What problem does it solve?
Continuous-learning captures recurring patterns, corrections, and preferences from conversations so that the assistant improves over time without manual reconfiguration. It reduces repeated mistakes, codifies heuristics, and helps the team benefit from past session knowledge.
Core Features & Use Cases
- Observe sessions: Scan a conversation for mistakes, user corrections, and repeatable patterns and surface candidate instincts for review.
- Record instincts: Parse user-provided descriptions into structured instincts and persist them to a line-delimited JSON memory store for future recall.
- Review & prune: Group instincts by category, flag contradictions, identify low-confidence or stale entries, and recommend merges or removals.
- Apply instincts: Filter and surface relevant instincts for the current task, boosting confidence for matches and integrating with pixl artifact storage when available.
- Use Case: After correcting a recurring React pattern in a session, record an instinct so future sessions proactively remind the developer to apply the preference.
Quick Start
Observe the current session for recurring mistakes and preferences, propose instincts for confirmation, and record confirmed instincts to the persistent memory store.