Import & Merge · v1.5.0

Import & Merge

Six tools for moving memory data in and out of VEKTOR — from a simple export CLI to a full drop-folder consolidation daemon with causal lineage reconstruction. All tools are additive: they never modify the core memory engine.

Memory Export

Exports all memories to JSON, Markdown, or CSV. Supports filtering by tag, importance, date, namespace, and limit. No wire-in required — standalone CLI only.

vektor-export.js CLI

CLI

bash

node vektor-export.js json                         # → memories-export.json
node vektor-export.js markdown                     # → memories-export.md
node vektor-export.js csv                          # → memories-export.csv
node vektor-export.js json --out ./backup.json
node vektor-export.js json --tag trading
node vektor-export.js json --min-importance 0.7
node vektor-export.js json --limit 100
node vektor-export.js json --since 2025-01-01
node vektor-export.js json --namespace research
node vektor-export.js json --db ./custom.db

Programmatic

javascript

const { exportMemory } = require('./vektor-export');
const result = await exportMemory({
  format:        'json',
  tag:           'trading',
  minImportance: 0.7,
  out:           './backup.json',
});
// → { dest: './backup.json', count: 42, format: 'json' }

DB auto-detection order

Checks VEKTOR_DB_PATH env var → ./slipstream-memory.db → ./vektor-slipstream-memory.db → AppData path. Override with --db.

Schema Migrations

Idempotent additive migrations — safe to run repeatedly. Never drops or modifies existing columns. Runs automatically on first boot of any module that requires new columns.

vektor-migrate.js CLI

Migrations

ID	Description
001	Add `tags` TEXT column to memories
002	Add `confidence` REAL column (default 1.0)
003	Add `source_db` TEXT column (tracks import origin)
004	Create `vektor_migrations` tracking table
005	Index on `memories.tags`
006	Index on `memories.confidence`
007	Add `namespace` TEXT column (required by v1.4+)
008	Add `pinned` INTEGER column
009	Index on `memories.namespace`

CLI

bash

node vektor-migrate.js
node vektor-migrate.js --db ./slipstream-memory.db

Programmatic

javascript

const { runMigrations } = require('./vektor-migrate');
await runMigrations('./slipstream-memory.db');

Drop-Folder Daemon

Watches an ./import/ folder. Drop any .db file in — it auto-merges into the main database. Debounced 1.5s to handle slow file copies. Processes files already in the folder on startup. Failed merges move to import/error/, successful merges move to import/done/.

vektor-import-watch.js DAEMON

Wire-in

javascript

try{const _iw=require2('./vektor-import-watch');_iw.startImportWatcher().catch(function(){});}
catch(_){}

Usage

Drop any .db file into the ./import/ folder at any time — while the agent is running or offline. The daemon picks it up on startup if dropped while offline.

Environment Variables

Variable	Default	Description
VEKTOR_IMPORT_DIR	./import	Watched folder path
VEKTOR_IMPORT_TARGET_DB	./slipstream-memory.db	Target DB to merge into

DB Merge Engine

Core merge engine used by the drop-folder daemon. Merges memories from any source .db file into the target using content-hash dedup. Dangling edges are guarded. Processed files auto-move to import/done/.

vektor-db-merge.js CLI

CLI

bash

node vektor-db-merge.js --folder ./import --target ./slipstream-memory.db

Programmatic

javascript

const { mergeDb } = require('./vektor-db-merge');
const report = await mergeDb({
  source: './import/old-agent.db',
  target: './slipstream-memory.db',
});
// → { merged: 142, skipped: 18, total: 160 }

Dedup Strategy

Uses content-hash dedup — prevents true duplicates even when source and target have different UUIDs for the same memory content.

Namespace Isolation

Manages per-project or per-user SQLite DB files under a namespaces/ directory. Routes remember() to the active namespace. recall() targets the active namespace by default, or queries all namespaces simultaneously with the /cross prefix.

vektor-namespace.js PROXY

Wire-in

javascript

try{
  const _ns=require2('./vektor-namespace');
  await _ns.initNamespaces(memRef,{push,getMemory});
}catch(_){}

Switching Namespaces

javascript

// Switch at runtime (e.g. from /ns chat command)
await _ns.setActiveNamespace('trading', memRef);

// Cross-namespace recall — queries ALL namespaces simultaneously
const results = await memory.recall('/cross your query');

Environment Variables

Variable	Default	Description
VEKTOR_NAMESPACE	default	Active namespace name
VEKTOR_NAMESPACE_DIR	./namespaces	Path to namespaces folder
VEKTOR_NS_CROSS_RECALL_LIMIT	5	Results per namespace on cross-query
VEKTOR_NAMESPACE_ENABLED	true	Set to false to disable

CLI

bash

node vektor-namespace.js list
node vektor-namespace.js create research
node vektor-namespace.js stats
node vektor-namespace.js query research "your query"
node vektor-namespace.js merge trading default

MAGMA Causal Diff

When merging imported DBs, reconstructs the causal lineage — detecting which memories evolved from which across DBs. Preserves the why a memory changed, not just deduplicates content. Three conflict resolution strategies: importance-wins, newest-wins, and llm-arbitrate.

vektor-magma-diff.js ASYNC

Wire-in (registers as global, called by import-watch)

javascript

try{const _mg=require2('./vektor-magma-diff');globalThis.__magmaDiff=_mg;}
catch(_){}

// In vektor-import-watch.js, after merge completes:
if(globalThis.__magmaDiff){
  await globalThis.__magmaDiff.magmaMerge(srcPath, targetPath, { push, callLLM });
}

Conflict Resolution Strategies

Strategy	Description
importance-wins	Higher importance score takes precedence
newest-wins	Most recently updated memory wins (default fallback)
llm-arbitrate	LLM compares both memories and picks the winner. Times out after 8s and falls back to newest-wins

Environment Variables

Variable	Default	Description
MAGMA_RESOLVE_STRATEGY	newest-wins	Conflict resolution strategy
MAGMA_LLM_TIMEOUT_MS	8000	LLM arbitration timeout (ms)

What gets written to the DB

Table	Description
magma_edges	Causal edges between memories (source → derived)
magma_lineage	Full provenance chain per memory
magma_diff_log	Audit log of every merge decision

Driver Compatibility

Supports both better-sqlite3 and sql.js (required for Claude DXT). Auto-detects Electron context and skips native binary if running inside the DXT sandbox.

← Memory Intelligence

Troubleshooting →