Technical Specifications
The engineering foundation that powers maenifold's persistent knowledge graph.
Core Technology Stack
- Language
- C# with .NET 9.0
- Database
- SQLite with vector extension
- Vector Embeddings
- 384 dimensions (all-MiniLM-L6-v2 via ONNX)
- Search Algorithm
- Reciprocal Rank Fusion (RRF, k=60) with Hybrid search (semantic + full-text)
- Graph Sync
- Incremental with file watching
- Memory Format
- Markdown with YAML frontmatter
- URI Scheme
- memory:// protocol
- Protocol
- MCP (Model Context Protocol) with full tool annotation support
Scale Testing Results
Relationships tested and verified
Graph Performance
maenifold has been stress-tested with over 1.1 million concept relationships. The graph layer efficiently handles:
- Incremental concept updates with automatic relationship detection
- Semantic similarity search across large knowledge bases
- Hybrid RRF search combining semantic and full-text results
- Concurrent agent access to the same memory location
Demonstrated Capability
In production demonstrations, maenifold supported 25 AI agents across 68 minutes of coordinated discovery, creating 171,506 new concept relationships without any orchestration code—demonstrating emergent multi-agent coordination at scale.
Graph Layer Architecture
Automatic Graph Construction
The graph emerges from WikiLink usage without pre-defined schema or ontology:
- WikiLink Detection: Every [[Concept]] reference in markdown files automatically creates a graph node
- Semantic Embeddings: Each concept is vectorized in 384-dimensional space for similarity search
- Edge Weights: Relationship strength increases with repeated mentions across memory files
- Concept Clustering: Emergent patterns reveal knowledge structure and relationships
- Incremental Sync: File watching ensures the graph stays synchronized with memory updates
Search & Discovery
Multiple search modes optimize for different retrieval patterns:
- Hybrid RRF Search: Fuses semantic similarity and full-text matching using Reciprocal Rank Fusion (k=60)
- Semantic Search: 384-dimensional embeddings find conceptually related content
- Full-Text Search: Traditional regex-based pattern matching for exact content queries
- Context Building: Graph traversal reveals connected knowledge at any depth
MCP Compliance & Integration
Protocol Support
maenifold is fully compliant with the Model Context Protocol (MCP), providing a standardized interface for AI tools and applications:
- Tool Annotation: Full support for MCP tool specifications with complete metadata
- Resource Handling: Proper resource management for long-lived agent interactions
- Workflow Integration: Seamless orchestration with MCP-compatible AI frameworks
- Feature Parity: MCP and CLI interfaces have identical capabilities—start via MCP, continue via CLI
Supported Platforms
maenifold integrates with major AI tools and platforms:
- Claude Code, Continue, Cline: Full MCP integration via stdio protocol
- Codex: TOML configuration with configurable timeouts
- CLI Interface: Direct command-line access for scripts and pipelines
- Concurrent Agent Access: Multiple agents can read/write the same memory location safely
Architectural Principles
Local First
All knowledge lives as markdown files on disk with persistent URIs. No cloud dependency, no data leaving your system.
Transparent
Every thought, revision, and decision is visible in human-readable markdown. No hidden internal state or opaque operations.
Composable
Multiple agents can safely access the same knowledge base concurrently. No conflicts, no coordination code needed.
Emergent
Graph structure emerges from WikiLink usage without pre-defined schema. Organization arises naturally from how agents think.
Performance Characteristics
Memory Operations
Write, read, and search operations are optimized for both small personal knowledge bases and large multi-agent systems. Incremental file watching prevents expensive full-graph rebuilds.
Graph Operations
The graph layer efficiently handles incremental updates. New WikiLinks are detected and indexed automatically without blocking concurrent access.
Scalability
Demonstrated capability at scale: 1.1M+ relationships, 25 concurrent agents, 171,506 new concept relationships created in a single demonstration run.
Additional Resources
For more detailed information about maenifold's architecture and design philosophy: