp_cache = self._get_compression_cache(openai_session_id)
@@ -431,11 +439,6 @@ class OpenAIHandlerMixin:
# Re-freeze boundary
openai_frozen_count = comp_cache.compute_frozen_count(messages)
- # F2.1 c5/5: per-request CompressionPolicy. See the
- # equivalent block in handlers/anthropic.py.
- from headroom.transforms.compression_policy import resolve_policy
-
- compression_policy = resolve_policy(getattr(request.state, "auth_mode", None))
result = await self._run_compression_in_executor(
lambda: self.openai_pipeline.apply(
messages=working_messages,