EARLY BIRD First 100 customers — LangChain v1+v2 included FREE with Pro
67 / 100 remaining
PERSISTENT MEMORY FOR AI AGENTS

Your agents
should remember
everything.

VEKTOR gives AI agents a persistent memory graph that survives session resets. Built on published research. One-time payment.

MEMORY_GRAPH // LIVE
ROOT EPISODIC SEMANTIC PROCEDURAL MEM_001 MEM_002 MEM_003 NODES: 7 EDGES: 10 STATUS: ACTIVE

Every session reset wipes your agent's memory. Completely.

WITHOUT VEKTOR // SESSION AMNESIA
SESSION_001
"User prefers Python. Working on data pipeline."
SESSION_002
✗ MEMORY WIPED — context lost
SESSION_003
✗ MEMORY WIPED — starting over again
SESSION_N
✗ Agent has no idea who you are
WITH VEKTOR // PERSISTENT GRAPH
SESSION_001
"User prefers Python. Working on data pipeline."
→ STORED IN GRAPH
SESSION_002
"Recalled: Python preference, pipeline context"
→ GRAPH UPDATED +3 NODES
SESSION_003
"Full context available. Continuing from last time."
→ GRAPH: 47 NODES, 89 EDGES
SESSION_N
→ COMPLETE AGENT MEMORY INTACT

Raw input → structured scene → persistent graph.

01
INPUT_LAYER

Raw Input

Conversation turns, tool outputs, observations — any unstructured agent context fed in as text.

CONVERSATION TOOL_OUTPUT OBSERVATION
02
SCENE_LAYER

Scene Graph

EverMemOS lifecycle extracts entities, relationships and temporal context into a structured scene representation.

ENTITY_EXTRACT AUDN_DEDUP TEMPORAL
03
GRAPH_LAYER

MAGMA Graph

Persisted as one of 4 MAGMA graph types in SQLite. Survives session resets. Queryable by the agent at any time.

EPISODIC SEMANTIC PROCEDURAL TEMPORAL

Four graph types. One unified memory system.

TYPE_01
HIERARCHICAL

Episodic

Time-ordered events and experiences. What happened, when, and in what sequence.

TYPE_02
ASSOCIATIVE

Semantic

Concepts, facts, and their relationships. What the agent knows about the world.

TYPE_03
SEQUENTIAL

Procedural

Step-by-step skills and workflows. How the agent accomplishes recurring tasks.

TYPE_04
CHRONOLOGICAL

Temporal

Time-weighted memory decay and reinforcement. Recent context weighted appropriately.

Drop into any Node.js agent in minutes.

QUICKSTART javascript
// 1. Import VEKTOR
const { VektorMemory } = require('./vektor-core');

// 2. Initialise with your agent
const mem = new VektorMemory({
  agent: process.env.AGENT_ID,
  graphType: 'episodic',
  licenceKey: process.env.VEKTOR_LICENCE
});

// 3. Store a memory
await mem.store({
  content: 'User prefers TypeScript',
  tags: ['preference', 'language']
});

// 4. Recall relevant context
const context = await mem.recall('coding preferences');
// → injects into your LLM prompt automatically
01

No external services

Runs entirely on SQLite. No cloud dependency, no API keys for memory, no data leaving your server.

02

Model agnostic

Works with OpenAI, Anthropic, Groq, Ollama, or any LLM. VEKTOR sits between your agent and the model.

03

Agent configurable

Set agent names via AGENTS= in your .env. Multi-agent systems supported out of the box.

04

D3.js visualizer included

Live graph visualizer ships with Studio tier. Watch your agent's memory grow in real time.

Implementation is original. Concepts are peer-reviewed.

READ FULL RESEARCH BREAKDOWN →

One-time. No subscriptions. Ever.

VEKTOR PRO
$59
One-time payment · Commercial licence
⚡ 67 of 100 early bird slots remaining
  • All 4 MAGMA graph types
  • AUDN deduplication loop
  • EverMemOS lifecycle
  • Local embeddings — no API cost
  • Groq, OpenAI, Ollama adapters
  • Private GitHub repo access
  • Use in commercial projects
  • LangChain v1+v2 adapter EARLY BIRD

Questions? [email protected]

STAY UPDATED

Get notified of updates

New features, early access, and nothing else.