Skip to main content

Admin

The admin console is where a tenant's administrators manage users, agents, records and security (its title on screen is “AGENTIC STAR ADMIN”).

The guides in this section focus on security settings. For each setting they cover the items on screen, the choices, the defaults and limits, where and how it takes effect, and what users see.

The screen where you create API clients and LLM Gateway keys is a different one, the Console (Console Guide). How the two differ is set out in the Admin Console Reference.

What you can do in the admin console​

The menu is split into twelve groups. For each group, the screens in it and what you can do on each. Names follow the screen. Some screens are grouped into one menu item, and you switch between them with the tabs at the top of the screen (for example, “Records and Audit” > “Operation Logs” > the “Operation audit logs” tab). If you can open only one screen in a group, the menu shows that screen's name instead.

Dashboard​

See the state of the whole tenant (such as how many runs are in progress now and how many runs there have been today).

Users and Organizations​

ScreenWhat you can do
Users > the “Users” tabCreate user accounts (one at a time, or in bulk from CSV), approve them, revoke their approval, suspend and resume their use, unlock them, reset their passwords, delete them, and grant system administrator rights
Users > the “Admins” tabSee the list of system administrators, and remove system administrator rights (organization admins and auditors are assigned in “Admin Roles”)
Restrictions for violationsSee the users and IP addresses whose use is restricted, and lift restrictions
Organizations and Job Types > the “Organizations” tabManage the organizations assigned to users
Organizations and Job Types > the “Job Types” tabManage the job types assigned to users
Admin RolesReview organization administrators, operators and auditors, and grant or remove the roles
Ownership handoverMove projects and agents whose owner has left or been disabled to an active user in the same organization. A moved project is paused, and the new owner resumes it after checking it. Each move is recorded in the audit log

For security, see: the operations that stop and restore users, and how they differ (Sign-in and External Auth (SSO))

For SSO users, organization and job type can also be set from values received from your identity provider (Sign-in and External Auth (SSO)). What organization administrators can see is in the Roles Reference.

Sign-in and Access Sources​

ScreenWhat you can do
Sign-inDecide whether self sign-up and password reset are allowed, and how long a login lasts (15 minutes to 24 hours without activity, 1 to 30 days at most)
External AuthConnect to your identity provider for SSO with OIDC or SAML 2.0. Also decide automatic user creation, auto-approval, and importing organizations and job types. You can let people in through SSO only. There is also a SCIM tab
Access Source Restrictions > the “Allowed IPs” tabRegister, per service, the IP addresses (CIDR allowed) that may reach the admin console and the chat, and turn the restriction on. Until it is on, people can connect from anywhere
Access Source Restrictions > the “External API Allowed IPs” tabDecide the IP addresses allowed to connect to the external integration API (/api/v1). Once it is on, calls from elsewhere get 403 (ip_not_allowed)
Access Source Restrictions > the “Country Restrictions” tabDecide, per country, whether connections are blocked or only recorded (up to 10 countries per rule). It applies to sign-in, the chat and the admin console. It does not apply to the external integration API or the LLM Gateway. Takes about an hour to take effect. Detections from the last 30 days can be viewed
Outbound IPsSee the source IP addresses of traffic going out from this service. Use them when allowing this service in your firewall

For security, see: all of them (Network, Sign-in and External Auth (SSO))

AI Agents​

ScreenWhat you can do
Agent inventorySee the agents in your organization across the board, with their placement, connections, playbooks, skills and latest run. Stop and resume agents from the directory (agents not yet in the directory are added by a system administrator with “Add all to directory”). For an agent stopped by a governance service, “Stopped by” shows who stopped it. Rows for tenant chats in the directory (“Chat (tenant)”) appear to system administrators and tenant-wide auditors, and only system administrators can stop and resume them (organization administrators do not see them)
Agent governance integrationRegister AGENTIC STAR Agents in the agent registries of Microsoft Agent 365 and AWS, send traces, and apply stops made at those services
Common AI Agent InstructionsDecide the instructions passed to the AI (user prompts and prohibition instructions)
Image and Video ModelsSet the model and connection for the generative AI that makes images and videos
AI PersonaDecide the role and tone given to agents
Slide TemplatesManage the slide (PPTX) templates used to generate documents
Share requestsReview requests to share skills and playbooks, then approve, return or stop sharing

Connections and Usage Scope​

ScreenWhat you can do
MCP Servers > the “MCP Servers” tabRegister the MCP servers agents use. You can set the allowed organizations and job types, create a server from a template, choose not to give flagged tools to Agents, and set the allowed account domains
MCP Servers > the “MCP Connections” tabList and delete the connections that users and projects made to MCP server providers
A2A AgentsRegister external agents called through agent-to-agent integration
Feature permissionsDecide who can use browser, web fetch, image generation, video generation, file attachment, sharing, Agent mode and scheduled run creation (“Everyone”, “Selected organizations and job types” or “Not allowed”). The switch that stops all shared links is here too

AI Agent Restrictions​

ScreenWhat you can do
Guardrails SettingsDecide how content passed to the AI is inspected (prompt attacks, harmful content, personal information) and the execution time limit. With “DLP Integration”, you can also have your company's DLP service check content before and after an agent uses tools
Banned URLDecide destinations agents must not open
Widget Connection RestrictionsDecide which domains the small interactive screens (widgets) in agents' answers cannot connect to

For security, see: Common AI Agent Instructions, Guardrails Settings, Banned URL and Widget Connection Restrictions (Guardrails and Banned URLs)

Project Management​

ScreenWhat you can do
Cross-project KPIsSee each project's success rate, human intervention, duration, failure reasons and review request backlog over the last 7 or 30 days, and against the previous period
Execution historyFollow an agent run step by step, from the request to the outcome. Organization administrators of that organization can open the text by entering a reason (who opened it and why is recorded, and the user can see it). Enter a reason to export the record, without the text, as JSON
Execution rulesDecide how far agents may operate in project runs (Open: broad, the default; Guarded: blocks deletion; Strict: mostly read-only) and forbidden operations no project can lift. Changes take effect once you press “Publish”, from the next run that starts. Each publish is numbered v1, v2… and kept in the history. They do not apply to the chat or scheduled runs. “Definition change approval”, which makes changes to a project's definition need approval, is also switched here (system administrators only)
Automation destinationsRegister the destinations that the project step “External automation” sends to. The signing key is shown only once, at registration
Review requestsSee review requests waiting on approvers in your organization, including overdue ones, and pass the review screen URL to the approver. You cannot answer them from the admin console
Organization projectsSee your organization's projects and their current owners, and add users in your organization as members or owners

How to use them is in the Project section.

Records and Audit​

ScreenWhat you can do
Operation Logs > the “Audit Logs” tabSee and export to CSV the record of operations performed in the admin console
Operation Logs > the “Operation audit logs” tabSee what users, agents and the system carried out, such as agents' tool calls, allows and denies by execution rules, and approvals. The text itself is not kept
Operation Logs > the “User action audit” tabSee when users opened, created or deleted conversations, files, shares and so on in the chat. The text itself is not kept (system administrators and auditors)
AI Agent LLM UsageSee what agents had the AI think through (how a request was understood, guardrail decisions, a summary of the result). Personal information is masked before it is kept
Guardrail AlertsSee what guardrails stopped (prompt attacks, harmful content and so on). The text itself is not kept, and nothing that identifies the user is shown on screen or in the API. Track each alert with a status, an assignee and an outcome
User EvaluationsSee reports users sent with “Report Issue” on an AI answer (six kinds, such as misinformation, information leakage and biased answers). Counts by kind and monthly trends are shown too
Outbound Communication RecordsSee each outbound connection agents made: the destination, whether it was allowed or denied, and why it was stopped. You can show only what was stopped

For security, see: all of them (Logs and Audit)

Record Retention and Delivery​

ScreenWhat you can do
Export and Legal Hold > the “Conversation Export” tabExport users' conversations as JSON, including the text and the list of attachments. You can narrow them by update period and by user. Each export is recorded in Audit Logs (system administrators only)
Export and Legal Hold > the “Retention and Legal Hold” tabDecide the days before conversations are archived, how long archived conversations are kept, and legal hold
SIEM IntegrationSend nine kinds of records, including Audit Logs, to your SIEM (HTTP push, Microsoft Sentinel, Splunk or AWS CloudWatch Logs). Choose which kinds to send and see the delivery status
Conversation archive destinationSet where conversation records are sent (your archive). Shown only to the person appointed as the records officer

For security, see: all of them (Logs and Audit)

Announcements and Notifications​

ScreenWhat you can do
AnnouncementsCreate announcements shown in users' chat screen (Inbox). You can also announce maintenance, and stop changes in the chat during that period
Notification SettingsTurn the three kinds of email (task completion, follow-up and weekly) on or off for the whole tenant. Changes are saved immediately

LLM Gateway Management​

ScreenWhat you can do
BYOK ModelsManage the models used through the LLM Gateway with API keys registered by administrators
LLM Gateway MCPRegister the MCP servers that requests through the LLM Gateway may have the AI use. Once at least one is registered, MCP servers that are not registered are removed from the request before it is passed to the AI. With none registered, nothing is narrowed. Separate from the “MCP Servers” used by chat agents
LLM Gateway GuardrailsDecide the restrictions applied to requests through the LLM Gateway
LLM Gateway Allowed IPsDecide the IP addresses allowed to connect to the LLM Gateway
LLM Gateway LogsSee the record of requests relayed by the LLM Gateway
LLM Gateway Operation AuditSee the record of operations performed through the LLM Gateway
Client SettingsOpens the “LLM Gateway” screen of the Console (where LLM Gateway clients are created and their credentials checked; developers use it in the Console)

How to use it is in the LLM Gateway Setup Guide.

License​

Register and refresh the license key. See the contract period, how many users and projects can be registered, and current usage. If the license becomes invalid, services for users stop.

Two kinds of administrator​

A system administrator can use almost every menu. However, “Review requests” and “Organization projects” under Project Management appear only when they are also an organization administrator, and “Conversation archive destination” appears only when they have been appointed as the records officer (Admin Console Reference).

An organization administrator can use only part of it: the Dashboard, Project Management (Cross-project KPIs, Execution history, viewing Execution rules, Review requests, Organization projects), the cross-cutting agent list and share requests under AI Agents, and MCP Connections under Connections and Usage Scope. What they can see is limited to what concerns the organizations they are in charge of (Roles Reference). An auditor can view only the record screens of the admin console. An auditor assigned to an organization can see what concerns that organization, and a tenant-wide auditor can see the entire tenant. Organization administrators are set under Users and Organizations, in “Admin Roles”.

Every setting the guides in this section deal with is made by a system administrator.

Guides​

  • Sign-in and External Auth (SSO) — How users and administrators get in. Deciding on passwords, passkeys and SSO, and how long a sign-in lasts
  • Network — Where people can come in from (IPs and countries), and traffic going out from this service (source IPs, ports, destinations)
  • Guardrails and Banned URLs — How to narrow what is passed to the AI and where agents can go. When settings take effect
  • Logs and Audit — What is recorded where, and how to get it out. Sending to a SIEM, and retention

Reference​