NeuraKeep

NeuraKeep provides a local-first memory layer designed for computational agents, ensuring reliable and source-cited recall. Key features include: * Source-cited memory tracking * Proposal-based update review * Failure memory and warnings * Unified memory across various agents * Local and hosted deployment options This platform transforms project notes, sessions, PDFs, and chats into a universally searchable and citable memory base for agents. Every durable claim is linked back to its original raw source, including specific sections, timestamps, and assigned trust/sensitivity levels. Agents can propose new facts, events, decisions, and past failures, with human or policy-controlled reviewers determining what becomes a permanent part of the memory. The system's unique failure memory functionality allows agents to identify and warn against previously unsuccessful paths, preventing repeated mistakes in coding or other actions. It offers a single, consistent memory layer accessible by different computational agents like Claude, Codex, and OpenClaw via MCP tools, eliminating the need for each agent to maintain a separate memory silo. NeuraKeep is built with robust governance features to manage duplicates, conflicts, stale facts, sensitivity, and provides full audit trails and undo capabilities. Whether run locally from GitHub for individual operators or used in hosted personal and team spaces with governed remote MCP access, NeuraKeep offers a controlled memory loop for engineering teams, content creators, and researchers who depend on accurate, verifiable information for their computational agents.