Built-in Components

Relevant source files

• agent/canvas.py
• agent/component/agent_with_tools.py
• agent/component/base.py
• agent/component/categorize.py
• agent/component/data_operations.py
• agent/component/exit_loop.py
• agent/component/invoke.py
• agent/component/iteration.py
• agent/component/llm.py
• agent/component/loop.py
• agent/component/loopitem.py
• agent/component/message.py
• agent/component/variable_assigner.py
• agent/tools/base.py
• common/data_source/rest_api_connector.py
• common/mcp_tool_call_conn.py
• docs/guides/agent/agent_component_reference/_category_.json
• docs/guides/agent/agent_component_reference/agent.mdx
• docs/guides/agent/agent_component_reference/categorize.mdx
• docs/guides/agent/agent_component_reference/code.mdx
• docs/guides/agent/agent_component_reference/iteration.mdx
• docs/guides/agent/agent_component_reference/retrieval.mdx
• docs/guides/agent/agent_component_reference/switch.mdx
• docs/guides/agent/agent_component_reference/transformer.md
• docs/guides/agent/agent_introduction.md
• docs/guides/agent/best_practices/_category_.json
• docs/guides/agent/embed_agent_into_webpage.md
• rag/prompts/generator.py
• test/testcases/test_web_api/test_canvas_app/test_invoke_component_unit.py
• test/unit_test/data_source/test_rest_api_connector.py
• web/src/components/highlight-markdown/__tests__/index.test.tsx
• web/src/constants/agent.tsx
• web/src/pages/agent/canvas/index.tsx
• web/src/pages/agent/canvas/node/dropdown/accordion-operators.tsx
• web/src/pages/agent/canvas/node/dropdown/operator-item-list.tsx
• web/src/pages/agent/canvas/node/extractor-node.tsx
• web/src/pages/agent/constant/index.tsx
• web/src/pages/agent/context.ts
• web/src/pages/agent/form-sheet/form-config-map.tsx
• web/src/pages/agent/hooks/use-add-node.ts
• web/src/pages/agent/hooks/use-cache-chat-log.ts
• web/src/pages/agent/hooks/use-show-drawer.tsx
• web/src/pages/agent/log-sheet/index.tsx
• web/src/pages/agent/log-sheet/workflow-timeline.tsx
• web/src/pages/agent/operator-icon.tsx

This page documents the standard components available in RAGFlow's Canvas workflow system. These components serve as building blocks for constructing AI agent workflows, ranging from simple LLM invocations to complex multi-tool agent orchestrations with retrieval and conditional routing.

Component Overview

RAGFlow provides a library of built-in components that inherit from ComponentBase agent/component/base.py407-420 Each component consists of a parameter class (inheriting from ComponentParamBase agent/component/base.py43-54) and an implementation class that provides the _invoke or _invoke_async method.

Component Categories

Built-in components fall into several functional categories:

Category	Components	Purpose
LLM Interaction	LLM, Agent, RewriteQuestion	Direct LLM calls with prompt templates and tool-augmented ReAct agents.
Information Retrieval	Retrieval, Search tools	Knowledge base search and context-aware content generation.
Control Flow	Categorize, Switch, Iteration, Loop	Conditional branching and repetitive execution logic.
Data & Logic	CodeExec, DataOperations, ListOperations, VariableAssigner	Manipulation of variables and data structures within the workflow.
Integration	Tool, Message, Invoke, DocGenerator, Crawler	External API calls, specialized tool functions, user communication, and web scraping.

Sources: agent/component/base.py43-54 agent/component/base.py407-420 web/src/pages/agent/constant/index.tsx60-70 web/src/constants/agent.tsx69-122

---

LLM Component

The LLM component agent/component/llm.py86-101 provides direct access to language models with configurable prompts and generation parameters.

Component Structure

Sources: agent/component/llm.py34-83 agent/component/llm.py86-101 agent/component/base.py407-420 api/db/services/llm_service.py27

Variable Resolution in Prompts

The LLM component supports variable references using the syntax {component_id@output_var} or {sys.variable}. Variables are resolved during invocation via string_format agent/component/llm.py127-134

Image Input Handling

The component automatically detects base64-encoded images in input variables via _extract_data_images agent/component/llm.py162-176 It handles vision-capable models by processing these images before sending the payload to the LLMBundle agent/component/llm.py180-214

Sources: agent/component/llm.py127-134 agent/component/llm.py162-176 agent/component/llm.py180-214

---

Agent Component

The Agent component agent/component/agent_with_tools.py74-95 implements a ReAct (Reasoning + Acting) loop that iteratively calls language models to plan actions, invoke tools, and reflect on results.

Architecture

Sources: agent/component/agent_with_tools.py40-71 agent/component/agent_with_tools.py74-111 agent/tools/base.py29 agent/component/agent_with_tools.py104-111 common/mcp_tool_call_conn.py34

Tool Loading and ReAct Loop

The agent loads tool components during initialization agent/component/agent_with_tools.py80-95 It handles tool calls by binding them to the LLMBundle agent/component/agent_with_tools.py113-116 The core execution logic follows the ReAct pattern where the LLM decides which tool to call based on prompts like kb_prompt or structured_output_prompt agent/component/agent_with_tools.py35 It supports MCP (Model Context Protocol) servers for external tool integration via MCPServerService agent/component/agent_with_tools.py104-112

Sources: agent/component/agent_with_tools.py80-95 agent/component/agent_with_tools.py113-116 agent/component/agent_with_tools.py104-112 agent/component/agent_with_tools.py35

---

Retrieval Component

The Retrieval component extracts information from specified datasets or memories to provide context for LLMs.

Key Features

• Hybrid Search: Combines weighted keyword similarity and vector similarity. The parameter initialKeywordsSimilarityWeightValue controls this balance web/src/pages/agent/constant/index.tsx95
• Reranking: Supports optional rerank models to improve result precision via rerank_id web/src/pages/agent/constant/index.tsx92
• Retrieval Sources: Can switch between Dataset and Memory sources web/src/pages/agent/constant/index.tsx82-85 In the UI, this is controlled via the retrieval_from field web/src/pages/agent/constant/index.tsx99
• Knowledge Graph: Supports multi-hop QA by enabling use_kg web/src/pages/agent/constant/index.tsx96
• Output Formatting: Formats retrieved chunks into a standardized structure using chunks_format rag/prompts/generator.py41-66

Sources: web/src/pages/agent/constant/index.tsx82-110 rag/prompts/generator.py41-66 agent/canvas.py42

---

Categorize Component

The Categorize component agent/component/categorize.py90-91 classifies user input into predefined categories and routes execution to different downstream components.

Execution Flow

Sources: agent/component/categorize.py30-88 agent/component/categorize.py101-154 agent/component/categorize.py129-132 web/src/pages/agent/constant/index.tsx122-133

---

Message Component

The Message component agent/component/message.py65-66 handles communication with the end-user, supporting streaming and variable interpolation.

Logic Flow

• Variable Interpolation: Uses regex variable_ref_patt to find and replace placeholders with actual values from the canvas agent/component/message.py193-216
• Streaming: Implements an asynchronous generator _stream to yield content chunks as they are resolved agent/component/message.py188-228
• Memory Integration: Can automatically save messages to the conversation memory via queue_save_to_memory_task agent/component/message.py43
• Download Extraction: Identifies document metadata (e.g., doc_id, filename, mime_type) in values to populate the downloads output agent/component/message.py88-101

Sources: agent/component/message.py46-63 agent/component/message.py188-228 agent/component/message.py88-101 agent/component/message.py43

---

Control Flow Components

Switch Component

The Switch component handles conditional branching based on logical operations. It supports multiple conditions with logical operators (AND/OR) web/src/constants/agent.tsx196-199

• Comparison Operators: Supports =, ≠, >, ≥, <, ≤, contains, empty, in, etc. web/src/constants/agent.tsx124-139
• Else Routing: The default "Else" path is tracked via end_cpn_ids web/src/pages/agent/constant/index.tsx80

Iteration and Loop Components

• Iteration: Processes lists of items by looping through a sub-graph. The IterationStartNode web/src/pages/agent/canvas/index.tsx64 marks the entry point of the iteration block.
• Loop: Structured repetitive logic within the workflow canvas. The LoopStartNode web/src/pages/agent/canvas/index.tsx67 represents the beginning of a loop cycle.
• ExitLoop: Terminates a loop based on specific conditions web/src/pages/agent/canvas/index.tsx61

Sources: web/src/pages/agent/canvas/index.tsx61-108 web/src/pages/agent/constant/index.tsx80 web/src/constants/agent.tsx124-199

---

Integration and Tool Components

Invoke Component

The Invoke component agent/component/invoke.py57-58 allows workflows to interact with external HTTP APIs.

• Methods: Supports get, post, and put agent/component/invoke.py50
• Data Types: Supports json and formdata agent/component/invoke.py54
• SSRF Guard: Validates URLs to ensure they are safe and pins DNS to prevent redirection attacks agent/component/invoke.py179-182
• Variable Rendering: Headers and request bodies support variable interpolation via _render_template agent/component/invoke.py125-133

Code Component

The Code component enables execution of Python or JavaScript snippets for dynamic data processing.

• Template: Provides standard templates for Python and JavaScript entry points web/src/constants/agent.tsx16-29
• Initialization: Initialized with default code values for the selected language web/src/pages/agent/hooks/use-add-node.ts165

Built-in Search Tools

The Tool component serves as a wrapper for specialized search plugins:

• DuckDuckGo: Text and news search web/src/pages/agent/constant/index.tsx164-178
• ArXiv/PubMed/GoogleScholar: Academic search integrations web/src/pages/agent/constant/index.tsx231-270
• ExeSQL: Executes SQL queries against MySQL/PostgreSQL databases web/src/pages/agent/constant/index.tsx335-343

Sources: agent/component/invoke.py50-182 web/src/constants/agent.tsx16-29 web/src/pages/agent/constant/index.tsx164-343 web/src/pages/agent/hooks/use-add-node.ts165

---

Component Base Logic

All components share a common lifecycle managed by ComponentBase agent/component/base.py407-420

Output and Error Handling

Components store results using set_output(key, value) agent/component/base.py458-461 Special output keys include:

• _ERROR: Captures execution errors agent/component/base.py463
• _elapsed_time: Tracks execution duration agent/component/base.py418

Exception Routing

Components can configure automated responses to failures via ComponentParamBase agent/component/base.py43-53:

• Method: Defines behavior on error (e.g., retry logic).
• Goto: Jump to a specific component on error agent/component/base.py575-582

Sources: agent/component/base.py43-53 agent/component/base.py407-585