Shared Organizational Memory for Enterprise Coding Agents: System Design and Deployment Snapshot
06:00 · August 4, 2026 · arXiv cs.AI RSS
Enterprise coding agents rely on tools and retrieval, yet enterprise knowledge often remains outside public training data and formal documentation: internal DSLs, proprietary platforms, local conventions, recent fixes, and tacit workflows. Existing knowledge interfaces expose stored resources but still depend on agents recognizing and explicitly recording lessons worth reusing, disconnecting capture from the coding workflow and leaving development experience repeatedly rediscovered. We report an ongoing production deployment of a shared organizational memory system that makes capture a platform-level part of coding work: it collects task-adjacent experience with contributor approval, curates it into reusable question-answer memories, gates obvious security and privacy risks, and retrieves memories for future agents. This short paper describes the deployed lifecycle and an operational snapshot. Effects on retrieval and coding tasks remain under evaluation.
Summary
The paper describes a production deployment at SAP SE of a shared organizational memory layer designed specifically for enterprise coding agents. Enterprise development environments often rely on internal domain-specific languages, proprietary platforms, local conventions, and tacit workflows that fall outside public training data and formal documentation. Existing agent interfaces provide access to files and tools but leave the capture of reusable lessons dependent on explicit agent recognition or manual intervention, so valuable experience is repeatedly rediscovered across teams and repositories.
The deployed system addresses this gap by embedding capture directly into the coding workflow. A Contributor Client monitors changes to a project-scoped local knowledge base through PostToolUse and Stop hook events. When edits occur, the client constructs a stable diff with surrounding context, obtains contributor approval, and forwards the candidate without requiring the coding agent itself to decide what is worth preserving. Approved contributions enter a Curation Pipeline that normalizes edits into self-contained question–answer memories, applies deterministic security and privacy scans for secrets, unsafe patterns, and personal data, assigns registry tags, and stores the results as governed, retrievable records.
Consumption occurs through the Model Context Protocol, allowing future agents to retrieve relevant memories during navigation, editing, testing, or debugging across repositories. The architecture keeps collection, curation, and retrieval as independently controllable stages connected by explicit API and MCP boundaries. A July 2026 operational snapshot records 1144 curated memories; quantitative effects on retrieval performance and task outcomes remain under evaluation.
Why it matters
Offers actionable design patterns for enterprise AI coding agents handling proprietary DSLs and tacit knowledge, directly applicable to Dutch/EU firms with similar internal platforms. Strong focus on privacy, security, and governance aligns with EU regulatory priorities.
