D3 4 Layer Magma Graph
Cloak Stealth Browser
Last verified:
What is D3 4 Layer Magma Graph?
D3 4 Layer Magma Graph is a local-first persistent memory layer and SDK for Node.js AI agents that provides a structured, four-layer associative memory graph (MAGMA) to store, curate, and recall agent experiences across sessions. It bundles an on-device engine (SQLite + ONNX), embedding model, curation loop (AUDN), and lifecycle manager (EverMemOS) so agents can remember conversation turns, tool outputs, and observations without cloud dependencies. Key features include automatic deduplication and contradiction handling, a scheduled multi-phase
D3 4 Layer Magma Graph pricing
Pricing model: Freemium
The website lists paid options and mentions a low-cost subscription tier (examples shown: Pro at $59 and Studio at $129 in docs snippets, and a separate reference to $9/month on marketing pages), plus a licence key required to download the CLI; exact plan names and feature mappings are presented on the product page and quickstart (free evaluation paths are not emphasised), and bundles include the Slipstream SDK, MAGMA engine, AUDN loop, local embedding model, and integration adapters. The documentation highlights that VEKTOR is designed to avoid ongoing cloud API costs by running locally; licence activation is required to access the full CLI and MCP tooling.
D3 4 Layer Magma Graph pros
- 4-layer MAGMA associative graph for episodic, semantic, procedural, and temporal memory
- Runs entirely local on SQLite — no cloud or external services required
- Bundled ONNX and a small local embedding model for offline operation
- AUDN automatic curation loop for deduplication and contradiction resolution
- Scheduled 7-phase dream cycle that compresses many fragments into Core Insight nodes
- createMemory/remember/recall API for simple integration into agent loops
- Low-latency recall performance (desktop-grade, ~8ms mentioned for recall)
- Built-in CLI and MCP server for quick developer workflows and deployment
- D3 force-graph dashboard to visualise memory with semantic/temporal filters
- Settings GUI with hot-reload and visual model/provider selection
- Prebuilt integrations and adapters for LangChain, OpenAI, Claude, Mistral, Groq, Ollama
- Namespace isolation for multi-agent deployments and safe memory separation
- Import/export and drop-folder DB merge tools with MAGMA causal lineage reconstruction
- Confidence scoring, recall tuning, and RL prioritisation for retrieval control
- Several developer conveniences: bundled examples, full API reference, and troubleshooting guides
D3 4 Layer Magma Graph cons
- Node.js v18+ requirement limits direct use from non-Node hosts
- Primary SDK and tooling focused on Node/JavaScript rather than Python-first stacks
- Local storage (SQLite) can still encounter locking/BUSY issues on heavy concurrent writes
- ONNX and embedding model size increases local install footprint (ONNX bundled)
- Some advanced features (MCP, Claude DXT, Cloak) require additional setup/licence activation
- GUI and dashboard auto-start on a fixed port may require network configuration
- Proprietary licence key required to download CLI and activate full features
- Limited cloud-native managed offering — no turnkey hosted persistence or multi-region sync
Frequently asked questions about D3 4 Layer Magma Graph
How do I get started and install VEKTOR Slipstream?
Install via npm or yarn (npm install vektor-memory or yarn add vektor-memory), ensure Node.js v18 or later is installed, activate your licence key during setup, then call createMemory() at agent startup and use remember() and recall() in your agent loop as described in the quickstart.
What memory types does VEKTOR store and how are they organised?
VEKTOR organises memory into a four-layer MAGMA graph comprising episodic, semantic, procedural, and temporal nodes so the system can store conversation turns, tool outputs, and observations with causal lineage and temporal context for richer recall and reasoning.
Does VEKTOR require a cloud service or external database?
No — VEKTOR is local-first and runs on an embedded SQLite database bundled with the package, including ONNX support and a small local embedding model so agents operate without cloud dependencies or ongoing hosting costs.
How does deduplication and contradiction handling work?
The AUDN curation loop continuously evaluates incoming memories and decides to ADD, UPDATE, DELETE, or NO_OP; it detects duplicates and contradictions and applies deduplication and contradiction resolution automatically as part of the memory lifecycle.
What is the dream cycle and what does it do?
The scheduled multi-phase dream cycle compresses raw conversational fragments into higher-density summaries and Core Insight nodes through a seven-phase process, typically turning many fragments into a single synthesized insight and reducing token costs for session recalls.
Which LLM providers and stacks are supported?
VEKTOR provides adapters and examples for multiple providers including OpenAI, Anthropic/Claude, Groq, Ollama, Mistral and integrates with LangChain and other agent frameworks so you can connect your preferred model provider via the Settings GUI or environment variables.
Can I visualise and explore agent memory?
Yes — the product includes a live D3 force-graph dashboard that visualises memory nodes and edges, lets you filter by semantic, causal, temporal, or entity nodes, and opens on a local dashboard endpoint (auto-starts on port 3847 by default).
How do multi-agent namespaces work?
When creating a memory instance you set an agentId which namespaces stored memories so multiple agents can share one database without collisions and supports cross-namespace recall and isolated imports/exports.
What developer tooling and CLI commands are available?
The CLI exposes commands such as vektor setup, activate, status, test, vektor mcp to start the MCP server, vektor rem and briefing utilities; the package also includes full API references and example integration files for quick developer onboarding.
How do I handle import, export, and migrations?
VEKTOR supports drop-folder DB consolidation, auto-merge of .db files placed into the import folder, exports in JSON/Markdown/CSV, schema migrations, and causal lineage reconstruction via MAGMA to safely merge or migrate memory stores.