Enzan

(name is still in progress) this is fomalization of Cognition into an available name / domain

Typed, structured, self-maintaining memory for AI agents.

Named for 演算 (enzan) — Japanese for computation. Also 遠山 — the distant mountain you can only see when you have enough memory to look back far.

Most AI memory products are flat vector stores. Enzan is different: a typed, curated, relationship-aware knowledge layer with confidence tracking, provenance, pattern recognition, and maintenance semantics built in. Your agents don't just retrieve — they reason over a cortex that gets sharper over time.

Related MCP server: Memsolus MCP Server

What makes Enzan different

Document types

• knowledge — facts, claims, concepts with confidence, source strength, and optional expiry

• skill — reusable techniques with steps, pitfalls, and source attribution

• pattern — recurring structures recognizable from signals[], with examples and counter-examples

• question — logged user queries for blindspot analysis

MCP tools

Connect via any MCP-compatible client (Claude, Cursor, Windsurf, OpenClaw, etc.):

Quickstart

# Install the Enzan MCP server
npx @sparksharе-io/enzan

# Or add to your MCP config manually:
{
  "mcpServers": {
    "enzan": {
      "command": "npx",
      "args": ["@sparksharе-io/enzan"],
      "env": {
        "ENZAN_API_KEY": "ez_your_key_here"
      }
    }
  }
}

Get your API key at enzan.ai — free tier available.

Architecture

AI Agent (Claude, GPT, etc.)
    ↓ MCP over HTTP/SSE
Enzan Gateway
    ↓ API key → tenant namespace
Azure Cosmos DB (per-tenant container)
    ↓
Azure OpenAI (embeddings)

Self-hosted

Enzan runs on any Node.js host with a Cosmos DB backend.

git clone https://github.com/SparkShare-io/enzan
cd enzan
cp .env.example .env   # fill in your Cosmos + Azure OpenAI credentials
npm install
npm start

Roadmap

See ROADMAP.md.

License

MIT — SparkShare.io