Unabyss is a universal, MCP-native context layer that maintains persistent, segmented context for AI agents and large language models. It automatically ingests and refreshes relevant context (files, events, and interaction history) so agents can access up-to-date information across sessions without manual prompt engineering. Unabyss exposes that context via the MCP connector protocol to any agent or LLM, helps solve transient 'goldfish memory' problems, and enforces segmentation and access boundaries by default to reduce accidental data exposure.