Skip to main content

Settings Guide

This page describes the settings available in the AGENTIC STAR marketplace edition admin console.

Accessing the Admin Console​

After deployment completes, open the admin console URL in your browser and sign in using the administrator email and password you provided at purchase.

Admin console dashboard

note

A self-signed certificate is shipped by default, so your browser will show a certificate warning. Allow the exception to proceed. To configure a proper SSL certificate, see the Domain Configuration Guide.

The admin console is organized as follows:

CategoryMenuDescription
—DashboardOverview of usage and system resources
Users and OrganizationsUsers > the “Users” tabManage end users
Users > the “Admins” tabManage admins who can access the admin console
Restrictions for violationsList of users and IP addresses whose use is restricted
Service AccountsManage service accounts for API use
Organizations and Job Types > the “Organizations” tabManage organization information
Organizations and Job Types > the “Job Types” tabManage the job-type master
Admin RolesAssign roles such as organization administrator
Ownership handoverHand over projects and agents whose owner has left or been disabled (system administrators only)
Sign-in and Access SourcesSign-inWhether self sign-up and password reset are allowed, and how long a login lasts
External AuthSSO with your identity provider
Allowed IPsThe IP addresses that may reach the admin console and the chat
AI AgentsAgent inventoryList of registered agents, where they are placed, and their latest runs
Generation Model ConnectionsConfigure LLM providers and models
Marketplace AI AgentsRegister and manage custom agents. The contract and certification used in project workflows are also registered here
Connections and Usage ScopeMCP Servers > the “MCP Servers” tabManage MCP servers
MCP Servers > the “MCP Connections” tabList of the connections users and projects made to MCP server providers
A2A AgentsManage the external A2A agents you integrate with
Feature permissionsWho can use each feature (“Everyone”, “Selected organizations and job types” or “Not allowed”), and the switch that stops all shared links
AI Agent RestrictionsGuardrails SettingsContent Safety and PII masking
Banned URLManage URLs that agents are not allowed to access
Widget Connection RestrictionsManage the domains widgets are blocked from connecting to
Project ManagementCross-project KPIsKPIs across projects
Execution historyList of project runs
Execution rulesExecution rules for the whole tenant. Whether definition changes need approval is also decided here (only system administrators can change them; organization administrators can only view)
Review requestsStatus of the organization's review requests (organization administrators only)
Organization projectsProjects that belong to the organization (organization administrators only)
Records and AuditOperation Logs > the “Audit Logs” tabHistory of admin operations
Operation Logs > the “Operation audit logs” tabRecords of operations performed by users, agents and the system
Operation Logs > the “User action audit” tabRecords of operations users performed on the screens (system administrators and auditors)
AI Agent LLM UsageHistory of LLM usage
Guardrail AlertsList of events detected by guardrails
User EvaluationsReview user reports
Outbound Communication RecordsRecords of outbound traffic from agents
Record Retention and DeliveryExport and Legal Hold > the “Conversation Export” tabExport conversations (system administrators only)
Export and Legal Hold > the “Retention and Legal Hold” tabDays until conversations are archived, how long archived conversations are kept, and legal hold
SIEM IntegrationSend records to your SIEM
Conversation archive destinationWhere conversation records are sent (only for the person appointed as the records officer)
Announcements and NotificationsAnnouncementsManage announcements for end users
Notification Settings > the “Notification Settings” tabConfigure notification email sending
Notification Settings > the “Mail Provider Settings” tabConfigure how emails are sent and the sender
—Version ManagementCheck the platform version and perform upgrades

Some screens are grouped into one menu item and switched with tabs at the top of the screen (for example, “Records and Audit” > “Operation Logs” > the “Audit Logs” tab). A person who can open only one of the grouped screens sees that screen's name instead.

For how to use projects, see Project.

Users and Organizations​

Users​

The “Users” screen switches between the “Users” and “Admins” tabs at the top.

“Users” tab​

Manage end users of AGENTIC STAR. You can add, edit, and disable users.

“Admins” tab​

Manage administrator accounts that can access the admin console. The initial administrator is auto-created at purchase time.

Service Accounts​

Create and manage service accounts for API access. Creating a service account issues a client ID and client secret used for API authentication.

For how to use the API, see the Marketplace API Quickstart.

Organizations and Job Types​

The “Organizations and Job Types” screen switches between the “Organizations” and “Job Types” tabs at the top.

“Organizations” tab​

Manage organization information.

“Job Types” tab​

Manage the job-type master.

AI Agents​

Generation Model Connections​

Configure the LLM provider and model used by agents.

LLM settings screen

SettingDescription
Agent TypeThe agent type the setting applies to
ModelModel name (LiteLLM format: provider/model-name)
API KeyAPI key for the LLM provider
Base URLEndpoint URL for the LLM provider
API VersionAPI version (YYYY-MM-DD format)
Additional SettingsExtra parameters in JSON format
note

Sensitive values such as API keys are displayed as ******** after they are saved. If you do not change them, the existing values are retained.

Marketplace AI Agents​

Register custom agents to make them available on the platform.

Basic Information​

SettingDescription
Agent IDAgent identifier (lowercase alphanumeric and hyphens, up to 63 characters)
Display NameDisplay name of the agent
DescriptionDescription of the agent
IconUpload an icon image (PNG / JPEG, 512 KB or less)
TagsTags for classification (comma-separated)

Docker Image​

SettingDescription
RegistryContainer registry URL (e.g., myacr.azurecr.io)
Image NameImage name (e.g., my-org/rag-agent)
Image TagImage tag (default: latest)
Image Pull SecretAuthentication secret for a private registry

Check image​

To see whether this cluster can pull the image you entered, use "Check image" in the Docker image section. It pulls the image from the registry with the selected Secret, in the same namespace and with the same security settings as when the agent runs, and checks that the container gets as far as startup preparation. It does not run the agent's code, so it does not check whether the agent's processing works. The result is not saved.

ResultWhat to do
The image was pulled and a digest is shownYou can use the displayed digest as the certified digest when you use the agent in workflows (certification information)
The image was pulled, but the digest could not be obtainedPass the "Checked reference" shown in the result to docker buildx imagetools inspect to look up the digest
The image was not foundCheck the registry, image name, and tag (the certified digest when you checked with a digest)
The registry refused the requestFor a private registry, check the Secret's credentials and that the Secret is selected
The image was pulled but cannot start (it may run as root, or USER may not be a numeric UID)Set USER in the Dockerfile to a numeric non-root UID (for example USER 1000) (Deploy Guide — Step 1)
Startup preparation stoppedCheck the image's USER and the security settings
The image could not be pulledCheck that this cluster can reach the registry
The pull did not finish in timeIf the image is large, check again. If it keeps happening, check the route to the registry
The image reference is not in a valid formatCheck how the registry, image name, and tag (or digest) are written
Could not connect to the check service, another check has not finished, or the check result is no longer availableCheck again after a while

Resources​

SettingDescription
CPU Request / LimitCPU allocation (e.g., 500m, 2000m)
Memory Request / LimitMemory allocation (e.g., 512Mi, 4Gi)

Storage​

Enable this when persistent storage is required.

SettingDescription
Storage SizeCapacity (e.g., 10Gi)
Mount PathMount path inside the container (absolute path)
Storage ClassKubernetes storage class
Access ModeAccess mode

Status​

SettingDescription
EnabledToggle the agent on or off

Environment Variables​

You can set custom environment variables. The following reserved variables are injected automatically by the platform and do not need to be configured manually (registering a variable with the same name results in an error).

Always injected

  • EXECUTION_ID
  • CONVERSATION_ID
  • USER_ID
  • AGENT_ID
  • REQUEST_SOURCE

Injected conditionally

  • MESSAGE_ID
  • USER_ORGANIZATION_ID
  • USER_ORGANIZATION_LABEL
  • USER_JOB_ROLE
  • USER_JOB_LABEL
  • USER_BIO
  • USER_LANGUAGE
  • PROJECT_ID (when the agent runs on a step of a project's cycle)
  • ASTER_NOTIFY_TOKEN (the notify token, issued per execution)

Custom environment variables can be marked as secret (with masked display) and as required.

Use in workflows​

This setting lets the agent be placed in the steps of a project workflow. When you turn it on, you register the contract, which decides the settings, files, outputs and connections the agent receives, together with the certification information. If you save with it off, the registered contract is deleted and the agent can no longer be used in workflows.

You do not need this setting if you use the agent only in cycle steps or in conversations.

SettingDescription
ReadinessCertified / Not certified (unavailable for workflows) / Not supported
Effect modeRead only / Idempotent write / Non-idempotent write. Only Read only and Idempotent write runs can be retried automatically after a failure. Agents that write externally must choose the value that matches their writes
Execution timeout (seconds)Defaults to 900. 0 means no limit (up to 86,400 seconds = 24 hours)
SDK version / Adapter version / Test results reference / Certified image digest / Supported run rulesCertification information (see below)
Connection slotsSlots for the connections the agent uses. Set the slot name, the allowed connections (MCP servers or A2A agents with OAuth configured) and whether the slot is required
Contract (JSON)Edit the settings (parameters_schema), default values (defaults), environment variable bindings (env_bindings), files (files) and outputs (outputs)
Load from YAMLLoad a contract written in YAML

When you turn it on, a contract template is filled in. After editing, check it with Validate and format before saving. For how to write the contract, see the Workflow Guide.

Certification information

Only agents that have all five of the following can be placed in project steps. If even one is missing, the project screen shows the agent as "Unavailable (not certified)" and it cannot be placed.

ItemContract keyValue
SDK versioncertification.sdk_versionSDK version of the tested image
Adapter versioncertification.adapter_versionVersion of the tested adapter
Test results referencecertification.tests_refLocation or ID where the test results can be found
Certified image digestcertification.image_digestsha256: followed by 64 hexadecimal digits. Runs use the image with this digest
Supported run rulescertification.supported_policiesThe basic rules the agent supports. Choose the ones confirmed in testing, from Open, Guarded and Strict

When all five are present, the readiness becomes "Certified". When items are missing, the readiness shows which ones.

If the basic rule of the project's execution rules is not in "Supported run rules", the agent does not start in that project (Execution Rules and Traffic Guide).

Certify again after you update the image. Runs use the image with the certified digest, so replacing only the tag does not make runs use the new image. You can look up the digest with, for example, docker buildx imagetools inspect <registry>/<image>:<tag>.

When "Use in workflows" is on, "Use as certified digest" in the result of "Check image" puts the digest of the pulled image into the certified digest field. "Check with this digest" next to the certified digest field checks the image with that digest in the same way. If the image for the tag differs from the certified digest, a message says so.

Additional Settings (JSON)​

Specify agent-specific extra parameters in JSON format.

ACR Secret Management​

When using a private Azure Container Registry, you can have an authentication secret created automatically from a username and password.

AI Agent Restrictions​

Guardrails Settings​

Configure content safety and personal information protection for AI outputs.

Guardrail settings screen

Content Safety​

Prompt Shield Detection

Detects and blocks jailbreak attacks. Can be turned on or off.

Moderation Detection

Sets the detection level for harmful content.

LevelDescription
0Disabled
2Recommended (detect moderately harmful content)
4Lenient (detect only severe content)
6Minimum (detect only extremely severe content)

Detection categories: violence, sexual content, hate speech, self-harm.

PII Masking​

Automatically detects and masks personally identifiable information (PII). Up to 189 categories are supported (the count varies by cloud provider).

Examples of detection targets:

  • Identifying info (phone numbers, email addresses, My Number, etc.)
  • Financial info (credit card numbers, account numbers, etc.)
  • Medical info (insurance numbers, medical records, etc.)
  • Location info (addresses, GPS coordinates, etc.)
  • Credentials (passwords, API keys, tokens, etc.)
Backend per Cloud Provider​
ProviderBackendSupported Languages
AzureAzure AI Language - PII DetectionMultilingual (including Japanese)
AWSBedrock Guardrails sensitiveInformationPolicyMultilingual (including Japanese)
GCPCloud DLPMultilingual (including Japanese)

On AWS, we recommend handling ContentSafety (content moderation / prompt shield) and PII detection in a single Bedrock Guardrail. Both are evaluated with one apply_guardrail API call, so the only IAM permission required is bedrock:ApplyGuardrail.

Banned URL​

Manage URLs that agents are forbidden to access.

Restriction Granularity​

FeatureGranularityDescription
Base feature (shipped by default)DomainOutbound traffic from the Pod is checked by ExtAuth Service (Envoy sidecar-based authorization). Because HTTPS hides the path inside the TLS tunnel, in practice this acts as a domain (+ port) level restriction.
Extension (optional implementation)URL (path)If the agent itself uses the SDK's ConfigAccess.get_banned_urls() and inspects URL strings inside its tools, path-level restriction becomes possible.
tip

Both features share the same “Banned URL” configuration (the banned_urls table), so registration in the admin console is unified. For how to implement the extension, see ConfigAccess.get_banned_urls() in the SDK Reference.

Announcements and Notifications​

Announcements​

Create and manage announcements shown to end users. You can set a title and body (Markdown), priority, and enabled / disabled state.

Notification Settings​

The “Notification Settings” screen switches between the “Notification Settings” and “Mail Provider Settings” tabs at the top.

“Notification Settings” tab​

Configure whether notification emails are sent to end users. Task-completion mail, follow-up mail, and Weekly mail can be toggled individually.

“Mail Provider Settings” tab​

Configure the email provider used to send notification emails. Review request and review-overdue emails for projects are also sent with the sending method and sender configured here. If it is not configured, these emails are not sent.

Basic Settings​

Toggle the email feature on or off.

SMTP Settings​
SettingDescriptionNotes
HostSMTP server hostnameRequired (e.g., smtp.example.com)
PortSMTP server portRequired (1–65535, default: 587)
EncryptionEncryption for the connectionNone / SSL / TLS (TLS recommended)
AuthenticationAuthentication methodNone / LOGIN / PLAIN / CRAM-MD5
UsernameSMTP authentication usernameRequired
PasswordSMTP authentication passwordRequired
From AddressSender email addressRequired
Sender display nameName shown as the email senderOptional

Records and Audit​

Operation Logs​

The “Operation Logs” screen switches between the “Audit Logs”, “Operation audit logs” and “User action audit” tabs at the top.

“Audit Logs” tab​

Review the history of operations performed in the admin console — who changed what and when.

AI Agent LLM Usage​

Review the history of LLM usage. Token consumption and request content can be traced.

User Evaluations​

Review user reports.

Outbound Communication Records​

Review the outbound traffic agents made: the destination, whether it was allowed or denied, and why it was stopped.

Version Management​

Check the platform's current version and upgrade to the latest available version. Versions must be upgraded one at a time, in order.