Platform Overview & Core Concepts

What Is LongbridgeAI Agent Platform

LongbridgeAI Agent Platform is Longbridge's AI Agent orchestration platform. On a visual canvas, you drag in nodes, connect them, and configure them to build AI applications that understand natural language, call tools, and execute business logic—no need to write code from scratch.

The Longbridge AI Product Landscape

Longbridge's AI offering consists of two parts:

Product	Positioning	Who It's For
LongbridgeAI Agent Platform (this platform)	Agent orchestration platform: build, debug, and publish AI applications	Agent builders (you)
Official Chatbot suite	LongbridgeAI conversational products: the AI Q&A entry points inside the App/Web	End users

How they relate:

• Agents you build on the Agent Platform reach end users through entry points such as the official Chatbot after publishing
• Usage quota is shared: the same user's usage quota applies to both tools—chatting in the Chatbot and running flows on the Agent Platform draw from the same quota (see Workspace & Plans)
• For capabilities on the official Chatbot side (conversation charts, voice input, long-term memory, etc.), see Chat Experience

Three Application Types: Agentic Chat, Chatflow, and Workflow

This is the platform's most important concept—it determines how you build, which nodes you can use, and how your application gets invoked:

	Agentic Chat	Chatflow Agent	Workflow Agent
Positioning	A simplified version of conversational applications	Multi-turn conversational applications (canvas orchestration)	Automated batch-processing applications
How you build	No canvas needed—just configure directly	Orchestrate nodes on the canvas	Orchestrate nodes on the canvas
Typical scenarios	Quickly spin up a conversational assistant	Conversations that need branching/tool orchestration, such as customer support or market Q&A	Research report generation, batch translation, data analysis, scheduled tasks
Entry point	User sends a message in the chat window	User sends a message in the chat window	Triggered by APIs / triggers / upstream systems, with parameters passed in
How it ends	Direct conversational output	Answer node (outputs the answer to the chat window)	End node (returns results in structured formats such as JSON)
Best for	Beginners	Users who need precise control over conversation flow	Users who need task automation (not yet available)

Rule of thumb: as a beginner, use Agentic Chat for a conversational assistant; when conversation logic is complex and you need to orchestrate branches and tools, use Chatflow; for one-in, one-out automated tasks, use Workflow.

• Agentic Chat is a simplified Chatflow: under the hood it is a fixed three-node chain—Start → Agent (invoke Agent) → Answer (direct reply). The platform hides the canvas; you only configure the role and capabilities and start chatting, making it the fastest way to get started. When you need multi-branch routing or multi-node data processing, migrate to Chatflow for free-form orchestration
• ⚠️ Workflow is not yet available: because Workflow demands more from users, it is currently not open to general users; availability will be announced in product updates. The Workflow-related content in this documentation (node scenarios, triggers, examples) can serve as advance learning material
• The node reference in this handbook (1.4) applies to canvas orchestration in Chatflow / Workflow; Agentic Chat does not involve canvas nodes

Core Concepts

Node

A single processing step in a flow. Each node does one thing: call a large language model, evaluate a condition, send an HTTP request, execute code, and so on. The platform currently provides 15 nodes; see the full list in the Node Reference.

UI elements common to every node:

• Icon + name: the default name is generated in the current language and can be edited
• Comments: annotate a node to help team collaboration
• More button: right-click menu actions such as copy and delete
• Run button: nodes that support test runs can be tested individually (see "Test Run" below)
• Help documentation link: jump directly to the relevant help doc from the node editing panel

Variable

The vehicle for passing data between nodes. Output variables from upstream nodes can be referenced in downstream nodes (for example, inserting the user input collected by the Start node into an LLM's prompt).

• Each variable has a variable name and a data type (String, Number, Object, Array, etc.)
• The LLM node outputs text (String) by default; with structured output enabled, it additionally outputs structured_output (Object)
• Variable transformation, extraction, and aggregation are handled by three dedicated nodes: Variables Transformer, Parameter Extractor, and Branch Aggregator

Connections and Branches

Connections between nodes define the execution order. Nodes like IF Else and Question Classifier produce multiple outgoing branches, with different conditions taking different paths; results from multiple branches can be merged back into a single main line with Branch Aggregator.

Test Run

Run a single node or the entire flow directly on the canvas without publishing, and check whether the inputs and outputs match your expectations. Apart from pure routing/aggregation nodes such as IF Else and Branch Aggregator, processing nodes (LLM, Agent, Tool, Code, Iteration, Loop, Parameter Extractor, etc.) all support test runs—the presence of a run button on the node is the authoritative indicator. Making "configure one node, test-run it once" a habit is the most effective way to debug.

Models

Models for LLM-related nodes are selected from service settings, centrally initialized by Longbridge with test accounts provided. Some input parameters can be adjusted (the adjustable parameters are whatever the node configuration panel actually shows); model management is not yet available. Models come in two tiers, Mainstream and SOTA; SOTA consumes more usage.

Workspace

Everything you do on the platform happens inside a Workspace: Agents, Workflows, knowledge bases, and usage consumption are all scoped to a Workspace. Each Workspace is bound to a plan, which determines what features are available in that space and how much quota you get. The same person may see different features in different Workspaces. See the detailed breakdown in Workspace & Plans.

Plan and Usage

The platform offers three plans: Starter (free) / Pro / Premium. They differ in monthly usage quota (LLM calls and other consumption are metered in USD) and fine-grained feature switches (tool scope, Skill count, knowledge base capacity, publishing permissions, etc.). Starter is activated automatically upon registration. Quota is allocated monthly and released on a rolling 7-day basis; runs are blocked once you exceed your quota. See Workspace & Plans for details.

Regional Services

The platform is operated by independent legal entities in Hong Kong, Singapore, and the United States (in preparation). Pricing currency, regulatory requirements, and available capabilities vary by region, and users cannot switch service entities across regions. See Regional Services for details.

Beyond Nodes: The Full Capability Landscape

Node orchestration is the skeleton; the platform also provides a full suite of capabilities around Agents, each with its own documentation (index at Platform Capabilities):

Capability	Problem It Solves	Docs
Knowledge Base	Agents only know general knowledge, not your private documents	Knowledge Base
Skill System	Reuse specialized capabilities without bloating the System Prompt	Skill System
Workflow Triggers	Tasks need to run on a schedule or be triggered by external systems	Workflow Triggers
Guardrail	Compliance risks in inputs/outputs need mechanism-level interception	Guardrail
Context & Memory Management	Multi-turn conversations lose memory; context keeps growing	Context & Memory Management
Planning & Execution	Complex tasks lack structure; full automation is uncontrollable	Planning & Execution
Debugging & Observability	Validate results in batches; track run quality after launch	Debugging & Observability
Security Verification & Localization	Asset data authentication; Simplified/Traditional Chinese adaptation	Security Verification & Localization
Canvas & Collaboration	Prevent accidental edits in team collaboration; reuse nodes	Canvas & Collaboration
Model & Tool Ecosystem	SOTA model selection; account/market-data tool integration	Model & Tool Ecosystem
Chat Experience	Understand how your Agent is presented to users after publishing	Chat Experience
Disaster Recovery	Keep Agents stable when vendors fail or calls error out	Disaster Recovery

The 15 Nodes at a Glance

Basic nodes        Start · Answer (outputs the answer, Chatflow only) · End (Workflow only)
AI nodes           LLM (large language model) · Agent (invoke an Agent) · Question Classifier
Logic & routing    IF Else (conditional) · Branch Aggregator · Iteration · Loop
Data processing    Code (code execution) · Variables Transformer (template transform) · Parameter Extractor
External access    Tool (tool call) · Http Request

What the Simplest Chatflow Looks Like

Start ──▶ LLM ──▶ Answer
(receive user message)  (call the LLM to generate a reply)  (output the reply to the chat window)

Next step: follow Quick Start and build it yourself.