AppBuilder

also known as Baidu AppBuilder, 千帆AppBuilder, 百度千帆 AppBuilder

Type: low-code · Vendor: Baidu (百度) — Baidu Intelligent Cloud / Qianfan · Language: Python (SDK) + hosted visual platform · License: Apache-2.0 (AppBuilder-SDK on GitHub); hosted Qianfan AppBuilder console is proprietary · Status: active · Status in practice: mature

Links: homepage docs repo

Provide an Agent-centred, one-stop application development platform on Baidu's Qianfan cloud where a builder composes a RAG-and-tool-using agent from Baidu-ecosystem Components and MCP services, with workflow orchestration exposed both as a hosted console and as an open-source Python SDK.

Description. AppBuilder is Baidu's hosted, low-code agent platform on the Qianfan large-model cloud. Per Baidu's product page it is the '以Agent为核心的一站式企业级大模型服务平台' (Agent-centred one-stop enterprise large-model platform), built around four pillars: 'Agent引擎、工具及MCP、模型服务与企业级服务' (Agent engine, tools and MCP, model services, enterprise services). The companion AppBuilder-SDK on GitHub is Apache-2.0 and exposes 'Message、Component、AgentRuntime多级工作流抽象' (Message / Component / AgentRuntime multi-level workflow abstractions) over '40+个源于百度生态的优质组件' (40+ premium components from the Baidu ecosystem) for building production RAG applications.

Agent loop shape. Two surfaces share one conceptual model. In the hosted Qianfan AppBuilder console, a builder assembles an Agent from a system prompt, a Knowledge Base (RAG), Components/Skills (callable tools), MCP services, and optionally a Multi-Agent orchestration graph that routes a turn across specialised sub-agents. In code, the AppBuilder-SDK exposes the same loop through three abstraction levels — Message (the data unit), Component (a callable RAG/tool node), and AgentRuntime (the orchestrator that runs a workflow over Components). RAG is realised as a chain of typed Components covering document parsing, chunking, embedding, indexing, retrieval, and answer generation.

Primary use cases

no-code/low-code agent building on Qianfan with Baidu-ecosystem components
production RAG over enterprise documents (parse → chunk → embed → index → retrieve → generate)
multi-agent orchestration through the Multi-Agent engine
tool/MCP integration via curated Skill marketplace and MCP services
code-level extension via the open-source AppBuilder-SDK

flowchart TD USR[User / API caller] --> AGT[Agent engine — Qianfan AppBuilder] AGT --> KB[Knowledge Base — RAG chain] KB --> P[Doc parsing] --> CHK[Chunking] --> EMB[Embedding] --> IDX[(Index)] --> RET[Retrieval] --> GEN[Answer generation] AGT --> COMP[Components / Skills<br/>40+ Baidu ecosystem] AGT --> MCP[MCP services] AGT --> MA{Multi-Agent<br/>orchestration engine} MA -.delegate.-> SUB1[Specialised sub-agent A] MA -.delegate.-> SUB2[Specialised sub-agent B] AGT --> SDK[AppBuilder-SDK<br/>Message / Component / AgentRuntime] GEN --> OUT[Response] COMP --> OUT MCP --> OUT

Key concepts

Agent engine (Agent引擎) (docs) — Core of Qianfan AppBuilder: the agent runtime that drives reasoning over components, tools, and MCP services.
Component (组件) / Skill → tool-use (docs) — Reusable callable unit; 40+ Baidu-ecosystem components plus curated Skills are composed into the agent.
Knowledge Base (知识库) / RAG chain → agentic-rag (docs) — Production RAG built from typed components: parse, chunk, embed, index, retrieve, generate.
MCP services → mcp (docs) — Tools and external systems integrated via Model Context Protocol; first-class alongside built-in components.
Multi-Agent orchestration (Multi-Agent编排引擎) → supervisor (docs) — Engine that distributes tasks across multiple specialised agents and coordinates their results.
AgentRuntime (SDK) (docs) — Top-level workflow runner in the Python SDK; sits above Component and Message abstractions.