Cross-cutting persistence abstractions and data stores stage-21 (cross-cutting infrastructure)

This stage is the system’s long-term memory and filing system. It is shared support used throughout startup, normal work, recovery, and cleanup, rather than one single user-facing flow. Its parts save different kinds of information so the app can resume work, avoid repeats, and keep private data safe.

Rollout files and thread-store persistence keep conversation records: raw session logs, searchable indexes, live writers for active chats, and tools to list, rebuild, archive, or delete threads. SQLite runtime state and agent graph storage keep structured records in a small local database, such as thread summaries, goals, agent jobs, progress checkpoints, audit views, and “which agent started which thread.” Caches and local lookup files store reusable facts like cloud settings, connector lists, plugin catalogs, models, update notices, and local plugin paths, while checking age, account, and version before trusting them. Plugin, secrets, and memory stores manage installed plugins, encrypted secret values, and saved user memories. External session import persistence reads conversation files from other tools and keeps a ledger so the same outside session is not imported twice.