Because a solution is linked to the problem it solved and the technology it applies to, "has this failed before" and "what replaced this" become traversals rather than keyword searches. The project marks core memory operations as current work and the intelligence layer as still ahead, so treat the analytics tools as newer ground than storage.
A Python server backed by Neo4j that stores development memories — tasks, code patterns, problems, solutions, projects and technologies — along with the relationships between them.
- Memories stored, fetched, searched, updated and removed — store_memory, get_memory, search_memories, update_memory, delete_memory
- Memories linked to each other and traversed — create_relationship, get_related_memories, analyze_relationships
- Seven categories of relationship, from causal (CAUSES, TRIGGERS, LEADS_TO) through to quality (EFFECTIVE_FOR, PREFERRED_OVER, DEPRECATED_BY)
- A project scanned into a contextual memory graph — analyze_codebase
- Workflow recorded, analogous past solutions surfaced, and success estimated — track_task_execution, suggest_similar_solutions, predict_solution_effectiveness
- The knowledge network read back: get_memory_graph, find_memory_paths, memory_effectiveness
Python 3.10 or higher and a Neo4j database, local or hosted. NEO4J_URI (default bolt://localhost:7687), NEO4J_USER (default neo4j) and NEO4J_PASSWORD; MEMORY_LOG_LEVEL is optional. Install with pip install -e . and initialise the schema with python -m claude_memory.setup before the first run.
One command plus a key — pip install -e ., then supply credentials