Version: 1.0.0 Last updated: 2026-03-22
MA (Memory Architect) is NekoCore OS's built-in AI development agent. It is not an external dependency — it is a first-class subsystem that lives at project/MA/ and serves two roles:
-
Builder — MA's blueprints define the architecture for two sub-projects that can be built from scratch:
- REM System Core (26 modules, 8 layers) — the cognitive memory engine at port 3860
- NekoCore Cognitive Mind (97 modules, 5 parts) — the full cognitive personality runtime at port 3870
Starter scaffolds (PROJECT-MANIFEST.json, BUILD-ORDER.md, package.json) are included in the repo. MA builds the full implementation from the blueprints.
-
Runtime service — MA runs as a persistent background server (port 3850) that NekoCore OS entities can call for tool execution, web search, model routing, and recurring task management (pulses/chores). The OS server proxies requests to MA through the
/maslash command bridge and the Resource Manager's pulse panel.
project/
├── server/ ← NekoCore OS server (port 3847)
│ └── brain/ ← Cognitive pipeline, tasks, blueprints
├── client/ ← OS GUI (desktop shell, apps, tabs)
├── MA/ ← Memory Architect (port 3850)
│ ├── MA-Server.js ← Self-contained Node.js server
│ ├── MA-server/ ← 15 core modules (zero npm deps)
│ ├── MA-workspace/ ← Sandboxed project workspace
│ │ ├── nekocore/ ← Starter scaffold (MA builds from blueprints)
│ │ └── rem-system/ ← Starter scaffold (MA builds from blueprints)
│ ├── MA-blueprints/ ← Build instructions (tracked)
│ ├── MA-knowledge/ ← Reference docs (tracked)
│ └── MA-skills/ ← Skill definitions (tracked)
├── entities/ ← [GITIGNORED] Runtime entity data
└── memories/ ← [GITIGNORED] Runtime system memory
MA itself (the engine, config templates, blueprints, skills, knowledge base) is tracked in git. The workspace projects include starter scaffolds (manifest, build order, package.json) that are tracked — everything MA builds from the blueprints (source code, tests, node_modules) is gitignored.
The nekocore/ and rem-system/ workspace directories ship as starter scaffolds — just a PROJECT-MANIFEST.json (all statuses "not-started"), a BUILD-ORDER.md (construction guide), and a package.json. MA reads the blueprints (MA-blueprints/nekocore/ and MA-blueprints/rem-system/) and builds the full implementation module-by-module, updating the manifest as it goes.
The generated source code, contracts, tests, client code, and node_modules are all gitignored. The blueprints remain tracked and serve as the authoritative build instructions.
NekoCore OS entities can call MA through the /ma slash command:
User: /ma search for Node.js best practices
The OS server (server/services/ma-bridge.js) handles this by:
- Checking if MA is running (health check on port 3850)
- Auto-booting MA via the process manager if offline
- Sending the message to
POST /api/chaton MA - Returning MA's response to the entity's conversation
The Resource Manager app proxies pulse and chore management through the OS server:
Client → OS Server (port 3847) → MA Server (port 3850)
GET /api/resources/pulses → GET /api/pulse/status + GET /api/chores/list
POST /api/resources/pulses/chores → POST /api/chores/add
POST /api/resources/pulses/:id/toggle → POST /api/pulse/start or /api/pulse/stop
This keeps the client talking to a single origin (the OS server) while MA handles the recurring task engine internally.
The OS server can start, stop, and health-check both MA and the NekoCore Cognitive Mind server via the Process Manager routes:
| Route | Action |
|---|---|
GET /api/servers/status |
Health check all managed servers |
POST /api/servers/:id/start |
Start MA or NekoCore server |
POST /api/servers/:id/stop |
Stop a managed server |
Managed servers: rem-system (port 3860), nekocore (port 3870), and MA itself (port 3850).
- Location:
MA-workspace/rem-system/(starter scaffold tracked; built source gitignored) - Blueprint:
MA-blueprints/rem-system/(tracked) - Scope: 26 modules across 8 layers — memory storage, NLP, pipeline, entity identity, dream/brain loop, integration, transport, entity management
- Port: 3860
- Purpose: Standalone cognitive memory engine that can be embedded or run as a microservice
- Location:
MA-workspace/nekocore/(starter scaffold tracked; built source gitignored) - Blueprint:
MA-blueprints/nekocore/(tracked) - Scope: 97 modules across 5 parts — Foundation, Memory+Knowledge, Cognition Engine, Identity+Generation, Services+Transport
- Port: 3870
- Pre-requisite: REM System must be running (port 3860)
- Purpose: Full cognitive personality runtime — the reference implementation of a NekoCore entity mind as a standalone server
Both projects are built by MA following the blueprint-driven development workflow: plan → decompose → build module-by-module → test → integrate. The repo ships starter scaffolds; ask MA to build them.
MA's workspace (MA-workspace/) is sandboxed — all file operations are restricted to this directory. To create a new project:
- Write a blueprint in
MA-blueprints/describing the architecture - Start MA (
node MA-Server.jsor via Process Manager) - Give MA the build instruction:
"Build the project following the blueprint at MA-blueprints/my-project/" - MA will create the project directory, scaffold modules, write tests, and iterate
The blueprint system supports both core blueprints (universal patterns like task decomposition, error recovery) and module-specific blueprints (research, code, writing, analysis, planning).
| Document | What It Covers |
|---|---|
| ARCHITECTURE-OVERVIEW.md | Full system map, subsystem table, design principles |
| PIPELINE-AND-ORCHESTRATION.md | Cognitive pipeline, orchestrator, task fork |
| APP-FOLDER-OWNERSHIP.md | File ownership rules for apps |
| HOW-TO-CREATE-AN-APP.md | Step-by-step guide for new NekoCore OS apps |
project/MA/README.md |
MA quick start, feature list, configuration |
project/MA/USER-GUIDE.md |
Full MA user guide |