continual-learning

To preserve agent memory across sessions, you can use automatic learning capture via hooks alongside manual memory inputs. This approach maintains both global memory across projects and local memory per repository to track patterns and project-specific rules.

Continual learning for AI coding agents automates the preservation and reuse of knowledge across sessions. It maintains multi-tier memory with automatic captures and compaction rules to prune stale data while keeping high-value learnings.

To set up automatic learning capture, copy the continual-learning hooks into your project's .github/hooks directory. These hooks automatically capture agent learnings and safely persist them to designated memory stores.

Yes, you can track cross-project conventions using global memory. It specifically monitors patterns, conventions, and preferences across multiple projects, while local memory captures project-specific rules and common mistakes per repository.

Memory compaction works by applying predefined rules to prune stale data from the memory stores. This process preserves high-value learnings and ensures the persistent agent knowledge remains relevant and efficient over time.

No, manual memory inputs are optional. The system supports automatic learning capture via hooks, but you can also supplement the global and local memory files with manual inputs to enforce specific rules.