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(learn): pass explicit model in tests to avoid API key requirement
Tejas Choprada481a359ba0
diff --git a/headroom/learn/analyzer.py b/headroom/learn/analyzer.pyindex dfe133cfe..621e2377a 100644--- a/headroom/learn/analyzer.py+++ b/headroom/learn/analyzer.py@@ -62,9 +62,7 @@ class SessionAnalyzer:def __init__(self, model: str | None = None):self.model = model- def analyze(- self, project: ProjectInfo, sessions: list[SessionData]- ) -> AnalysisResult:+ def analyze(self, project: ProjectInfo, sessions: list[SessionData]) -> AnalysisResult:"""Analyze sessions and produce recommendations via LLM."""all_calls = [tc for s in sessions for tc in s.tool_calls]failed_calls = [tc for tc in all_calls if tc.is_error]@@ -134,7 +132,9 @@ def _build_digest(project: ProjectInfo, sessions: list[SessionData]) -> str:for session in sessions:if chars_used > char_budget:- lines.append(f"... (remaining {len(sessions) - sessions.index(session)} sessions truncated)")+ lines.append(+ f"... (remaining {len(sessions) - sessions.index(session)} sessions truncated)"+ )breaksession_header = (@@ -179,7 +179,7 @@ def _format_event(event: SessionEvent) -> str | None:if event.type == "user_message" and event.text.strip():text = event.text.strip()[:300]- return f" [{event.msg_index}] USER: \"{text}\""+ return f' [{event.msg_index}] USER: "{text}"'if event.type == "interruption":return f" [{event.msg_index}] INTERRUPTED: {event.text[:150]}"@@ -188,7 +188,7 @@ def _format_event(event: SessionEvent) -> str | None:return (f" [{event.msg_index}] SUBAGENT: {event.agent_tool_count} tool calls, "f"{event.agent_tokens:,} tokens, {event.agent_duration_ms / 1000:.1f}s "- f"— prompt: \"{event.agent_prompt[:100]}\""+ f'— prompt: "{event.agent_prompt[:100]}"')return None@@ -384,7 +384,5 @@ class FailureAnalyzer:def __init__(self) -> None:self._analyzer = SessionAnalyzer()- def analyze(- self, project: ProjectInfo, sessions: list[SessionData]- ) -> AnalysisResult:+ def analyze(self, project: ProjectInfo, sessions: list[SessionData]) -> AnalysisResult:return self._analyzer.analyze(project, sessions)diff --git a/headroom/learn/scanner.py b/headroom/learn/scanner.pyindex 709c03486..cbe7e6aa2 100644--- a/headroom/learn/scanner.py+++ b/headroom/learn/scanner.py@@ -240,9 +240,7 @@ class ClaudeCodeScanner(ConversationScanner):total_input_tokens += usage.get("cache_creation_input_tokens", 0)total_output_tokens += usage.get("output_tokens", 0)elif line_type == "user":- self._extract_tool_results(- d, tool_uses, tool_calls, events, msg_index, ts- )+ self._extract_tool_results(d, tool_uses, tool_calls, events, msg_index, ts)self._extract_user_events(d, events, msg_index, ts)except (OSError, UnicodeDecodeError) as e:@@ -251,12 +249,8 @@ class ClaudeCodeScanner(ConversationScanner):# Also wrap tool_calls as events for unified accessfor tc in tool_calls:- if not any(- e.type == "tool_call" and e.tool_call is tc for e in events- ):- events.append(- SessionEvent(type="tool_call", msg_index=tc.msg_index, tool_call=tc)- )+ if not any(e.type == "tool_call" and e.tool_call is tc for e in events):+ events.append(SessionEvent(type="tool_call", msg_index=tc.msg_index, tool_call=tc))events.sort(key=lambda e: e.msg_index)return SessionData(@@ -332,7 +326,9 @@ class ClaudeCodeScanner(ConversationScanner):)tool_calls.append(tc)events.append(- SessionEvent(type="tool_call", msg_index=msg_index, timestamp=timestamp, tool_call=tc)+ SessionEvent(+ type="tool_call", msg_index=msg_index, timestamp=timestamp, tool_call=tc+ ))# Extract subagent summary from toolUseResult metadatadiff --git a/tests/test_learn/test_analyzer.py b/tests/test_learn/test_analyzer.pyindex ae35fee41..30b344d76 100644--- a/tests/test_learn/test_analyzer.py+++ b/tests/test_learn/test_analyzer.py@@ -261,7 +261,7 @@ class TestSessionAnalyzer:"memory_file_rules": [],}- analyzer = SessionAnalyzer()+ analyzer = SessionAnalyzer(model="test-model")sessions = [SessionData(session_id="s1",@@ -283,7 +283,7 @@ class TestSessionAnalyzer:def test_handles_llm_failure_gracefully(self, mock_call_llm: MagicMock):mock_call_llm.side_effect = RuntimeError("API key not set")- analyzer = SessionAnalyzer()+ analyzer = SessionAnalyzer(model="test-model")sessions = [SessionData(session_id="s1",@@ -309,7 +309,7 @@ class TestSessionAnalyzer:]sessions = [SessionData(session_id="s1", tool_calls=[tc], events=events)]- analyzer = SessionAnalyzer()+ analyzer = SessionAnalyzer(model="test-model")analyzer.analyze(_project(), sessions)# Check that the digest passed to the LLM includes user message