Show HN: I built a pocket dimension for notes and context in LLM-space (www.kept.fyi)

Keptis is a lightweight “pocket dimension” for storing notes and context specifically tuned for LLM workflows: stash context anywhere and recall it everywhere. It lets you save quick mementos, links, chat summaries, full codices and visual assets, then expose that collection to models or agent tools on demand. The project plugs into MCP-supported environments — currently Claude, Claude Code, Cursor, and VS Code — enabling ad‑hoc insight or agentic access so an LLM can read from and act on your personal knowledge collection during a session. For the AI/ML community this is notable because it formalizes persistent, model-accessible memory and makes retrieval-augmented generation and agent behavior more practical across tools. Technically it’s about indexing arbitrary artifacts (text, images, summaries) into a reusable context layer that can be surfaced to different models and toolchains; that enables richer, stateful agents, faster developer workflows, and more consistent personal knowledge retrieval. Key implications include easier creation of agentic workflows, improved prompt grounding, and potential privacy/security considerations around granting models access to sensitive stores. Keptis is a building block for composable memories in LLM ecosystems and points toward more interoperable personal knowledge layers across model hosts.