structured-memory-engine
Structured memory system for AI workspaces. Indexes markdown memory files into SQLite FTS5 for fast, cited search. Extracts structured facts, maintains memory health, and provides an MCP server with live search + write-path for Claude Code integration. --- # Structured Memory Engine ## MCP Tools (v4) When running as an MCP server (`node lib/mcp-server.js`), exposes: - `sme_query` — Search memory. Supports `query`, `limit`, `since`, `type`, `minConfidence`, `includeStale`. - `sme_context` — Get relevant context for a message. Returns ranked, token-budgeted, formatted context for injection. Supports `message`, `maxTokens`. - `sme_remember` — Save a fact/decision/preference to today's memory log. Auto-indexed. - `sme_index` — Re-index workspace. Use `force: true` for full rebuild. - `sme_reflect` — Run maintenance: decay, reinforce, stale detection, contradictions, prune. Use `dryRun: true` to preview. - `sme_status` — Index statistics. ## CLI Commands ```bash # Index workspace memory files node lib/index.js index [--workspace PATH] [--force] [--include extra.md,other.md] # Search indexed memory node lib/index.js query "search terms" [--limit N] [--since 7d|2w|3m|2026-01-01] [--context N] [--type fact|confirmed|inferred|...] [--min-confidence 0.5] [--include-stale] # Show index status node lib/index.js status [--workspace PATH] # Memory maintenance node lib/index.js reflect [--workspace PATH] [--dry-run] node lib/index.js contradictions [--workspace PATH] [--unresolved] node lib/index.js archived [--workspace PATH] [--limit N] node lib/index.js restore <chunk-id> [--workspace PATH] ``` ## Configuration
更新日志: Source: GitHub https://github.com/Bryptobricks/Structured-Memory-Engine
还没有评论,快来第一个发言吧。