// MoorAI Management Server · tech specs
MoorAI Management Server: tech specs
What the console sets and what it shows. The short version is on the MoorAI Management Server page.
More: Sample report · Content-free data lineage · AI-BOM for the AI Act · How it works · Workflow
// policy and fleet
What the console sets, and what it shows
Scroll sideways →
| View | What it does |
|---|---|
| Policy: business actions | Sets money movement, bulk data export, identity and access changes, destructive changes and hand-offs to other agents to allow, alert, ask or block, plus rules for functions by name. Read by the model proxy. |
| Policy: when MoorAI can’t decide | Fail open or fail closed, for all surfaces or per surface: the coding-agent hook, the model proxy, the MCP gateway and Claude Inference hooks. Each setting is validated and signed into the policy. |
| Policy: living-off-the-AI detections | AI coding tools started by an unexpected program (off, alert or block, with an allow list), decoy AI keys on devices (opt-in), local model servers reachable without MoorAI, and unknown programs calling AI provider APIs. |
| Presets | Developer-safe, Trusted workspace, Regulated, Forensics and Industrial / OT, each previewed before it is applied. Regulated asks before money, identity and destructive changes, blocks bulk export and fails closed. Industrial / OT blocks register, coil, setpoint, controller and safety-system writes by tool name and fails closed. Both block unexpected launchers of AI CLIs. No preset places decoys. |
| Inventory and Coverage | Each device’s effective fail mode, flagged when it fails open against a closed policy; local model servers reachable without MoorAI (red when they listen beyond loopback), unknown programs calling AI APIs, and the decoy count. |
| Alerts | Living-off-the-AI alerts in plain words; a decoy key used counts as Critical and goes to the SIEM with the other Critical alerts. |
| Evidence | Alerts missing from a device, counter resets, and a run of alerts sent without a sequence number as its own low-severity finding. |
| Runtime AIBOM | What was actually used over 1 to 365 days: which agent used which model, MCP server and tool, how often and on which devices, with CSV export. A value no source measures is shown as not measured, never as zero. |
| Trace view | The span tree of an agent session across the hook, the model proxy, the MCP gateway and guard, opened from any Timeline event with a trace id. A surface that sent nothing for the trace is listed as not reported. Trace ids go to the SIEM on schema 6. |
| Granted vs used | What each agent can reach on each device (MCP servers, permission settings, built-in tool classes, AI keys at rest) against what it used over 30 or 90 days, with recommendations and an apply-as-policy that is previewed, never widens a list, and is logged. |
| Agent ownership | Owner, purpose and approval for each agent, MCP server and AI key, with Ownerless and stale findings where use is measured. |
| Context tab and patient zero | Each session’s content-free context leaves, every record’s root re-verified (verified, mismatch, or leaves not held), the leaf behind an injection alert, and every session that took in a file, found by its SHA-256. A sub-agent’s records are verified against its own tree, linked to the parent session. |
// what the plan includes
Everything in MoorAI Management Server
Fleet governance for security and compliance teams, hosted at app.moorai.dev or as a private deployment in your own environment — source-available under the Elastic License 2.0. Self-hosted: free, no user limit. Hosted by MoorAI: contact us for pricing. Self-hosted: free, no user limit. Hosted by MoorAI: contact us. Pricing →
- Everything in MoorAI for Agents, across the whole fleet
- Enforcement — block, require sign-off, redact output, end the session on a critical threat, per signed policy (built-in defaults when none is set)
- Fleet governance — policy, tenants, and per-device control from one console
- MCP approval — discovered servers approved or denied, with tool changes re-approved per tool
- Policy presets — Developer-safe, Trusted workspace, Regulated, Forensics and Industrial / OT, previewed before they apply
- Fail mode per surface — fail open or fail closed when MoorAI can’t decide, with each device’s effective mode in Inventory
- Living-off-the-AI detections — AI CLIs started by an unexpected program, opt-in decoy AI keys, exposed local model servers and unknown AI callers
- Runtime AIBOM & right-sizing — which agent used which model, MCP server and tool, granted vs used per device, and a previewed apply-as-policy
- Trace and context evidence — one trace per agent session, a context tree per session, and patient-zero lookup by file hash
- Agent ownership — owner, purpose and approval per agent, MCP server and AI key, with ownerless and stale findings
- Capture tiers — opt-in, admin-enabled content capture for incident response
- UBA & assurance — trust scoring, anomaly signals, board AI-readiness report
- Compliance mapping and a signed evidence pack — OWASP LLM Top 10, MITRE ATLAS, NIST AI RMF, ISO 42001, EU AI Act records
- Alerts · SIEM streaming · AIBOM export · SSO (SAML / OIDC)
See your fleet,
not your prompts.
The MoorAI Management Server is free self-hosted, with no user limit; hosted by MoorAI, contact us. Enrolled devices enforce: block, sign-off, session kill.