AI Studio Introduction

Overview

• AI Studio is the AI capability hub of the Bonree ONE platform. It provides unified management of foundational capabilities including models, tools, Skills, and knowledge bases, supports the full lifecycle of Agent building and operations, and uses Sage AI as the unified interaction entry point for Agents. Before use, please import the AI Studio Token license.
• The capability framework of AI Studio consists of three layers:

Foundational Capability Layer: Models, tools, Skills, and knowledge bases — providing the core capabilities of reasoning, action, and retrieval for Agents.

Agent Layer: Supports three build modes — Workflow, Autonomous Decision, and Knowledge Planning — to accommodate diverse scenarios ranging from fixed processes to autonomous reasoning.

Interaction & Operations Layer: Sage AI serves as the unified interaction entry point. Alert diagnosis history and diagnosis statistics form the operations feedback loop for alert diagnosis, enabling continuous tracking and optimization of diagnostic effectiveness.

Value

1. Unified AI capability foundation, lowering the barrier to adoption

Models, tools, Skills, and knowledge bases are centrally managed within AI Studio. All functional modules share the same resource system, eliminating duplicate configuration and resource fragmentation.

The platform includes built-in models, MCP tools, Skills, and knowledge documents that are ready to use out of the box — no setup from scratch required. Even non-expert users can quickly enable AI diagnostic capabilities.

2. Flexible Agent building to cover diverse operations scenarios

Three build modes — Workflow, Autonomous Decision, and Knowledge Planning — allow operations teams to choose the most suitable Agent type based on their scenario requirements and technical capabilities.

Agents can simultaneously reference tools, Skills, and knowledge bases, enabling end-to-end automation across the full "Query → Reason → Execute → Notify" pipeline, significantly reducing manual intervention.

3. Knowledge-driven, enabling AI to understand business context

Upload team-accumulated troubleshooting experience, remediation SOPs, system architecture documentation, and more to the knowledge base. This gives AI contextual business understanding during diagnosis, significantly improving diagnostic accuracy and relevance.

A single knowledge base can be referenced by multiple Agents simultaneously — maintain once, reuse across all scenarios, with no redundant effort.

4. Open tool ecosystem to continuously expand diagnostic capabilities

Connect custom tools (log queries, metric analysis, change retrieval, etc.) via the MCP protocol, forming a complete diagnostic toolchain alongside built-in tools.

The Skill framework supports action-oriented capabilities such as sending notifications, creating tickets, and pushing logs — enabling Agents to not only "diagnose" but also "remediate."

5. Operations feedback loop for data-driven continuous optimization

Diagnosis history provides a complete record of each diagnostic execution, user ratings, and reasons for negative feedback — enabling precise issue identification for strategy optimization and tool enhancement.

Diagnosis statistics present key metrics such as diagnostic trigger volume, success rate, and mode distribution in visual charts, providing management with quantitative data to support AI capability operations.

Module Descriptions

Sage AI

• Sage AI is the unified interaction entry point of AI Studio, with Insight Mode and Action Mode available for switching.
• In Insight Mode, gain one-click visibility across the entire observability domain. Out of the box and with no binding to knowledge bases, Agents, Skills, or tools, users can query platform-internal data and knowledge through conversational Q&A for data insights.
• In Action Mode, one click drives a self-closing operations loop. Users can directly invoke published Agents through the conversational interface to complete tasks such as fault diagnosis, intelligent Q&A, automated reporting, and remediation. For custom scenarios, users can add custom resources and Agents to enable custom Q&A and execution.
• Key capabilities:

Conversational invocation: Describe the problem in natural language; the Agent automatically interprets the intent and invokes tools, knowledge, and models to complete the task.

Action Mode multi-Agent selection: Select the required Agent or Skill (fault diagnosis, intelligent Q&A, etc.) in the input box and switch flexibly based on the scenario.

Session history: Supports resuming context to continue a conversation.

Result display: Diagnostic conclusions, charts, and tool invocation processes are clearly presented within the conversation flow, with Markdown rendering support.

Models

• The Models module supports connecting various large models compatible with the OpenAI API protocol, and serves as the foundation for all intelligent reasoning capabilities in AI Studio. Connected models can be used across multiple scenarios including workflow nodes, Autonomous Decision Agents, and knowledge retrieval.
• Supported model types:

LLM (Large Language Model): Used for reasoning, summarization, Q&A, and diagnostic conclusion generation.

Embedding Model: Used for vectorizing and retrieving knowledge base documents. Must be selected when creating a knowledge base.

Reranking Model: Used to improve the precision of knowledge retrieval results. Used in Knowledge Planning Agents.

• Key features: Add / edit / delete models; filter by model type, name, or ID; automatic connectivity validation upon saving to detect connection issues early.

Tools

• The Tools module connects external service tools to the platform via the Model Context Protocol (MCP) for Agents to invoke during reasoning, while also providing a built-in out-of-the-box toolset. Tools are the core capability carrier for Agents performing "Query and Analysis" type operations.
• The Tools module provides two management views:

MCP Service View (default): Manages tool services by MCP Server, displaying basic information and the number of tools for each Server. This is the primary entry point for adding, editing, and deleting MCP Servers. Built-in MCP Servers are ready to use out of the box, always pinned to the top, and cannot be deleted.

Tool View: Displays all tools across all MCP Servers in a flat list, with filtering by name, description, tag, or owning Server. Supports adding tags to tools for categorized management and quick location.

• Key features: Add custom MCP Servers (with request header configuration and timeout settings); one-click update of the tool list for all services; click a service card to view full tool details including parameter descriptions.

Skills

• A Skill is a reusable unit encapsulating a specific operational capability, enabling Agents to not only "query" but also "execute remediation actions." Skills are either built-in to the platform or imported by users from external sources.
• Typical built-in Skill examples:

Send DingTalk / WeCom / email notifications

Automatically create fault tickets in an ITSM system

Push operations logs to a logging platform

• Skill management features:

Built-in Skills only require parameter configuration (e.g., Webhook address, SMTP settings) before they can be invoked in Agents — no development required.

Supports importing external Skill packages (.zip / .skill format) to extend the platform's operational capabilities.

When configuring an Agent, each Skill can be assigned dedicated parameters to override global settings, enabling differentiated invocation.

Knowledge Base

• The knowledge base is a digital asset repository for team operations knowledge. It supports consolidating technical expertise, troubleshooting SOPs, system architecture documentation, and more. After vectorization, the content is available for Agents to retrieve and reference, giving AI the ability to understand business context.
• Document processing pipeline: Upload → Queued → Indexing → Available. Only documents in the "Available" state can be referenced by Agents.
• Key features:

Select a vector model when creating a knowledge base; supports uploading multiple document files.

Documents support renaming, exporting the original file, and deletion; batch deletion is supported.

Once documents exist in a knowledge base, the vector model cannot be changed (a new knowledge base must be created if a different model is needed).

A single knowledge base can be referenced simultaneously by multiple Agents (workflow nodes, Autonomous Decision Agents) — maintain once, effective across all scenarios.

Agents

• An Agent is the core capability carrier of AI Studio, encapsulating the reasoning logic, tool invocation chain, and knowledge sources for a specific scenario. The platform provides built-in Agents ready to use out of the box, and also supports user-built Agents.

Build Modes

Build Mode	Core Mechanism	Applicable Scenarios	Typical Configuration
Workflow	Visual node orchestration with a fixed execution path; LLM can be used as one of the nodes	Diagnostic tasks with clearly defined steps and stable processes	Model node + Tool node + Knowledge node + Skill node
Autonomous Decision	ReAct framework; LLM autonomously determines tool invocation and reasoning rounds	Complex scenarios with unknown fault causes requiring multi-turn dynamic reasoning	Reasoning model + MCP tools + Skills + Associated knowledge

Agent Status

Draft: Not yet published; can continue to be edited; cannot be invoked in Sage AI or fault diagnosis.

Published: Can be directly invoked in Sage AI and alert fault diagnosis; further edits must be republished to take effect.

Key Operations

Create / Clone an Agent: Completed through a two-step process (Step 1: Basic Settings → Step 2: Build Configuration). When cloning, the configuration of the source Agent is automatically pre-filled.

Edit: User-built Agents support full editing; built-in Agents only support editing certain fields such as reasoning model and associated knowledge.

Export (Workflow mode only): Export the workflow canvas content as a YAML file for backup and migration.

Inter-Module Collaboration

The modules of AI Studio collaborate closely to form a complete AI capability chain:

Collaboration	Description
Models → Agents	The reasoning capabilities of Agents (workflow nodes, Autonomous Decision) depend on LLMs connected in the Models module. Knowledge base indexing depends on the embedding model.
Tools → Agents	Autonomous Decision Agents invoke MCP tools configured in the Tools module during reasoning to perform operations such as log queries, metric retrieval, and change queries.
Skills → Agents	Agents can invoke Skills to directly query or execute remediation actions — such as sending alert notifications or creating tickets — enabling full end-to-end automation from "detection" to "remediation."
Knowledge Base → Agents	Both knowledge nodes in workflows and the "Associated Knowledge" field in Autonomous Decision Agents can reference knowledge base documents to provide business context for AI reasoning.
Agents, Skills → Sage AI	Published Agents and Skills can be directly invoked by users in Sage AI for interactive fault diagnosis and Q&A.
Agents → Alert Diagnosis	Published Agents can be selected in fault diagnosis policies; when an alert is triggered, the corresponding Agent is automatically executed to complete the diagnosis.
Diagnosis Execution → Operations Analytics	All diagnosis execution records triggered by alerts are aggregated into diagnosis history and diagnosis statistics, forming an operations data feedback loop that drives continuous optimization of strategies and knowledge.

Permissions

The AI Studio permission framework consists of two dimensions: feature permissions and data permissions.

Feature Permissions

AI Read-Write Access: Allows creating, deleting, editing, and importing across all AI Studio pages.

AI Read-Only Access: Allows viewing content on all AI Studio pages, querying, and exporting; write operations are not permitted.

Data Permissions

Models: Isolated by environment. All resource domains within the same environment share the same model list; switching environments changes the model list accordingly.

Tools, Skills, Knowledge Bases, Agents: Isolated by resource domain. Content created within a resource domain is not visible to other resource domains by default. Content created under "All" resource domains is visible and usable by all resource domains with permission.

Sage AI, Diagnosis History, Diagnosis Statistics: Isolated by resource domain. Data is isolated between different primary accounts; data is not isolated between a primary account and its sub-accounts, or between sub-accounts. Under "All" resource domains, data from all resource domains the current account has permission to access can be viewed.

Quick Start

The following is the recommended configuration path to enable AI Studio fault diagnosis capabilities from scratch:

Step	Action	Description
1	Add a model	Add at least one LLM on the "Models" page. If you plan to use a knowledge base, also add an embedding model.
2	Configure tools (optional)	Connect the required MCP Servers on the "Tools" page. Built-in tools are available immediately without configuration.
3	Configure Skills (optional)	Fill in the configuration parameters for the built-in Skills you need (e.g., notification channel Webhook address) on the "Skills" page.
4	Build a knowledge base (optional)	Create a knowledge base and upload troubleshooting documents on the "Knowledge Base" page. Wait for the status to change to "Available" before referencing.
5	Create an Agent (optional)	Create a new Agent on the "Agents" page, select a build mode, configure the model, tools, Skills, and knowledge base, then publish.
6	Configure a diagnosis policy (optional)	Configure alert trigger rules in "Fault Diagnosis Policy," associating published Agents with alert filter conditions to enable automated diagnosis.
7	Trigger diagnosis interactively	View diagnostic conclusions in "Sage AI" or the Alert List; review past diagnostic records and submit ratings in "Diagnosis History."