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
refactor(memory,learn,cache): remove unreferenced classes, methods and constants (#4112)
Tejas Chopraabbf36f052fc
diff --git a/headroom/cache/compression_feedback.py b/headroom/cache/compression_feedback.pyindex dbb8d5bd8..9032d3886 100644--- a/headroom/cache/compression_feedback.py+++ b/headroom/cache/compression_feedback.py@@ -344,13 +344,6 @@ class CompressionFeedback:# Extract potential field names from queryself._extract_field_hints(pattern, event.query)- def _truncate_strategy_dicts(self, pattern: LocalToolPattern) -> None:- """Truncate strategy counters using the shared strategy outcome domain."""- outcomes = pattern.strategy_outcomes- outcomes.prune()- pattern.strategy_compressions = outcomes.compressions- pattern.strategy_retrievals = outcomes.retrievals-def _extract_field_hints(self, pattern: LocalToolPattern, query: str) -> None:"""Extract potential field names from search queries.diff --git a/headroom/learn/analyzer.py b/headroom/learn/analyzer.pyindex 9ce974579..dd8fe0939 100644--- a/headroom/learn/analyzer.py+++ b/headroom/learn/analyzer.py@@ -1162,18 +1162,3 @@ def _safe_int(val: object) -> int:except (ValueError, TypeError):return 0return 0---# =============================================================================-# Legacy compatibility alias-# =============================================================================---class FailureAnalyzer:- """Legacy alias for SessionAnalyzer — used by existing CLI code."""-- def __init__(self) -> None:- self._analyzer = SessionAnalyzer()-- def analyze(self, project: ProjectInfo, sessions: list[SessionData]) -> AnalysisResult:- return self._analyzer.analyze(project, sessions)diff --git a/headroom/memory/core.py b/headroom/memory/core.pyindex 776d5152c..0b7131b54 100644--- a/headroom/memory/core.py+++ b/headroom/memory/core.py@@ -894,15 +894,6 @@ class HierarchicalMemory:f"Bubbled memory {memory.id} from {current_scope.value} to {target_scope.value}")- def _scope_level_value(self, level: ScopeLevel) -> int:- """Get numeric value for scope level (lower = broader scope)."""- return {- ScopeLevel.USER: 0,- ScopeLevel.SESSION: 1,- ScopeLevel.AGENT: 2,- ScopeLevel.TURN: 3,- }[level]-# =========================================================================# Properties# =========================================================================diff --git a/headroom/memory/inline_extractor.py b/headroom/memory/inline_extractor.pyindex d04b9ae63..aef53a466 100644--- a/headroom/memory/inline_extractor.py+++ b/headroom/memory/inline_extractor.py@@ -173,95 +173,3 @@ def parse_response_with_memory(response_text: str | None) -> ParsedResponse:memories=memories,raw=response_text,)---class InlineMemoryWrapper:- """Wrapper that extracts memories from LLM responses inline.-- This is the zero-latency approach - memories are extracted- as part of the response, not in a separate call.-- Usage:- wrapper = InlineMemoryWrapper(openai_client)- response, memories = wrapper.chat(- messages=[{"role": "user", "content": "I prefer Python"}],- model="gpt-4o-mini"- )- # response = "Great choice! Python is excellent..."- # memories = [{"content": "User prefers Python"}]- """-- def __init__(self, client: Any):- """Initialize wrapper.-- Args:- client: OpenAI-compatible client- """- self.client = client-- def chat(- self,- messages: list[dict[str, Any]],- model: str = "gpt-4o-mini",- short_instruction: bool = True,- **kwargs: Any,- ) -> tuple[str | None, list[dict[str, Any]]]:- """Send chat request and extract memories inline.-- Args:- messages: Chat messages- model: Model to use- short_instruction: Use shorter memory instruction- **kwargs: Additional args for chat completion-- Returns:- Tuple of (response_content, extracted_memories)- """- # Inject memory instruction- modified_messages = inject_memory_instruction(messages, short=short_instruction)-- # Call LLM- response = self.client.chat.completions.create(- model=model,- messages=modified_messages,- **kwargs,- )-- raw_content = response.choices[0].message.content-- # Parse response and extract memories- parsed = parse_response_with_memory(raw_content)-- return parsed.content, parsed.memories-- def chat_with_response(- self,- messages: list[dict[str, Any]],- model: str = "gpt-4o-mini",- **kwargs: Any,- ) -> tuple[Any, str | None, list[dict[str, Any]]]:- """Send chat request and return full response object.-- Args:- messages: Chat messages- model: Model to use- **kwargs: Additional args for chat completion-- Returns:- Tuple of (response_object, content, memories)- """- modified_messages = inject_memory_instruction(messages)-- response = self.client.chat.completions.create(- model=model,- messages=modified_messages,- **kwargs,- )-- raw_content = response.choices[0].message.content- parsed = parse_response_with_memory(raw_content)-- # Modify response to have clean content- response.choices[0].message.content = parsed.content-- return response, parsed.content, parsed.memoriesdiff --git a/headroom/memory/traffic_learner.py b/headroom/memory/traffic_learner.pyindex 468071369..2939b06a9 100644--- a/headroom/memory/traffic_learner.py+++ b/headroom/memory/traffic_learner.py@@ -125,17 +125,6 @@ class PatternCategory(str, Enum):ARCHITECTURE = "architecture" # File structure, dependencies, conventions-class AgentType(str, Enum):- """Supported coding agent types."""-- CLAUDE = "claude"- CURSOR = "cursor"- CODEX = "codex"- AIDER = "aider"- GEMINI = "gemini"- UNKNOWN = "unknown"--# =============================================================================# Extracted Pattern Model# =============================================================================@@ -258,11 +247,6 @@ def _is_error(content: str) -> bool:# Tool Call Extractors# =============================================================================-# Extract command from Bash tool calls-_COMMAND_RE = re.compile(r"^(?:source\s+\S+\s*&&\s*)?(.+)", re.I)--# Extract file paths-_FILE_PATH_RE = re.compile(r"(?:/[\w./-]+(?:\.\w+)?)")# Extract package/module names from errors_MODULE_RE = re.compile(r"No module named ['\"]?(\w[\w.]*)['\"]?")@@ -758,14 +742,6 @@ class TrafficLearner:return list(by_hash.values())- def get_learned_patterns(self) -> list[ExtractedPattern]:- """Return patterns from the in-memory accumulator.-- Retained for backwards compatibility. Reads only the accumulator;- does not consult persisted rows. Use flush_to_file() for full data.- """- return [pattern for pattern, count in self._pattern_counts.values() if count >= 1]-async def on_tool_result(self,tool_name: str,diff --git a/tests/test_learn/test_analyzer.py b/tests/test_learn/test_analyzer.pyindex abd6b0e1c..d4a6d6498 100644--- a/tests/test_learn/test_analyzer.py+++ b/tests/test_learn/test_analyzer.py@@ -21,7 +21,6 @@ from headroom.learn.analyzer import (_strip_fenced_json,)from headroom.learn.models import (- AnalysisResult,ErrorCategory,ProjectInfo,RecommendationTarget,@@ -1445,20 +1444,3 @@ class TestCallLlmRouting:mock_cli.return_value = {}_call_llm("digest", "codex-cli")mock_cli.assert_called_once_with("digest", "codex-cli", on_progress=None)---# =============================================================================-# Legacy Compatibility-# =============================================================================---class TestFailureAnalyzerCompat:- @patch("headroom.learn.analyzer._call_llm")- def test_legacy_alias_works(self, mock_call_llm: MagicMock):- from headroom.learn.analyzer import FailureAnalyzer-- mock_call_llm.return_value = {"context_file_rules": [], "memory_file_rules": []}-- analyzer = FailureAnalyzer()- result = analyzer.analyze(_project(), [])- assert isinstance(result, AnalysisResult)