Sift

소개

Sift solves the critical problem of large language models forgetting context by giving Claude Code persistent, queryable memory that survives across sessions. It intelligently stores user preferences, coding patterns, project plans, and hard-won lessons in a structured format, enabling Claude to recall past interactions and decisions. This system grounds Claude in reality, combating hallucination by providing verifiable, exact information from indexed sources and enforcing strict reality checks during file operations, thereby ensuring accurate and auditable output.

주요 기능

• Persistent and queryable memory for Claude Code sessions
• Structured knowledge types: patterns, preferences, plans, tasks, gotchas, and reflections
• Full-text search with boolean operators across all memories for efficient retrieval
• 0 GitHub stars
• Intelligent and reality-checked file operations (read, edit, write) with exact file paths and line numbers
• Decision tracking and reflection logging for historical context and learning from past choices and corrections

사용 사례

• Ensuring Claude remembers coding preferences and architectural decisions across multiple development sessions
• Helping Claude avoid repeating past mistakes by recalling project-specific 'gotchas' and previous corrections
• Allowing Claude to reflect on its work, provide transparent reasoning, and log observations about the codebase