Metanet and Skelter Labs Partner on Enterprise AI Controls

Metanet and Skelter Labs are collaborating to address permission management and security issues that arise when companies use AI agents in their work. Metanet Group’s global consulting firm AT Kearney will set clients’ AI transformation (AX) strategies and initiatives, while Skelter Labs will develop the core AI engine and platform. Metanet Digital and Metanet Global will handle cloud infrastructure and systems integration, providing services from planning and development through delivery.

Enterprise agents running under shared accounts can access information beyond a user’s permissions or operate systems incorrectly. Skelter Labs’ enterprise AI agent orchestration platform, BELLA, manages agent creation, deployment and collaboration in a controlled environment. It has four layers: identity, policy, AI gateway and assets. The identity layer connects to the company’s single sign-on and identity provider systems (SSO/IDP) to identify end users. The policy layer uses user information to apply access control lists (ACLs) to documents, tools and agents.

The AI gateway handles calls to large language models (LLMs), including in-house models and external APIs, as well as calls between agents. It applies cost tracking and budget limits by user and department, model routing and security guardrails. The asset layer registers document search indexes, MCP tools, workflows, agents, prompts and skills, and manages sharing and isolation across the company, teams and individuals.

BELLA also uses identity-based access controls and a four-stage defense system. At the data and retrieval-augmented generation (RAG) integrity stage, it prevents unauthorized changes to data chunks and filters prompt injections and hidden characters when documents are registered. Input and output guardrails block prompt injections and mask Korean personal-information patterns, API keys, internal IP addresses and internal domains. Agent behavior controls require human approval (HITL) based on risk and run code in isolated sandboxes to limit infinite loops and unauthorized tool use. Gateway resource protection sets token and request limits by user and key, and detects and blocks abnormal consumption or resource exhaustion attacks.

BELLA offers three environments tailored to user roles. The AI portal for employees supports natural-language questions and document review, while AI Studio provides a low-code builder for non-developers and a pro-code builder for developers. The administrator portal centrally manages usage, budget limits and guardrails. The platform is designed to let teams safely reuse and connect verified AI assets across departments.

Metanet and Skelter Labs have also worked on an enterprise-wide AI agent platform for a large Korean company. They established a shared AI foundation to support agent development and operation across the organization, bringing infrastructure, agent execution, data and models, security, and operations into one platform. Each function can be replaced or expanded independently, so only the needed parts can change when work requirements or security policies shift. The AI portal and agent builder were connected to existing employee login and permission systems, which restrict data access and work according to each user’s permissions. The platform also keeps execution logs that cannot be edited or deleted afterward, allowing issues to be traced. The companies provided an AI adoption roadmap, an operational handover environment and role-based training so the company can develop and operate the platform itself.

Skelter Labs CTO Jonghun Seo said control, rather than model performance, is the bottleneck to adopting AI agents. When teams directly call different LLMs, he said, it is difficult to identify where costs arise, and differing guardrails make it hard to maintain consistent controls across a company. He said the focus had previously been on training foundation models and improving performance, while the agent era makes it important to control validated open models flexibly and safely in enterprise environments.

According to Skelter Labs, BELLA follows open standards, can be deployed on premises or in a private cloud using Kubernetes, and is not tied to a particular model or cloud provider. Seo said platforms must absorb change because models and tools change quickly, and emphasized the importance of operating corporate AI assets safely through open standards.

Metanet and Skelter Labs are collaborating to address permission management and security issues that arise when companies use AI agents in their work. Metanet Group’s global consulting firm AT Kearney will set clients’ AI transformation (AX) strategies and initiatives, while Skelter Labs will develop the core AI engine and platform. Metanet Digital and Metanet Global will handle cloud infrastructure and systems integration, providing services from planning and development through delivery.

Enterprise agents running under shared accounts can access information beyond a user’s permissions or operate systems incorrectly. Skelter Labs’ enterprise AI agent orchestration platform, BELLA, manages agent creation, deployment and collaboration in a controlled environment. It has four layers: identity, policy, AI gateway and assets. The identity layer connects to the company’s single sign-on and identity provider systems (SSO/IDP) to identify end users. The policy layer uses user information to apply access control lists (ACLs) to documents, tools and agents.

The AI gateway handles calls to large language models (LLMs), including in-house models and external APIs, as well as calls between agents. It applies cost tracking and budget limits by user and department, model routing and security guardrails. The asset layer registers document search indexes, MCP tools, workflows, agents, prompts and skills, and manages sharing and isolation across the company, teams and individuals.

BELLA also uses identity-based access controls and a four-stage defense system. At the data and retrieval-augmented generation (RAG) integrity stage, it prevents unauthorized changes to data chunks and filters prompt injections and hidden characters when documents are registered. Input and output guardrails block prompt injections and mask Korean personal-information patterns, API keys, internal IP addresses and internal domains. Agent behavior controls require human approval (HITL) based on risk and run code in isolated sandboxes to limit infinite loops and unauthorized tool use. Gateway resource protection sets token and request limits by user and key, and detects and blocks abnormal consumption or resource exhaustion attacks.

BELLA offers three environments tailored to user roles. The AI portal for employees supports natural-language questions and document review, while AI Studio provides a low-code builder for non-developers and a pro-code builder for developers. The administrator portal centrally manages usage, budget limits and guardrails. The platform is designed to let teams safely reuse and connect verified AI assets across departments.

Metanet and Skelter Labs have also worked on an enterprise-wide AI agent platform for a large Korean company. They established a shared AI foundation to support agent development and operation across the organization, bringing infrastructure, agent execution, data and models, security, and operations into one platform. Each function can be replaced or expanded independently, so only the needed parts can change when work requirements or security policies shift. The AI portal and agent builder were connected to existing employee login and permission systems, which restrict data access and work according to each user’s permissions. The platform also keeps execution logs that cannot be edited or deleted afterward, allowing issues to be traced. The companies provided an AI adoption roadmap, an operational handover environment and role-based training so the company can develop and operate the platform itself.

Skelter Labs CTO Jonghun Seo said control, rather than model performance, is the bottleneck to adopting AI agents. When teams directly call different LLMs, he said, it is difficult to identify where costs arise, and differing guardrails make it hard to maintain consistent controls across a company. He said the focus had previously been on training foundation models and improving performance, while the agent era makes it important to control validated open models flexibly and safely in enterprise environments.

According to Skelter Labs, BELLA follows open standards, can be deployed on premises or in a private cloud using Kubernetes, and is not tied to a particular model or cloud provider. Seo said platforms must absorb change because models and tools change quickly, and emphasized the importance of operating corporate AI assets safely through open standards.