Verifies and injects high-confidence context into LLM sessions to prevent fragmentation and improve continuity.
Sponsored
Large language models often lack verified context from previous interactions, leading to repeated explanations and broken workflows. CognOS Session Memory addresses this by acting as a plan-mode gateway, extracting structured context from recent traces, and scoring its quality using a unique epistemic trust formula. Only context exceeding a defined trust threshold is injected as a system prompt, preventing the propagation of hallucinations. This system also flags low-trust context for manual review and audits all injections, ensuring reliable and compliant long-running LLM conversations.
Key Features
01Scores context quality using a proprietary epistemic trust formula (C = p · (1 − Ue − Ua)).
02Conditionally injects context as a system prompt only if its trust score exceeds a defined threshold.
03Flags low-trust context for manual review, preventing untrustworthy information from propagating.
04Audits every context injection with trace IDs for compliance and transparency.
05Provides a flexible API gateway and MCP server integration for LLMs like Claude Code.
061 GitHub stars
Use Cases
01Maintaining verified, high-confidence session context across multiple LLM conversations to enable long-running workflows.
02Preventing the injection of unverified or hallucinated context into LLM sessions to improve reliability.
03Ensuring compliance and auditability for context injected into AI systems, addressing requirements like the EU AI Act.
