OAgent is an intelligent operation and maintenance diagnostic assistant for the OceanBase database ecosystem. Through LLM-driven conversational interaction, it transforms tasks such as inspection, diagnosis, tuning, and alert analysis that database administrators (DBAs) and operations engineers rely on daily into a combination of natural language question answering and automated tool invocation. This lowers the professional threshold for database operations and maintenance and improves problem identification and resolution efficiency.
Product positioning
Traditional database operations rely on operations personnel repeatedly switching between multiple platforms such as OceanBase Cloud Platform (OCP), OceanBase Autonomy Service (OAS), and OceanBase Developer Center (ODC). They manually collect metrics, compare data, and draw conclusions, which is highly dependent on experience and difficult to standardize. OAgent unifies the capabilities of these platforms into a single conversational entry point: you only need to describe your goal (for example, "Inspect this cluster" or "Which SQL statements were the slowest in the last hour"), and OAgent automatically selects the required tools and skills to collect data, cross-verify, and output evidence-based analysis conclusions and remediation suggestions.
Architecture
- Client layer: The service entry point, prioritizing Web Chat dialog calls and gradually opening up CLI and API tool calls to facilitate integration with customer-owned systems.
- Control platform: Supports management of tools, skills, components, knowledge bases, and provides commercial capabilities in authentication, security, and operational design.
- Core service layer: Supports session management, scheduled task scheduling, knowledge and memory management, supports multi-agent collaboration, and supports multiple different roles.
- OceanBase component access layer: Encapsulates the capabilities of different tools through tools and skills, providing a unified entry point for intelligent access to various tool products.
- Knowledge base: Includes proprietary knowledge bases (OceanBase knowledge, component knowledge, expert experience, expert skills, etc.) and custom knowledge bases (user-recorded knowledge, supporting import and management).
- Model and inference layer: Relying on LLM capabilities, it executes inference requests initiated by agents.
Technical features
Connection credentials and cluster binding relationships for OCP/OAS/ODC are centrally maintained in Product Management. Cluster/tenant information is automatically injected during tool invocation based on the session context, reducing the need for repeated parameter filling.
Tool invocation results are automatically rendered into charts (line charts, bar charts, pie charts, etc.) based on the scenario, reducing the manual effort required to draw charts from text.
Built-in retry limits and error categorization (authentication failure, network unreachable, timeout, etc.) for OCP/OAS interface failure scenarios are provided to prevent the agent from repeatedly and invalidly retrying the same fault.