Zikkaron

Acerca de

Large language models like Claude Code often struggle with retaining context across sessions, leading to repetitive explanations of project specifics. Zikkaron addresses this by giving Claude Code a sophisticated, biologically-inspired memory engine. Running locally over SQLite, it automatically stores, consolidates, and intelligently recalls architectural decisions, debugging history, and project conventions, ensuring that relevant knowledge is always available, evolving, and optimized for retrieval. This system is built on computational neuroscience principles, allowing Claude Code to learn and adapt its understanding of a codebase over time, all while keeping data securely on your machine.

Características Principales

• Biologically-Inspired Memory Dynamics: Implements predictive coding for smart ingest, heat-based salience for natural decay, and reconsolidation for dynamic memory updates.
• Advanced Knowledge Structuring: Utilizes fractal hierarchies for multi-scale summarization, a knowledge graph for entity-relationship linking, and causal discovery for understanding cause-effect.
• Metacognitive Capabilities: Features self-assessment of knowledge coverage, gap detection across five dimensions, and neuro-symbolic rules for fine-grained control over memory behavior.
• 4 GitHub stars
• Comprehensive MCP Toolset: Exposes 18 specialized tools via the Model Context Protocol (MCP) for memory management, context retrieval, prospective triggers, and more.
• Local & Private Persistent Storage: Stores all data, including embeddings and knowledge graphs, in a single SQLite database with FTS5 and `sqlite-vec` for secure, offline functionality.

Casos de Uso

• Allow Claude Code to consolidate and evolve its understanding of a codebase over time, generating autobiographical project narratives and identifying knowledge gaps.
• Enable Claude Code to automatically retain and recall architectural decisions, debugging history, and project conventions across development sessions.
• Provide contextually relevant past work, such as solutions to previous bugs or feature implementations, to assist Claude Code in new tasks without re-explanation.