Overview
ai-memory treats memory as governed continuity infrastructure for coding agents, not a private notebook or unquestioned source of truth. Across the series, bounded lifecycle events become session evidence and owner-aware handoffs, selected knowledge is compiled into a human-readable Markdown wiki, and SQLite supplies derived indexes, state, and retrieval. Hybrid search, versioning, retention, review, scope, privacy, backups, and optional native-session bridges extend the system without erasing uncertainty. Current code, tests, instructions, and runtime behavior remain the authority: memory preserves rationale and reduces repeated work, but operators must verify what is true and authorized now.
Learning outcomes
- ai-memory carries work across sessions and harnesses through bounded evidence, typed handoffs, and searchable project knowledge.
- Markdown is the human-readable knowledge surface; SQLite provides derived indexes and operational state, so both layers need distinct recovery plans.
- Retrieval can improve recall through lexical, entity, graph, and optional vector signals, but ranked history does not establish present authority.
- Current code, instructions, tests, permissions, and runtime behavior decide what is true and allowed now, not historical memory.
- Use the smallest continuity layer that solves the workflow, adding native resume, enrichment, or shared deployment only when its trade-offs fit.
Keywords
Preview every chapter · Unlock full text, audio, and video from chapter 4
PlansFAQ
What problem does ai-memory solve for coding agents?
ai-memory solves the loss of project reasoning across sessions, harnesses, machines, and teammates by preserving bounded evidence, searchable knowledge, and typed handoffs. It reduces repeated investigation without replacing current-code verification.
Why does ai-memory use Markdown with SQLite?
ai-memory uses Markdown as the inspectable, versioned knowledge surface and SQLite as the derived index and operational store. The wiki and database support different recovery needs, so a complete backup must account for both.
Does ai-memory replace current code and project instructions?
No. ai-memory memory preserves historical rationale and open context, while the current checkout, instructions, tests, permissions, and runtime behavior determine what is true and authorized now.
When should a team centralize ai-memory?
A team should centralize ai-memory when lost reasoning across machines or people costs more than operating a governed shared service. Remote use requires appropriate transport protection, credentials, project scope, backups, and acceptance that project pages are shared.