Pasar al contenido principal
No se pierda lo que está por venir en Microsoft Ignite
No se pierda lo que está por venir en Microsoft IgniteAcceda directamente a las personas, los productos y los conocimientos que hacen posible la tecnología de Microsoft, con oportunidades para hacer preguntas, desarrollar aptitudes y aprender junto a la comunidad.
Microsoft
separator
https://catalogartifact.azureedge.net/publicartifacts/sbmpn.agenticstar-45c41e73-ae5a-462b-ad74-8ea2efbbfea5/image1_logo256x256.png

AGENTIC STAR – Enterprise AI Agent Platform with Human Review

por ソフトバンク株式会社_法人事業

Put AI agents to work on business processes, with human review steps and a record of every run.

Overview

Put AI agents to work on your business processes, and keep people in charge of the decisions that matter. AGENTIC STAR is an enterprise AI agent platform that runs in your own Azure subscription. With the Advanced plan, chain agents into workflows with human-in-the-loop review, run agents on a schedule or from a webhook, and keep a record of every run with its deliverables.

Use cases

  • Customer inquiries (Advanced plan) – An agent drafts the reply, and the next step runs only after a person approves it in the inbox.
  • Scheduled checks and reports (Advanced plan) – Agents run on a schedule or when a webhook arrives, and report each run.
  • Research and documents – An agent researches a topic in chat and writes a report.

Why AGENTIC STAR

  • Runs in your Azure subscription – One Azure Managed Application deployment provisions the infrastructure and the platform. Connect your LLM endpoint after deployment.
  • People decide at the points that matter (Advanced plan) – A review step holds the work until a person approves it, sends it back, or rejects it.
  • Built-in safety – Content filtering and personal information detection with Azure AI Services.
  • 14 languages – Web UI for chat, agents, and administration.

Plans

  • AGENTIC STAR Advanced (v3.3.0 and later) – Recommended for new deployments. Adds projects for AI agents on business processes.
  • AGENTIC STAR Basic (v2.x) – For existing v2.x environments. The AI agent platform without projects; SoftBank Corp. continues to maintain v2.x with bug-fix releases.

Projects (Advanced plan)

  • Workflow projects – Place registered agents on the steps of a project and take each case through them, with review steps and waits for external input.
  • Cycle projects – Start agents on a schedule or from a webhook and report each run.
  • Execution records – Outcomes and an audit trail of every run, with deliverables kept in the library by version.
  • Inbox and notifications – Review requests and notifications in one place, with a webhook when a run ends.
  • Project roles and versions – Ownership handover, operators, delegates for absences, and restorable versions of project definitions.
  • Project API – Start workflows (with input files), fetch deliverables and execution records, and cancel runs from external systems.
  • Administration – Organization administrators manage project membership, and administrators create MCP connections, including OAuth, from the admin console.
  • Chat and devices – Queue requests, rate answers, export conversations, and sign out of other devices.

Platform (both plans)

  • AI-ready data – PostgreSQL with pgvector and the Qdrant vector database.
  • Observability – Prometheus, Grafana, Loki, Alloy, Container Insights, and Log Analytics; Datadog is optional.
  • Security – Managed identity, VNet-integrated networking with Calico network policy, Keycloak authentication, and NGINX Ingress.

What Gets Deployed (Advanced plan, v3.3.0)

  • Azure Kubernetes Service (Kubernetes 1.35, Azure CNI Overlay, two autoscaling node pools)
  • Azure Database for PostgreSQL Flexible Server 16 with pgvector (high availability optional)
  • Azure AI Services, Storage Accounts, Virtual Network, managed identity with RBAC, and Log Analytics
  • 23 Kubernetes services, including the web UI, admin console, agent executor, authentication, Qdrant, monitoring, and ingress

Prerequisites

  • Azure subscription with Owner or Contributor role
  • vCPU quota in the target region for the Standard EDSv5 family: at least 12 vCPUs, 16 recommended (3 to 4 nodes of the default Standard_E4ds_v5), plus at least the same amount of Total Regional vCPUs. If you choose another VM size, secure the same amount for that family. Azure Database for PostgreSQL does not consume this quota.
  • An LLM endpoint such as Azure OpenAI

Plan limits

  • Advanced plan: up to 5 projects per environment. Archived and deleted projects are not counted.
  • Environments deployed from the Basic plan (v2.x) cannot be updated to v3.x, and updating them to v3.x by other means, such as manually applying v3.x Helm charts, is not supported. To use projects, deploy the Advanced plan.

Support

Contact SoftBank Corp. for deployment support, documentation, and updates.

De un vistazo

https://catalogartifact.azureedge.net/publicartifacts/sbmpn.agenticstar-45c41e73-ae5a-462b-ad74-8ea2efbbfea5/image5_01aiagenthumanreviewinbox.png
https://catalogartifact.azureedge.net/publicartifacts/sbmpn.agenticstar-45c41e73-ae5a-462b-ad74-8ea2efbbfea5/image6_02aiagentworkflowaudittrail.png
https://catalogartifact.azureedge.net/publicartifacts/sbmpn.agenticstar-45c41e73-ae5a-462b-ad74-8ea2efbbfea5/image7_03aiagentprojectsbusinessprocesses.png
https://catalogartifact.azureedge.net/publicartifacts/sbmpn.agenticstar-45c41e73-ae5a-462b-ad74-8ea2efbbfea5/image4_04aiagentchatdeliverables.png
https://catalogartifact.azureedge.net/publicartifacts/sbmpn.agenticstar-45c41e73-ae5a-462b-ad74-8ea2efbbfea5/image8_05aiagentadminconsole.png
Español (Chile)
Icono de exclusión de opciones de privacidad Tus opciones de privacidad
Privacidad de la salud del consumidor Sitemap Contact Us Privacy & Cookies Terms of Use About our ads Manage cookies