Egc gives ai agents persistent memory

Egc is a local runtime that installs a pair of MCP servers and creates a ~/.egc/state directory where each project gets a plain-Markdown file storing decisions, failures, preferences, and next steps [DevTo]. The runtime loads this file when any supported AI coding tool starts a new session and writes back updates at the end of the session. Installation is a single npm command (npm install -g @egchq/egc && egc install) or an npx one-off [GitHub]. Egc detects which of the 12 supported assistants are present and wires them to the same state file. The tool includes 479 components (63 agents, 229 skills, 76 commands) that can be added with npx skills add Fmarzochi/EGC. Core APIs such as get_state, update_state, store_decision, and compress_observations compress hook events to save token budget. Validation tools (validate_command, validate_write) guard against unsafe operations. Cross-session lesson storage (lesson_save, lesson_recall) with confidence decay keeps knowledge fresh. Egc's MIT-licensed source code is hosted on GitHub and maintained by a single developer. By compressing observations and reducing context payloads before each prompt, Egc trims the token count that each model must process, directly lowering API costs for high-frequency coding assistants. A unified memory layer ensures consistent decisions and avoids divergent documentation across tools. Persistent memory forces developers to think about data ownership, versioning, and security in a way that pure-cloud services have avoided.