Memory Survival Kit

bitacorista-py has been open-sourced. To keep context from being lost when compaction hits during long Claude Code sessions, it mechanically detects important events using regex and keyword analysis and stores them on disk in structured form.

The core design is as follows.

• No LLM runtime dependency: it does not offload the decision of what to remember to another LLM call.
• SQLite storage: all structured memory is persisted to local disk.
• Connection graph / blast radius / confidence score: computes relationships and impact scope between entries.
• Thread merge detection: finds the right moment to merge scattered threads.

Figures were also provided.

• 29 Python scripts
• zero deps core
• 81ms/cycle
• all 55 automated tests passing

The author explained that this system is a structure designed to solve the problem of context loss in long conversations, and also released the GitHub repository.