March 22, 2026

AI Core & Goal Ecology Major Updates

The encryption key service now stores the key directly in the database, ensuring the key and encrypted data are never separated, and it automatically migrates f

The encryption key service now stores the key directly in the database, ensuring the key and encrypted data are never separated, and it automatically migrates from the legacy .key file on upgrade. Token analytics dashboard received fixes for field name mismatches, dict key access in stats queries, and added definitions for several LLM job types to ensure full visibility.

A new short message gate was introduced to prevent false topic boundaries on terse answers, which were previously flagged as outliers. This gate, coupled with a lower default threshold, significantly improved boundary detection accuracy and F1 score.

Several database migrations were addressed to prevent crashes when upgrading from v0.1.0, specifically by handling the missing reliability column in the user_traits schema.

The persistent storage of the encryption key was stabilized by falling back to a Docker volume path for new installations, and provider cache poisoning was prevented by stopping caching empty provider results.

The core goal ecology system saw major integration, wiring the signal pipeline and KNN search functionality, and adding an is_muted column for goal filtering.

The system began supporting cross-goal inference by detecting clusters of related goals and suggesting consolidation, along with ambient-informed goal formation.

The goal architecture was deepened by adding goal hierarchy via embedding similarity, and allowing users to retrieve narrative summaries of their goal evolution using LLM storytelling.

The invocation tracker was unified to track every execution path consistently across tools and skills, feeding metrics to both tool performance and procedural memory.

  • Encryption key is now stored in the database (encryption_keys table).

  • Short message gate introduced to reduce false topic boundaries on terse answers.

  • Goal ecology wired to live via signal pipeline and KNN search integration.

  • Cross-goal inference detects and suggests consolidation for related goals.

  • Goal narrative synthesis allows users to receive story-based summaries of their goals.

  • Invocation tracker unified to monitor all execution paths consistently.