Vex — Vector Exchange v0.4.0

Cross-standard vector DB migration tool with cryptographic memory portability. Export, sign, verify, and migrate agent memory between vector stores using the open .vmig.jsonl interchange format.

https://medium.com/ai-in-plain-english/your-agent-memory-is-trapped-heres-the-key-87224b483534

npx @vektormemory/vex export --from vektor --db slipstream-memory.db --output memories.vmig.jsonl
npx @vektormemory/vex sign memories.vmig.jsonl
npx @vektormemory/vex verify memories.vmig.jsonl
npx @vektormemory/vex import --from memories.vmig.jsonl --to pinecone --api-key $KEY --index my-index --host $HOST
npx @vektormemory/vex migrate --from vektor --to qdrant --db memory.db --url http://localhost:6333 --collection memories

Why

Every vector DB has a different API, a different format, and zero interop. Moving your agent memory from VEKTOR to Pinecone, or Qdrant to Weaviate, means writing a one-off script every time.

vex fixes that with a single open format, a growing connector library, and — as of v0.4.0 — cryptographic signing so you can prove your agent memory hasn't been tampered with in transit.

Your memory is always exportable, always portable, always verifiable, always yours.

What's New in v0.4.0

vex sign — BLAKE3 content-addresses every record, Ed25519 signs the root hash. Writes a .vmig.sig sidecar.
vex verify — Recomputes all hashes, verifies the signature. Exits 0 (valid) or 1 (tampered). Pipeline-safe.
--sign flag — Auto-sign immediately after export in one command.
--components flag — Selective disclosure. Export only working, semantic, procedural, episodic, or identity memories.
LangChain adapter — VektorMemory class, drop-in BaseMemory for any LangChain chain.
Dynamic schema detection — vektor connector now handles all SDK versions automatically via PRAGMA table_info.

Connectors

Connector	Export	Import	Status
`vektor`	✅	✅	Stable — dynamic schema, sqlite-vec ANN optional
`jsonl`	✅	✅	Stable
`pinecone`	✅	✅	Stable — tested 4,900 vectors
`qdrant`	✅	✅	Stable — tested 3,917 vectors, auto-create
`chroma`	✅	✅	Stable — auto-create collection
`weaviate`	✅	✅	Stable — GraphQL cursor pagination, extractStream
`pgvector`	✅	✅	Stable — schema introspection, extractStream

Install

# Global install
npm install -g @vektormemory/vex

# Or run without installing
npx @vektormemory/vex --help

# Signing support (BLAKE3 + Ed25519)
npm install @noble/hashes @noble/ed25519

# Vex Adapter (vec2vec projection — optional, premium)
npm install @vektormemory/vex-adapter

Requirements: Node.js >= 18. No extra dependencies for Pinecone, Qdrant, Chroma, or Weaviate — connectors use the built-in fetch API. pgvector requires npm install pg.

Commands

Sign & Verify (v0.4.0)

# Sign an export — generates .vmig.sig + .vmig.key
vex sign memories.vmig.jsonl

# Sign with an existing key
vex sign memories.vmig.jsonl --key memories.vmig.key

# Verify integrity (exit 0 = valid, exit 1 = tampered)
vex verify memories.vmig.jsonl

# Pipeline: verify before import
vex verify memories.vmig.jsonl && vex import --from memories.vmig.jsonl --to qdrant --collection mem

# Export + auto-sign in one step
vex export --from vektor --db memory.db --output memories.vmig.jsonl --sign

The .vmig.sig sidecar contains:

BLAKE3 hash of every individual record (canonical JSON)
Root hash (BLAKE3 of the array of record hashes)
Ed25519 signature of the root hash
Public key (verifier needs no private key)

Tamper with any byte in any record and vex verify will catch it.

Selective Disclosure (v0.4.0)

Export only specific memory component types — useful for multi-agent handoffs where different agents need different context.

# Export only working memory (current goals, TODOs, status)
vex export --from vektor --db memory.db --components "working" --output todos.vmig.jsonl

# Export working + procedural (skills + state) for a coding agent
vex export --from vektor --db memory.db --components "working,procedural" --output coding-ctx.vmig.jsonl

# Export semantic only (facts, knowledge)
vex export --from vektor --db memory.db --components "semantic" --output facts.vmig.jsonl

Memory types are stored via VEKTOR Slipstream SDK v1.6.1+:

await memory.remember('TODO: ship vex v0.4', { importance: 5, memory_type: 'working' });
await memory.remember('JWT tokens expire after 24 hours', { importance: 4, memory_type: 'semantic' });

LangChain Adapter (v0.4.0)

Drop-in BaseMemory implementation backed by VEKTOR Slipstream:

import { createVektorMemory } from '@vektormemory/vex/adapters/langchain';
import { ConversationChain } from 'langchain/chains';

// From a DB path
const vektorMem = await createVektorMemory({
  dbPath: './agent.db',
  topK: 5,
  importance: 3,
});

// Use in any LangChain chain
const chain = new ConversationChain({ llm, memory: vektorMem });
await chain.call({ input: 'What is our auth setup?' });

// Or use standalone
const ctx = await vektorMem.loadMemoryVariables({ input: 'JWT configuration' });
console.log(ctx.history); // top-5 relevant memories as formatted string

await vektorMem.saveContext(
  { input: 'What DB do we use?' },
  { output: 'PostgreSQL 15 with pgvector' }
);

Options:

createVektorMemory({
  dbPath:         './agent.db',   // path to VEKTOR SQLite DB
  topK:           5,              // memories to inject per turn
  memoryKey:      'history',      // key written into chain variables
  minScore:       0.0,            // min similarity score to include
  includeScores:  false,          // show similarity scores in output
  importance:     3,              // importance for saved turns
  returnMessages: false,          // return as ChatMessage[] for chat models
})

Export

# Export VEKTOR memory
vex export --from vektor --db ./slipstream-memory.db --output memories.vmig.jsonl

# Export + sign in one step
vex export --from vektor --db ./slipstream-memory.db --output memories.vmig.jsonl --sign

# Export specific components only
vex export --from vektor --db ./memory.db --components "working,procedural" --output ctx.vmig.jsonl

# Export specific namespace
vex export --from vektor --db ./memory.db --namespace trading --output trading.vmig.jsonl

# Export from Qdrant
vex export --from qdrant --url http://localhost:6333 --collection memories --output memories.vmig.jsonl

# Export from Pinecone
vex export --from pinecone --api-key $PINECONE_API_KEY --index my-index --host $HOST --output memories.vmig.jsonl

# Export from ChromaDB
vex export --from chroma --collection memories --output memories.vmig.jsonl

# Export from Weaviate
vex export --from weaviate --url http://localhost:8080 --collection MyDocs --output memories.vmig.jsonl

# Export from pgvector
vex export --from pgvector --url postgres://user:pass@host/db --table vex_vectors --output memories.vmig.jsonl

Import

# → Pinecone
vex import --from memories.vmig.jsonl --to pinecone \
  --api-key $PINECONE_API_KEY \
  --index my-index \
  --host https://my-index-xxxx.svc.pinecone.io

# → Qdrant (auto-creates collection if missing)
vex import --from memories.vmig.jsonl --to qdrant \
  --url https://xxxx.cloud.qdrant.io:6333 \
  --collection my-collection \
  --api-key $QDRANT_API_KEY

# → Qdrant local (no auth)
vex import --from memories.vmig.jsonl --to qdrant --collection memories

# → ChromaDB
vex import --from memories.vmig.jsonl --to chroma --collection memories

# → Weaviate
vex import --from memories.vmig.jsonl --to weaviate --url http://localhost:8080 --collection MyDocs

# → pgvector (auto-creates table + ivfflat index)
vex import --from memories.vmig.jsonl --to pgvector --url postgres://user:pass@host/db

# → VEKTOR (re-import for cross-model transfer)
vex import --from memories.vmig.jsonl --to vektor --db ./target.db

Migrate (direct — no intermediate file)

# VEKTOR → Qdrant
vex migrate --from vektor --to qdrant \
  --db ./memory.db --url http://localhost:6333 --collection memories

# VEKTOR → pgvector
vex migrate --from vektor --to pgvector \
  --db ./memory.db --url postgres://user:pass@host/db

# VEKTOR → Weaviate
vex migrate --from vektor --to weaviate \
  --db ./memory.db --url http://localhost:8080 --collection Memories

Embedding flags

# Re-embed from text field when moving between models (requires OpenAI key or Ollama)
vex migrate --from vektor --to qdrant --db memory.db --collection memories \
  --reembed --embed-model text-embedding-3-small

# Ollama re-embed (local, no API key)
vex migrate --from vektor --to qdrant --db memory.db --collection memories \
  --reembed --embed-model nomic-embed-text --ollama-url http://localhost:11434

# vec2vec projection — translate embeddings without any API call (premium)
vex migrate --from memories.vmig.jsonl --to pinecone \
  --adapter --adapter-model text-embedding-3-small

# List available projection pairs
vex adapters

.vmig.jsonl Format

One JSON object per line. UTF-8. Portable across any vector store.

{
  "id": "1234",
  "text": "JWT tokens expire after 24 hours. Algorithm: RS256.",
  "vector": [0.021, -0.043, 0.018, "...384 or 768 floats"],
  "model": "bge-small-en-v1.5",
  "dims": 384,
  "namespace": "default",
  "metadata": {
    "tags": "auth,security",
    "importance": 4,
    "memory_type": "semantic",
    "agent_id": "default"
  },
  "created_at": "2026-05-29T10:23:00.000Z",
  "source_store": "vektor",
  "vex_version": "1.0.0"
}

Key decisions:

Metadata is flat — Pinecone compatible out of the box
namespace is top-level — structural routing, not descriptive metadata
text field always preserved — enables cross-model re-embedding via --reembed
memory_type in metadata — enables --components selective disclosure
Sidecar .vmig.meta.json — record count, SHA-256 checksum, source store
Sidecar .vmig.sig — BLAKE3 Merkle root + Ed25519 signature (v0.4.0+)

Cryptographic Integrity (v0.4.0)

memories.vmig.jsonl     — your exported memories
memories.vmig.meta.json — record count, SHA-256, timestamps
memories.vmig.sig       — BLAKE3 Merkle root + Ed25519 signature
memories.vmig.key       — your Ed25519 private key (keep safe)

Verification is fully self-contained — the .vmig.sig file contains the public key, so anyone can verify without needing your private key:

vex verify memories.vmig.jsonl
# [vex verify] hashing 5725 records...
# ✓  Signature valid — file has not been tampered with

Verification checks three things:

Every record's BLAKE3 hash matches the hash stored in .vmig.sig
The root hash (BLAKE3 of all record hashes) matches
The Ed25519 signature of the root hash is valid

Embedding Handling

Scenario	Behaviour
Same model, same dims	Vectors copied directly — no re-embedding
Dim mismatch + `--reembed`	Re-embeds from `text` field via OpenAI or Ollama
Dim mismatch + `--adapter`	vec2vec projection — no API call (premium)
Dim mismatch, no flag	Records skipped with warning + count in summary
`null` vector	Record skipped with warning

Progress & Summary

[████████████████████] 100% pinecone (4900/4900)

┌─ pinecone summary ─────────────────────────
│  total records   : 4900
│  upserted        : 4900
│  skipped         : 0
│  duration        : 87.3s
└────────────────────────────────────────────

sqlite-vec (Optional — ANN Search)

The vektor connector supports sqlite-vec for native approximate nearest neighbour search inside SQLite. Without it, vex works fine — export falls back to standard ORDER BY created_at.

npm install sqlite-vec

# Backfill existing database
node scripts/migrate-vec.mjs --db slipstream-memory.db

# ANN-ordered export
vex export --from vektor --db memory.db \
  --vec-query '[0.021, -0.043, 0.018, ...]' \
  --limit 50 \
  --output results.vmig.jsonl

Roadmap

v0.0.1 — shipped — VEKTOR export, JSONL round-trip, format spec v1.0.0

v0.1.0 — shipped — Pinecone + Qdrant import, SHA-256 checksum, batch retry, progress bar

v0.2.0 — shipped — Pinecone + Qdrant export, ChromaDB connector, --namespace + --limit flags

v0.3.0 — shipped — Weaviate + pgvector connectors, --reembed, vec2vec adapter, streaming for >100k vectors, sqlite-vec ANN

v0.4.0 — shipped

vex sign / vex verify — BLAKE3 + Ed25519 cryptographic signing
--components — selective disclosure by memory type
--sign — auto-sign on export
LangChain adapter — VektorMemory BaseMemory implementation
Dynamic schema detection — handles all VEKTOR SDK DB versions

v0.5.0 — planned

vex diff — compare two .vmig.jsonl snapshots, surface added/removed/modified memories
Capability tokens — scoped read/export permissions per component type
TypeScript type definitions
Mem0 + Letta import connectors

Contributing

PRs welcome — especially new connectors.

Each connector is a single file in connectors/ implementing:

{ extract(opts), load(records, opts), extractStream(opts, onPage) }

See connectors/qdrant.js as the reference implementation. The Vex core handles batching, dimension filtering, retry, progress, and sidecar generation.

License

Apache 2.0 — free to use, fork, and build on.

npm · Docs Built by VEKTOR — persistent semantic memory for AI agents.

Name		Name	Last commit message	Last commit date
Latest commit History 58 Commits
adapters		adapters
connectors		connectors
core		core
formats		formats
scripts		scripts
utils		utils
CHANGELOG.md		CHANGELOG.md
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
SPEC.md		SPEC.md
package-lock.json		package-lock.json
package.json		package.json
vex.mjs		vex.mjs

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Vex — Vector Exchange v0.4.0

Why

What's New in v0.4.0

Connectors

Install

Commands

Sign & Verify (v0.4.0)

Selective Disclosure (v0.4.0)

LangChain Adapter (v0.4.0)

Export

Import

Migrate (direct — no intermediate file)

Embedding flags

.vmig.jsonl Format

Cryptographic Integrity (v0.4.0)

Embedding Handling

Progress & Summary

sqlite-vec (Optional — ANN Search)

Roadmap

Contributing

License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

Vex — Vector Exchange v0.4.0

Why

What's New in v0.4.0

Connectors

Install

Commands

Sign & Verify (v0.4.0)

Selective Disclosure (v0.4.0)

LangChain Adapter (v0.4.0)

Export

Import

Migrate (direct — no intermediate file)

Embedding flags

.vmig.jsonl Format

Cryptographic Integrity (v0.4.0)

Embedding Handling

Progress & Summary

sqlite-vec (Optional — ANN Search)

Roadmap

Contributing

License

About

Topics

Resources

License

Security policy

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages