KERNL

A 128-tool MCP server for session persistence, semantic memory, cross-project learning, adversarial testing, and crash recovery — built to give a stateless assistant continuity across sessions instead of starting cold every time.

The problem

Every new conversation with an LLM starts from zero. No memory of decisions made yesterday, no awareness of what was tried and failed last week, no ability to notice a pattern that shows up across five different projects. That's a real ceiling on what an assistant can actually do for ongoing work.

Related MCP server: Self-Improving Memory MCP

What it does

KERNL provides the infrastructure layer underneath a persistent working relationship with an AI assistant: semantic recall over past sessions, cross-project pattern matching, structured checkpointing so a crash doesn't lose hours of context, and adversarial self-testing so conclusions get challenged rather than accepted at face value. 128 tools across session management, memory, testing, and project coordination.

Part of a system

KERNL is one piece of a larger cognitive-infrastructure stack — memory, maintenance, adversarial testing, and session continuity as separate, composable layers rather than one monolith. See davidkirsch.me/builds for the rest.