Skip to content

CLAUDE_API_OPTIMIZATION_SUITE.md

Latest commit

 

History

History
87 lines (65 loc) · 3.16 KB

CLAUDE_API_OPTIMIZATION_SUITE.md

File metadata and controls

87 lines (65 loc) · 3.16 KB
Raw

Claude API Optimization — Canonical Module

Canonical module: autobot-backend/utils/claude_api_integration.py

The former claude_api_optimization_suite.py was retired in #10796 (permanently broken import from the SLM-only monitoring.claude_api_monitor; zero callers). All unique logic — OptimizationMode enum, OptimizationMetrics dataclass, TodoWrite optimization, graceful degradation, tool-pattern analysis, dynamic mode switching — has been folded into the canonical module.

Overview

autobot-backend/utils/claude_api_integration.py provides a unified optimization layer for all outbound Claude/LLM requests in AutoBot. It is wired into backend startup during Phase 2 (initialization/lifespan._init_claude_api_integration) and stored on app.state.claude_api_adapter.

Key Classes

Class Purpose
AutoBotClaudeAPIAdapter Production singleton; use get_autobot_claude_adapter()
ClaudeAPIBatchManager Core manager: rate limiting, batching, optimization, metrics
ClaudeAPIConfig Configuration dataclass
OptimizationMode Enum: CONSERVATIVE / BALANCED / AGGRESSIVE / EMERGENCY
OptimizationMetrics Snapshot metrics dataclass

Quick Start

# Production usage — singleton already started at backend startup
from utils.claude_api_integration import get_autobot_claude_adapter

adapter = await get_autobot_claude_adapter()

# Chat
response = await adapter.process_chat_request("Explain async/await")

# Code analysis
result = await adapter.process_code_analysis(code_str)

# TodoWrite optimization (batches/deduplicates todos)
await adapter.submit_todowrite([
    {"content": "Task 1", "status": "pending"},
    {"content": "Task 2", "status": "pending"},
])

# Metrics / status
status = await adapter.get_performance_stats()

Optimization Modes

Mode req/min req/hour When
CONSERVATIVE 60 2500 Light usage
BALANCED (default) 50 2000 Regular development
AGGRESSIVE 30 1500 High-frequency sessions
EMERGENCY 15 800 Rate-limit recovery

Modes switch automatically when the background pattern-analysis task detects critical inefficiencies (priority_score > 0.8 on more than 3 recommendations).

Components

  1. Conversation Rate Limiter (conversation_rate_limiter.py) — sliding-window limiting
  2. Payload Optimizer (payload_optimizer.py) — compression, chunking
  3. Intelligent Request Batcher (request_batcher.py) — time-window + similarity batching
  4. Graceful Degradation (graceful_degradation.py) — 5-level fallback system
  5. TodoWrite Optimizer (todowrite_optimizer.py) — consolidation and deduplication
  6. Tool Pattern Analyzer (tool_pattern_analyzer.py) — efficiency scoring

Startup Wire-in

initialization/lifespan.initialize_background_services()
  └─ _init_claude_api_integration(app)   ← Phase 2, non-critical
       └─ get_autobot_claude_adapter()   ← lazy singleton
            └─ ClaudeAPIBatchManager.start()

Testing

from utils.claude_api_integration import AutoBotClaudeAPIAdapter

AutoBotClaudeAPIAdapter.reset_instance()  # isolate test