Luumen is the governed agent platform for regulated industries and SAP operations. Teams use it to run AI agents against production infrastructure and enterprise systems with governance built in: approvals, policy, role-based access, audit log, and human-in-the-loop control before anything touches a live system.
Why Luumen Exists
Enterprise AI is moving from answering questions to executing work. Once an agent can run a command or change a system, governance has to be enforced at the moment the action runs. Most AI tools that can reach infrastructure run with the full scope of whoever connected them, with no per-action permissions and no independent record. Luumen was built to close that gap.
Luumen is built by Apiphani, a managed services company and SAP partner that operates mission-critical SAP and infrastructure estates for manufacturing, energy, public sector, aerospace, and other regulated industries. The controls in Luumen are the controls Apiphani requires before letting AI touch a system it is accountable for.
What Luumen Does
Keeps every AI action inside your rules. Every action runs under policy guardrails enforced at runtime. Permissions scope down to the individual tool call, governed actions wait for human-in-the-loop approval, and role-based access control decides who can manage people, workspaces, and credentials.
Connects your existing tools under the same governance. LuumenAI reads from and acts in the systems you already use, such as ServiceNow, Jira, and GitHub. Each integration is a governed capability with defined risk levels, approval requirements, and evidence requirements, and each is scoped to only the agents that need it.
Gets operational work done from one conversation. Ask LuumenAI to investigate an issue or run a procedure and it can work across one or many servers and systems in the same conversation. It runs Skills, such as patching, across whole groups of hosts, and it pauses for inline approval whenever policy requires it.
Grounds answers in your own documentation. Documentation, runbooks, and SOPs become AI context through selective ingestion. Teams choose which sources LuumenAI and agents can draw from, answers cite the sources they use, and knowledge can be scoped at the organization, workspace, and individual user level.
Puts every action on one record. Every governed action, including runs, tool calls, approvals, and guardrail events, lands in one queryable audit log. Owners and admins can filter it by workspace, session, tool, risk, status, approval state, and date, so audit, incident review, and compliance questions are answered from the log.
Luumen Governance Controls
Control | What This Control Means in Luumen |
Human-in-the-loop approval | Your AI policy decides which actions require an explicit human answer before they run. When LuumenAI proposes a governed action, it shows you exactly what it wants to do and waits, and a decision applies to that one action only. |
No standing approvals | There is no always allow. Luumen has no way to pre-approve a command or auto-approve a category of host action. Every command is its own decision, every time, including every step of a Skill. |
Runtime policy | Every AI action runs under defined policy guardrails, enforced at the moment it runs. Permissions scope down to the individual tool call. |
Deterministic tool blocking | An integration tool that is not enabled is excluded from the set of tools exposed to the AI. It is blocked in code, not by prompting, so the AI cannot see or attempt the action. |
User-scoped access | LuumenAI has no credentials of its own. On a host, it works inside the session you opened, with the credential you used to connect and your permissions. It cannot reach a server you could not reach yourself. |
Owner, Admin, and Member roles control what a person can change, from day-to-day host and Skill work up to user management and billing. Which servers they can reach is controlled separately, by workspace membership. | |
Each integration is enabled by an admin or owner with permissions set per tool. Each carries defined risk levels and approval requirements, and is scoped to only the agents that need it. | |
Safe defaults | Integration tools are classified read, write, or destructive, and every AI action carries a risk level. By default, read actions run automatically and write and destructive actions require human approval. |
Credentials saved to Luumen are encrypted with AES-256-GCM before they are written, and Luumen decrypts one only to make a connection for you. No person can read a secret back out, including you. A credential can also point at the SSH key already on your computer, which is never uploaded. | |
Nothing on your servers | Luumen uses the SSH or WinRM access you already authorized. No agent, no inbound access, no new firewall rules. |
Workspace separation | Workspaces separate environments, business units, or clients. Hosts, integrations, and members are scoped to their workspaces, and admin-set permissions apply to every workspace. |
On Enterprise, every governed action, including runs, tool calls, approvals, and guardrail events, is recorded append-only for the life of your organization. |
Next Steps
Download the desktop app, add your hosts, and start a conversation with LuumenAI. These articles cover the rest of the setup:
Apiphani also builds and runs custom agents around a customer's specific operations and environments, delivered on Luumen under the same approvals, access control, and audit log. Custom agents, and the SAP and Dynatrace connectors, are part of an Enterprise engagement. To learn more please contact us at [email protected].