CognOS Session Memory

Verified context injection via epistemic trust scoring for session continuity between conversations.

CognOS Session Memory solves session fragmentation by maintaining verified, high-confidence context between LLM conversations. It extracts structured context from recent session traces, scores quality using the epistemic formula C=p*(1-Ue-Ua), and injects context as a system prompt only when confidence exceeds a configurable threshold.

The system provides a plan-mode gateway with three operating modes (auto, force, dry_run) and exposes save_session and load_session tools. Context that falls below the trust threshold is flagged for manual review rather than injected, preventing hallucination propagation. All injections are audited with trace IDs for compliance.