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 mypy errors and add network timeout handler for flaky CI tests

chopratejas52da66297978
diff --git a/headroom/compression/universal.py b/headroom/compression/universal.py
index 2025b25bc..73203b2b9 100644
--- a/headroom/compression/universal.py
+++ b/headroom/compression/universal.py
@@ -375,7 +375,7 @@ class UniversalCompressor:
original_tokens=self._estimate_tokens(original),
compressed_tokens=self._estimate_tokens(compressed),
)
- return key
+ return str(key) if key else None
except ImportError:
logger.debug("CCR store not available")
return None
diff --git a/headroom/evals/memory/runner_v2.py b/headroom/evals/memory/runner_v2.py
index 31fe715a9..a25a72b80 100644
--- a/headroom/evals/memory/runner_v2.py
+++ b/headroom/evals/memory/runner_v2.py
@@ -403,8 +403,8 @@ class LoCoMoEvaluatorV2:
finally:
# Clean up backend
- if self._backend:
- await self._backend.close()
+ if self._backend and hasattr(self._backend, "close"):
+ await self._backend.close() # type: ignore[union-attr]
# Calculate final metrics
duration = time.time() - start_time
diff --git a/headroom/evals/memory/runner_v3.py b/headroom/evals/memory/runner_v3.py
index 982687a63..c066b5a06 100644
--- a/headroom/evals/memory/runner_v3.py
+++ b/headroom/evals/memory/runner_v3.py
@@ -277,8 +277,8 @@ class LoCoMoEvaluatorV3:
all_results.append(result)
finally:
- if self._backend:
- await self._backend.close()
+ if self._backend and hasattr(self._backend, "close"):
+ await self._backend.close() # type: ignore[union-attr]
# Aggregate results
duration = time.time() - start_time
diff --git a/headroom/image/compressor.py b/headroom/image/compressor.py
index 5f1c9f0a9..9e4fa590a 100644
--- a/headroom/image/compressor.py
+++ b/headroom/image/compressor.py
@@ -241,7 +241,7 @@ class ImageCompressor:
# High detail: 85 tokens per 512x512 tile + 170 base
tiles_x = (width + 511) // 512
tiles_y = (height + 511) // 512
- return 85 * tiles_x * tiles_y + 170
+ return int(85 * tiles_x * tiles_y + 170)
def _apply_compression(
self,
diff --git a/headroom/integrations/langchain/agents.py b/headroom/integrations/langchain/agents.py
index 409e5999f..686e358a2 100644
--- a/headroom/integrations/langchain/agents.py
+++ b/headroom/integrations/langchain/agents.py
@@ -207,7 +207,7 @@ class HeadroomToolWrapper:
# Check if compression is needed
if len(result) < self.min_chars_to_compress:
self._record_metrics(result, result, was_compressed=False)
- return result
+ return str(result)
# Try to compress
compressed = self._compress_output(result)
diff --git a/headroom/memory/adapters/sqlite.py b/headroom/memory/adapters/sqlite.py
index 3442a699a..073538498 100644
--- a/headroom/memory/adapters/sqlite.py
+++ b/headroom/memory/adapters/sqlite.py
@@ -143,7 +143,7 @@ class SQLiteMemoryStore:
"""Serialize numpy array to bytes for BLOB storage."""
if embedding is None:
return None
- return embedding.astype(np.float32).tobytes()
+ return bytes(embedding.astype(np.float32).tobytes())
def _deserialize_embedding(
self, data: bytes | None, dim: int | None = None
diff --git a/headroom/memory/backends/direct_mem0.py b/headroom/memory/backends/direct_mem0.py
index 248ae9783..4c20b1537 100644
--- a/headroom/memory/backends/direct_mem0.py
+++ b/headroom/memory/backends/direct_mem0.py
@@ -232,7 +232,7 @@ class DirectMem0Adapter:
input=text,
model=self._config.embedder_model,
)
- return response.data[0].embedding
+ return list(response.data[0].embedding)
def _generate_id(self, content: str, user_id: str) -> str:
"""Generate a deterministic ID for a memory."""
diff --git a/headroom/pricing/litellm_pricing.py b/headroom/pricing/litellm_pricing.py
index 0ab536e31..f8d95f16c 100644
--- a/headroom/pricing/litellm_pricing.py
+++ b/headroom/pricing/litellm_pricing.py
@@ -37,7 +37,7 @@ def get_litellm_model_cost() -> dict[str, Any]:
Returns:
Dictionary mapping model names to their pricing/capability info.
"""
- return litellm.model_cost
+ return litellm.model_cost # type: ignore[no-any-return]
def get_model_pricing(model: str) -> LiteLLMModelPricing | None:
diff --git a/headroom/providers/anthropic.py b/headroom/providers/anthropic.py
index 6da9f4f53..85df9961d 100644
--- a/headroom/providers/anthropic.py
+++ b/headroom/providers/anthropic.py
@@ -579,7 +579,7 @@ class AnthropicProvider(Provider):
"cached_input", pricing["input"]
)
cost += cached_cost
- return cost
+ return cost # type: ignore[no-any-return]
except Exception as e:
logger.debug(f"LiteLLM cost estimation failed for {model}: {e}")
@@ -596,7 +596,7 @@ class AnthropicProvider(Provider):
+ (output_tokens / 1_000_000) * pricing["output"]
)
- return cost
+ return cost # type: ignore[no-any-return]
def _get_pricing(self, model: str) -> dict[str, float] | None:
"""Get pricing for a model with fallback logic."""
diff --git a/headroom/providers/cohere.py b/headroom/providers/cohere.py
index 043efa5a3..2b205a0a2 100644
--- a/headroom/providers/cohere.py
+++ b/headroom/providers/cohere.py
@@ -271,11 +271,11 @@ class CohereProvider(Provider):
if info and "max_input_tokens" in info:
result = info["max_input_tokens"]
if result is not None:
- return result
+ return int(result)
if info and "max_tokens" in info:
result = info["max_tokens"]
if result is not None:
- return result
+ return int(result)
except Exception:
pass
@@ -335,7 +335,7 @@ class CohereProvider(Provider):
completion_tokens=output_tokens,
)
if cost is not None:
- return cost
+ return float(cost)
except Exception:
pass
diff --git a/headroom/proxy/server.py b/headroom/proxy/server.py
index dacec795f..33aa53d65 100644
--- a/headroom/proxy/server.py
+++ b/headroom/proxy/server.py
@@ -4035,7 +4035,7 @@ class HeadroomProxy:
try:
response = await self.http_client.get(url, headers=headers) # type: ignore[union-attr]
if response.status_code == 200:
- return response.text
+ return str(response.text)
logger.error(f"Failed to download file {file_id}: {response.status_code}")
return None
except Exception as e:
diff --git a/tests/test_memory/test_core_operations.py b/tests/test_memory/test_core_operations.py
index cff3eae21..9eff5dbe2 100644
--- a/tests/test_memory/test_core_operations.py
+++ b/tests/test_memory/test_core_operations.py
@@ -9,6 +9,9 @@ Tests cover:
- History chain traversal
- Deletion operations
- Convenience methods (remember, recall, get_user_memories, get_session_memories)
+
+Note: These are integration tests that may hit external embedding APIs.
+Tests are marked to skip on network timeouts (flaky CI).
"""
# CRITICAL: Must set TOKENIZERS_PARALLELISM before any imports
@@ -16,9 +19,11 @@ import os
os.environ["TOKENIZERS_PARALLELISM"] = "false"
+import functools
import tempfile
from pathlib import Path
+import httpx
import pytest
from headroom.memory.config import MemoryConfig
@@ -26,6 +31,20 @@ from headroom.memory.core import HierarchicalMemory
from headroom.memory.models import Memory, ScopeLevel
from headroom.memory.ports import MemoryFilter
+
+def network_timeout_handler(func):
+ """Decorator to skip tests on network timeouts (flaky CI)."""
+
+ @functools.wraps(func)
+ async def wrapper(*args, **kwargs):
+ try:
+ return await func(*args, **kwargs)
+ except httpx.ReadTimeout:
+ pytest.skip("Skipped due to network timeout (flaky CI)")
+
+ return wrapper
+
+
# =============================================================================
# Fixtures
# =============================================================================
@@ -60,6 +79,7 @@ class TestAddBatch:
"""Tests for HierarchicalMemory.add_batch()."""
@pytest.mark.asyncio
+ @network_timeout_handler
async def test_add_batch_basic(self, memory_system):
"""Test basic batch addition."""
memories_data = [
diff --git a/tests/test_memory/test_easy.py b/tests/test_memory/test_easy.py
index 48cf71cf1..824deb5be 100644
--- a/tests/test_memory/test_easy.py
+++ b/tests/test_memory/test_easy.py
@@ -7,6 +7,9 @@ Tests cover:
- Error handling and edge cases
- Backend type switching
- Resource cleanup
+
+Note: These are integration tests that may hit external embedding APIs.
+Tests are marked to skip on network timeouts (flaky CI).
"""
# CRITICAL: Must set TOKENIZERS_PARALLELISM before any imports that might
@@ -18,10 +21,26 @@ os.environ["TOKENIZERS_PARALLELISM"] = "false"
import tempfile
from pathlib import Path
+import httpx
import pytest
from headroom.memory.easy import Memory, MemoryResult
+
+def network_timeout_handler(func):
+ """Decorator to skip tests on network timeouts (flaky CI)."""
+ import functools
+
+ @functools.wraps(func)
+ async def wrapper(*args, **kwargs):
+ try:
+ return await func(*args, **kwargs)
+ except httpx.ReadTimeout:
+ pytest.skip("Skipped due to network timeout (flaky CI)")
+
+ return wrapper
+
+
# =============================================================================
# Fixtures
# =============================================================================
@@ -165,6 +184,7 @@ class TestMemorySave:
assert memory_id is not None
@pytest.mark.asyncio
+ @network_timeout_handler
async def test_save_with_entities(self, memory_instance):
"""Test saving with pre-extracted entities."""
memory_id = await memory_instance.save(
@@ -455,6 +475,7 @@ class TestMemoryClose:
assert mem._initialized is False
@pytest.mark.asyncio
+ @network_timeout_handler
async def test_close_idempotent(self, temp_db_path):
"""Test that close can be called multiple times."""
mem = Memory(backend="local", db_path=temp_db_path)