Build Your Own Program Memory Layer
A starter kit of step-by-step instructions, not code, for building a shared program-memory layer for AI agents using Claude Code, native MCP integrations and git, covering the pull, map and loop levels of the architecture.
This repository is a guide rather than a codebase: it documents how to build the first three levels - full context pull, manual source mapping and an auto-ingestion loop - of a live, shared program-memory layer that lets every AI agent in an engineering organization answer questions about past decisions, ongoing work and open action items, using Claude Code and native MCP integrations rather than proprietary infrastructure.
Each family folder walks through the decision points and tradeoffs the authors found while building the same system for their managed product, Serro, so a team with the engineering bandwidth to operate it themselves can follow the same path. The proactive layer, a widget layer, and a semantic or graph index are explicitly out of scope and only pointed to, not documented.
The material is published by the company that sells a managed version of the same architecture, as a way for a team to fully understand what operating it themselves would involve before deciding whether to build or buy.