Frona Releases Ontology Memory

Self-hosted personal AI assistant Frona introduced an Ontology-based memory system via the v2026.8.0 update. Existing summary-based memory suffered from issues where model-generated errors were stored as facts and continuously influenced subsequent conversations; this update addresses this by adopting a method that verifies individual Claims before storing them.

Verified Memory Storage Process Frona does not unconditionally treat model outputs as truth, performing the following four verifications before storing them in long-term memory.

• Source Verification: All long-term memories maintain clear evidence such as conversation content, user confirmation, completed tasks, or successful tool results. Agents are prohibited from using their own previous memories as evidence.
• Type Classification: Entities such as people, projects, places, and services are systematically classified through the ontology to prevent the model from arbitrarily generating new labels.
• Identity Verification: Since names alone are insufficient, aliases, types, relationships, attributes, and contexts are compared to determine whether to merge with existing entities or create new ones.
• Change Tracking: Tracks how new information replaces or modifies existing knowledge.

This system runs as a single Rust process, enhancing security through a sandboxed execution environment, policy-based access control, and Vault-based credential management.