IBM Storage Scale AI Agents

AI agent for IBM Storage Scale administration built on BeeAI AgentStack.

Overview

Scale Agents provides an intelligent, conversational interface to IBM Storage Scale cluster management. It uses a single unified agent that routes natural language requests to specialized internal handlers, each with access to specific MCP (Model Context Protocol) tools.

Architecture

┌─────────────────────────────────────────────────────────────────────────────┐
│                         AGENTSTACK PLATFORM LAYER                           │
│  ┌─────────────────────────────────────────────────────────────────────┐   │
│  │                     SCALE_AGENT (Single Entry Point)                 │   │
│  │  - Intent classification (pattern or LLM)                           │   │
│  │  - Internal routing to specialized handlers                         │   │
│  └─────────────────────────────────────────────────────────────────────┘   │
│                                    │                                        │
│         ┌──────────────┬───────────┼───────────┬──────────────┐            │
│         ▼              ▼           ▼           ▼              ▼            │
│  ┌────────────┐ ┌────────────┐ ┌────────────┐ ┌────────────┐ ┌──────────┐ │
│  │   Health   │ │  Storage   │ │   Quota    │ │Performance │ │  Admin   │ │
│  │  Handler   │ │  Handler   │ │  Handler   │ │  Handler   │ │ Handler  │ │
│  │ (internal) │ │ (internal) │ │ (internal) │ │ (internal) │ │(internal)│ │
│  └────────────┘ └────────────┘ └────────────┘ └────────────┘ └──────────┘ │
│         │              │           │           │              │            │
│  ┌──────┴──────────────┴───────────┴───────────┴──────────────┴──────┐    │
│  │                     MCP TOOL INTEGRATION LAYER                     │    │
│  │            (Optional: BeeAI Framework RequirementAgent)            │    │
│  └────────────────────────────────────────────────────────────────────┘    │
└────────────────────────────────────────────────────────────────────────────┘
                                     │
                                     ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│                        SCALE-MCP-SERVER (FastMCP)                           │
│  HTTP Transport @ :8000/mcp                                                 │
└─────────────────────────────────────────────────────────────────────────────┘
                                     │
                                     ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│                    IBM STORAGE SCALE REST API (v2/v3)                        │
└─────────────────────────────────────────────────────────────────────────────┘

Internal Handlers

The single scale_agent automatically routes requests to specialized internal handlers:

Handler	Purpose	Example Queries
Health	Monitoring, diagnostics, alerting	"Show cluster health", "List node events"
Storage	Filesystem/fileset lifecycle	"List filesystems", "Create fileset data01"
Quota	Capacity governance	"Check quota for project01", "Set quota limit"
Performance	Bottleneck analysis	"Analyze performance", "Show throughput metrics"
Admin	Cluster topology, snapshots	"Create snapshot", "Add remote cluster"

Prerequisites

• Python 3.12+
• uv package manager
• Running instance of scale-mcp-server
• (Optional) Ollama with a function calling model for LLM reasoning

Installation

cd scale-agents
uv venv
source .venv/bin/activate

# Basic installation (pattern based routing)
uv pip install -e .

# With LLM reasoning support
uv pip install -e ".[llm]"

# Full development setup
uv pip install -e ".[all]"

Configuration

Scale Agents uses a hierarchical configuration system:

1. config.yaml (primary configuration file)
2. Environment variables (override YAML settings)
3. Default values

Configuration File

Copy the template and customize:

cp config.yaml.template config.yaml

Edit config.yaml:

# MCP Server Connection
mcp:
  server_url: "http://localhost:8000/mcp"

# LLM Configuration (optional)
llm:
  enabled: false
  provider: "ollama"
  model: "qwen3:30b-a3b"
  base_url: "http://localhost:11434"

# Server Configuration
server:
  host: "0.0.0.0"
  port: 8080

Environment Variables

Override any setting with environment variables:

export SCALE_AGENTS_MCP_SERVER_URL="http://scale-mcp:8000/mcp"
export SCALE_AGENTS_LLM_ENABLED="true"
export SCALE_AGENTS_LLM_PROVIDER="ollama"
export SCALE_AGENTS_LLM_MODEL="qwen3:30b-a3b"

Running

Standalone Server

# Using the module
python -m scale_agents.server

# Or using the run function
python -c "from scale_agents import run; run()"

With AgentStack

# Add to AgentStack
agentstack add /path/to/scale-agents

# List registered agents
agentstack list

# Run the agent
agentstack run scale_agent "Show cluster health"

Docker

# Build
docker build -t scale-agents:latest .

# Run
docker run -p 8080:8080 \
  -e SCALE_AGENTS_MCP_SERVER_URL=http://scale-mcp:8000/mcp \
  scale-agents:latest

Usage Examples

User: Show cluster health
Agent: [Routes to Health handler] -> Returns health overview

User: List all filesystems
Agent: [Routes to Storage handler] -> Returns filesystem list

User: What's the quota for fileset data01?
Agent: [Routes to Quota handler] -> Returns quota information

User: Analyze performance bottlenecks
Agent: [Routes to Performance handler] -> Returns performance analysis

User: Create a snapshot of fs01
Agent: [Routes to Admin handler] -> Creates snapshot

AgentStack Architecture Note

AgentStack SDK only supports one agent per server instance. Scale Agents implements this constraint by exposing a single scale_agent that internally routes to specialized handlers based on intent classification.

License

Apache 2.0