@alsania-io/eme — Echo Memory Engine

Sovereign Memory System for AI Agents — Local, Encrypted, Multi-Agent Memory with MCP Protocol Support

What is EME?

Echo Memory Engine (EME) is a sovereign memory system for AI agents that provides:

• Local & Encrypted: Memory stays on your machine, encrypted at rest
• MCP Protocol Native: Fully compatible with Model Context Protocol
• Multi-Agent Support: Isolated namespaces for different agents
• Semantic Search: Vector-based memory retrieval with SQLite or Qdrant backend
• Knowledge Graph: Relationship tracking between memories
• Memory Gate: Noise filtering to prevent memory bloat
• Configurable Embeddings: Switch between embedding models and dimensions at runtime

Quick Start

Installation

npm install @alsania-io/eme@latest

Start the MCP Server

# Start EME as an MCP server
npx eme server

# Or with custom configuration
npx eme server --embeddingDimension 768 --logLevel debug

Use in Your Code

import { createMemoryManager } from '@alsania-io/eme';

// Create memory manager
const mm = await createMemoryManager({
  embeddingModel: 'local',
  embeddingDimension: 384,
  vectorStore: 'memory',    // or 'sqlite', 'qdrant'
  graphStore: 'memory',
  snapshotStore: 'filesystem',
  snapshotPath: './storage/snapshots',
  memoryGateEnabled: false,
  memoryGateThreshold: 0.3,
  maxMemoryEntries: 10000,
  similarityThreshold: 0.1,
  logLevel: 'info',
});

// Add a memory
const id = await mm.addMemory(
  'User prefers dark mode and uses a mechanical keyboard',
  'assistant-1',
  'user-preferences',
  [],
  'shared'
);

// Search memories
const results = await mm.searchMemories(
  'What keyboard does the user have?',
  5,           // limit
  'user-preferences'  // namespace (optional)
);

MCP Tools

EME exposes these tools via the Model Context Protocol:

Core Memory

| Tool | Description | |------|-------------| | add_memory | Store a new memory with embedding | | search_memories | Semantic search across memories | | get_memory | Retrieve a memory by ID | | update_memory | Update an existing memory | | delete_memory | Delete a memory by ID |

Knowledge Graph

| Tool | Description | |------|-------------| | create_entities | Create entities in the knowledge graph | | create_relations | Create relationships between entities | | add_observations | Add observations to existing entities | | delete_entities | Remove entities from the graph | | delete_relations | Remove relationships | | delete_observations | Remove specific observations | | read_graph | Get the full knowledge graph | | search_nodes | Search graph nodes by text | | open_nodes | Retrieve specific nodes by name | | get_graph | Get complete graph with stats | | get_graph_stats | Get graph statistics |

Snapshots

| Tool | Description | |------|-------------| | create_snapshot | Snapshot current state | | list_snapshots | List available snapshots | | load_snapshot | Restore from snapshot | | delete_snapshot | Remove a snapshot |

Configuration

| Tool | Description | |------|-------------| | get_config | Get current config (secrets redacted) | | update_config | Update a config value at runtime |

Local RAG

| Tool | Description | |------|-------------| | query_documents | Search ingested documents | | ingest_file | Ingest a file for RAG | | ingest_data | Ingest raw text content | | delete_file | Remove an ingested file | | list_files | List ingested files | | local_rag_status | Get RAG system status |

Large File Handling

| Tool | Description | |------|-------------| | read_large_file_chunk | Read a chunk of a large file | | search_in_large_file | Search within a large file | | get_file_structure | Analyze file structure | | navigate_to_line | Jump to a specific line | | get_file_summary | Get file statistics | | stream_large_file | Stream file in chunks |

Configuration

EME can be configured via:

1. Environment variables (highest priority)
2. Config file (eme-config.json)
3. CLI flags
4. Defaults (lowest priority)

Embedding Models

# Use local embedding (default, 384 dimensions)
EMBEDDING_MODEL=local EMBEDDING_DIMENSION=384

# Use OpenRouter (requires API key, 2048 dimensions for Llama Nemotron)
EMBEDDING_MODEL=openrouter \
  EMBEDDING_MODEL_PATH=nvidia/llama-nemotron-embed-vl-1b-v2:free \
  EMBEDDING_DIMENSION=2048 \
  OPENROUTER_API_KEY=sk-or-v1-...

You can switch embedding models at runtime via the update_config MCP tool. The system will reinitialize all subsystems with the new configuration.

Vector Stores

| Store | Description | Default | |-------|-------------|---------| | memory | In-memory, ephemeral | ✓ | | sqlite | SQLite with persistence | | | qdrant | Qdrant vector database | |

Environment Variables

See .env.example for all available configuration options.

Architecture

┌─────────────────────────────────────────────┐
│                 MCP Interface               │
├─────────────────────────────────────────────┤
│              Memory Manager                 │
│         (Embedding + Gate + Graph)          │
├─────────────┬──────────────┬───────────────┤
│ Vector Store│ Graph Store  │ Snapshot Store│
│ (memory/    │ (memory/     │ (filesystem/  │
│  sqlite/    │  sqlite)     │  ipfs)        │
│  qdrant)    │              │               │
└─────────────┴──────────────┴───────────────┘

Build & Development

# Build
make build

# Run tests
make test

# Start server
make start

# Development (watch mode)
make dev

# Security audit
make audit

# Build Podman container
make container

Security

• Local First: All data stays on your machine
• No Telemetry: No data leaves your system
• Secret Redaction: get_config never exposes API keys
• Open Source: MIT licensed, fully auditable

License

MIT License — see LICENSE file for details.

Made by Alsania I/O — Building sovereign AI infrastructure.