STAROps is a full-domain intelligent O&M platform developed by Alibaba Cloud based on large language models and AI agent technologies. It deeply integrates cross-domain observability data with large language model reasoning to overcome the limitations of traditional O&M tools, such as steep learning curves and severe data silos. You can define goals in natural language, and O&M agents autonomously complete a closed loop of dynamic planning, secure execution, and result verification. STAROps fundamentally transforms O&M from passive response to proactive autonomy, providing enterprises with 24/7 autonomous O&M capabilities to safeguard the resilience of business systems in real time.
What does STAR stand for?
STAR in STAROps represents the following four core design principles:
-
S — Sense: Connect cross-domain data such as logs, metrics, traces, events, changes, and code through connectors. Use the observability model (UModel) to maintain entities, relationships, and dynamic topologies, providing unified O&M context for queries and diagnostics.
-
T — Target: Shift from alert response to goal achievement. Define O&M goals in natural language, continuously evaluate deviations from the goals, identify influencing factors, and drive decisions and execution.
-
A — Autonomy: Digital Employees collaborate to complete the sense-decide-execute-verify loop and provide continuous 24/7 operations. Queries and analyses can proceed automatically based on policies. High-risk operations require Human-in-the-Loop (HIL) confirmation, and operations that match deny rules are blocked.
-
R — Resilience: Use Missions to continuously inspect systems and proactively identify issues. When a fault occurs, advance investigation, remediation, and verification based on execution policies to reduce mean time to recovery (MTTR). Execution traces and verification results can be retained as references for subsequent decisions.
Core capabilities
AI Chat
Use natural language to quickly analyze alerts, query data, interpret metrics, and query and analyze logs. AI Chat turns complex command-line operations into immediate insights and significantly lowers the barrier to O&M. You can start a conversation with the intelligent assistant from multiple entry points, including STAROps, CloudMonitor, and Simple Log Service (SLS).
Missions
A Mission is a human-machine interaction mode designed for long-cycle asynchronous execution. After you align on a goal, STAROps can plan and autonomously execute an asynchronous O&M plan that spans days or months based on schedules, events, and other mechanisms. Missions convert repetitive manual intervention into reliable automated processes and significantly improve O&M efficiency. Missions support scheduled, manual, and event-based triggering. A built-in Human-in-the-Loop (HIL) mechanism ensures the safety of high-risk operations.
Digital Employees
A Digital Employee (SRE Agent) is the intelligent execution entity of STAROps. You can build enterprise-specific SRE agents by customizing responsibilities, permissions, tools, and skills to create O&M assistants tailored to your business scenarios. This significantly reduces customization costs and accelerates enterprise R&D and O&M productivity. A Digital Employee is both the counterpart in AI Chat and the executor of Missions.
Intelligent access
STAROps automatically discovers cloud resources and connects logs, metrics, traces, and other data. You can also describe your requirements in natural language, and a Digital Employee matches the appropriate data sources and access solutions. After data is connected, STAROps automatically parses and aggregates the data so that it can be directly used for queries and diagnostics, significantly lowering the barrier to data access.
Unified connection and modeling
Connect data from separate systems, including monitoring data, logs, resources, changes, and code, and automatically discover and associate relationships among applications, services, and resources. During troubleshooting, you do not need to switch among multiple systems. A Digital Employee can analyze data, troubleshoot issues, and perform O&M operations in the same context, eliminating data silos.
Continuous issue discovery and closed-loop handling
STAROps continuously inspects systems based on configured O&M goals, proactively identifies exceptions, automatically completes investigations and analyses, and provides remediation suggestions. This drives issues through a complete closed loop from discovery to recovery verification. Discovery, investigation, execution, and verification are continuously tracked within the same issue. High-risk operations require Human-in-the-Loop (HIL) confirmation before a Digital Employee executes them.
Multi-cloud and on-premises support
Through direct cloud connections and environment proxies, STAROps can connect O&M data from Alibaba Cloud, enterprise data centers, other clouds, and internal networks. Cross-account access is performed based on the specified role and its permission scope, allowing you to centrally manage multiple environments on the same platform.
Core strengths
|
Benefit |
Description |
|
Unified O&M context |
Connect logs, metrics, traces, changes, code, and other data. Complete queries and troubleshooting within the same context without switching among systems. |
|
O&M digital twin |
Use UModel to clarify the relationships among applications, services, resources, alerts, and changes, helping you quickly determine the scope and possible causes of an issue. |
|
Cross-domain analysis and diagnostics |
Automatically filter, compare, and correlate data across exception detection, log analysis, trace analysis, and change backtracking to accelerate root cause identification. |
|
Flexible integration solutions |
Provides OpenAPI, page integration, IM integration for DingTalk, Lark, and WeCom, as well as Skill and CLI solutions, allowing STAROps to fit into existing workflows. |
Security and compliance
-
Fine-grained authorization policies: Authorize users and Digital Employees separately to control what users can do and what Digital Employees can access, reducing the risk of excessive permissions.
-
Human intervention: High-risk write operations require your confirmation. Disallowed operations are blocked to prevent accidental operations.
-
Agent behavior auditing: Conversations, tool calls, executed commands, and data access are recorded for post-event review and compliance audits.
-
End-to-end data encryption: Data in transit is encrypted by using HTTPS/TLS. Observability data can be encrypted at rest by using Key Management Service (KMS), and runtime artifacts are also encrypted.
Common scenarios
-
Scheduled intelligent inspection of Kubernetes clusters: Automatically inspect cluster health every day, generate structured reports, and compare historical differences.
-
High-availability assurance for core services: Continuously monitor core services and automatically perform root cause analysis when an alert is triggered.
-
Natural language-driven fault diagnosis: Gradually narrow down the troubleshooting scope through multi-turn conversations and correlate relevant data with topology information.
-
Closed-loop handling of release and change exceptions: Automatically correlate change records and follow up on exceptions that occur after a release or configuration change.
-
Periodic data quality checks: Regularly inspect data pipeline health and automatically send notifications when exceptions are detected.
-
Automated O&M report generation: Automatically aggregate O&M data on a weekly or monthly basis and generate structured reports.
-
Cost analysis and optimization: Identify cost changes and usage spikes, and obtain cost allocation statistics and optimization suggestions.
FAQ
Are the session records, Digital Employees, and other resources of a RAM user isolated from those of the Alibaba Cloud account?
No. STAROps resources belong to an Alibaba Cloud account and are not isolated between RAM users and the Alibaba Cloud account. Session records, Digital Employees, Missions, and other resources created by RAM users still belong to the same Alibaba Cloud account.
To restrict the resources that different RAM users can view or manage, use RAM policies for isolation. For example, control the authorization scope of Actions such as starops:ListThreads, starops:GetThread, starops:ListDigitalEmployees, starops:GetDigitalEmployee, starops:ListMissions, and starops:GetMission. Existing policies can continue to use the corresponding cms:* Actions.
Do I need to activate STAROps separately?
No. Click the STAROps entry point in the CloudMonitor 2.0 (CMS 2.0) or Simple Log Service (SLS) console. STAROps is automatically activated by default, and no additional operations are required.