maenifold maenifold
Persistent Graph-of-Thoughts

Your Agent is ephemeral.

Knowledge persists.

maenifold transforms isolated AI sessions into a persistent knowledge graph with WikiLinks, vectors, and emergent structure. Test-time reasoning that compounds.

Get Started Usage & Docs

maenifold

Your AI is ephemeral. Your knowledge shouldn’t be.

What maenifold does

maenifold enhances AI agents with persistent graphs of thought that compound over time. Every tool creates [[WikiLink]] connections that survive conversations. Every session builds on the last. Knowledge compounds instead of resets. It transforms ephemeral AI sessions into continuous collective intelligence.

graph

Our demo shows this at scale: 25 AI agents across 68 minutes discovered a critical production bug not through programmed coordination, but through emergent understanding. No orchestration code was written, yet agents perfectly orchestrated themselves across 4 waves, building on each other’s discoveries through a shared knowledge graph.

The critical move operation bug emerged from the intersection of multiple test patterns seen across different agent sessions - something no single agent could have found alone. This is maenifold’s core: making every AI session additive rather than isolated.

How it worked:

85% test success rate. Real production bug found. Zero orchestration code written.

That’s what maenifold does: It provides the substrate (WikiLinks, memory, graph) and steps back. Intelligence fills the space.

Cognitive Assets

maenifold ships with 54 pre-configured cognitive frameworks that agents can adopt dynamically:

🔄 28 Workflows - Structured Methodologies

From reasoning patterns to development processes:

🎭 7 Roles - Professional Perspectives

Each with personality, principles, and transition triggers:

🎨 7 Colors - De Bono’s Six Hats + Gray

Thinking modes for different cognitive approaches:

🗣️ 12 Perspectives - Sapir-Whorf Linguistic Patterns

Language-influenced cognitive styles based on linguistic relativity:

These aren’t just templates - they’re cognitive substrates that combine with the knowledge graph to create emergent reasoning capabilities. An agent can start with strategic thinking, switch to red team perspective when finding vulnerabilities, then adopt the writer role to document findings - all while building on the persistent knowledge graph.

The Cognitive Stack

Tool Relationships

graph TD
    subgraph OVERALL["maenifold"]
        subgraph AGENT["AI Agent"]
            LLM[🧠 LLM Agent]
        end

        subgraph PERSP["Perspectives"]
            A[🎭 Adopt]
        end

        subgraph THINK["Thinking"]
            ST[🌊 SequentialThinking]
            WF[📋 Workflow]
        end

        subgraph MEMORY["Memory System"]
            MEM[🗄️ Memory]
                GRPH[🕸️ Graph]
        end
    end


    A --> LLM
    LLM --> WF
    LLM --> ST
    WF -.-> ST
    ST --> MEM
    WF --> MEM
    MEM --> GRPH
    GRPH --> LLM
    MEM --> LLM

Reasoning Layer (Tools + Workflows) - Where Information is Processed

Where test-time computation happens:

Memory Layer (memory://) - Where Data is Stored

Graph Layer (SQLite + vectors) - Where Knowledge Emerges

Automatic graph construction from [[WikiLinks]] with:

Technical Specifications

Quick start

Install

npm install -g @ma-collective/maenifold

MCP Interface

Claude Code, Continue, Cline - Add to MCP config:

{
  "mcpServers": {
    "maenifold": {
      "command": "maenifold",
      "args": ["--mcp"],
      "env": {"MAENIFOLD_ROOT": "~/maenifold"}
    }
  }
}

Codex - Add to ~/.codex/config.toml:

[mcp_servers.maenifold]
type = "stdio"
command = "maenifold"
args = ["--mcp"]
startup_timeout_sec = 120
tool_timeout_sec = 600
env = { MAENIFOLD_ROOT = "~/maenifold" }

Try it: "Write a memory about our architecture decisions"

CLI Interface

Use maenifold directly in scripts, pipelines, or with non-MCP clients:

# Write a memory with WikiLinks
maenifold --tool WriteMemory --payload '{
  "title": "Architecture Decisions",
  "content": "Our [[microservices]] use [[event-sourcing]] for [[audit-trails]]"
}'

# Continue a sequential thinking session
maenifold --tool SequentialThinking --payload '{
  "sessionId": "session-1234567890",
  "response": "After analyzing the architecture...",
  "nextThoughtNeeded": true
}'

# Search memories with hybrid mode
maenifold --tool SearchMemories --payload '{
  "query": "authentication patterns",
  "mode": "Hybrid",
  "pageSize": 10
}'

MCP and CLI have full feature parity. Start a session via MCP and continue it via CLI, or vice versa. The system supports concurrent agents using the same memory location - perfect for multi-agent pipelines or parallel workflows.

Learn more

.NET 9.0 · SQLite · ONNX · MCP · MIT License

Your AI’s extended mind