Glossary

Comprehensive index of LangChain concepts, APIs, and types for quick lookups.

Need task-based examples? See the Task Index for direct "I want to..." → code mappings.

Core Concepts

Agent

A ReAct (Reasoning + Acting) agent that combines language models with tools to iteratively work towards solutions. Created with createAgent().

See: Agent Guide | Agent API

Tool

A function that agents can call to take actions. Created with tool() or by extending StructuredTool.

See: Tool Guide | Tool API

Middleware

Composable components that extend agent behavior with cross-cutting concerns like logging, retries, or human-in-the-loop.

See: Middleware Guide | Middleware Overview

Structured Output

Type-safe, validated responses from agents using Zod schemas or JSON schemas.

See: Structured Output Guide | API Reference

State

Agent state that persists across invocations when using a checkpointer. Defined with stateSchema.

See: Agent Guide - Custom State

Context

Read-only data provided per invocation that is not persisted. Defined with contextSchema.

See: Agent Guide - Context

Checkpointer

Storage mechanism for persisting agent state across invocations. Enables conversation memory.

See: Agent Guide - Persistence

API Functions

createAgent()

Creates a production-ready ReAct agent with configuration for models, tools, prompts, state, and middleware.

Signature: function createAgent<TConfig>(params: CreateAgentParams): ReactAgent<TConfig>

See: Agent API | Quick Reference

tool()

Creates a structured tool from a function and Zod schema.

Signature: function tool<T>(func: (input: T) => any, fields: { name, description, schema }): StructuredTool<T>

See: Tool API | Quick Reference

initChatModel()

Initializes a chat model from any of 18+ supported providers using a universal interface.

Signature: function initChatModel(model?: string | ChatModel, fields?: InitChatModelFields): ChatModel

See: Model API | Quick Reference

createMiddleware()

Creates custom middleware with lifecycle hooks and state/context schemas.

Signature: function createMiddleware<TSchema, TContextSchema, TTools>(config: MiddlewareConfig): AgentMiddleware

See: Middleware Overview | Custom Middleware Guide

toolStrategy()

Creates a tool-based structured output strategy.

Signature: function toolStrategy<T>(schema: ZodType<T>, options?: ToolStrategyOptions): ToolStrategy<T>

See: Structured Output API

providerStrategy()

Creates a provider-native structured output strategy.

Signature: function providerStrategy<T>(schema: ZodType<T>): ProviderStrategy<T>

See: Structured Output API

filterMessages()

Filters message arrays based on types or custom criteria.

See: Message API

trimMessages()

Trims message arrays to fit within token limits.

See: Message API

Classes

ReactAgent

Agent instance returned by createAgent() with methods for execution.

Methods:

• invoke() - Synchronous execution
• stream() - Streaming execution
• streamEvents() - Event streaming
• batch() - Batch processing

See: Agent API - ReactAgent Class

Tool

Base class for creating tools with string input.

See: Tool API - Base Tool Class

DynamicTool

Dynamically created tool with string input.

See: Tool API - DynamicTool Class

StructuredTool

Tool with structured input schema (Zod).

See: Tool API - StructuredTool Class

DynamicStructuredTool

Dynamically created structured tool.

See: Tool API - DynamicStructuredTool Class

HumanMessage

Message class for user/human messages.

See: Message API

AIMessage

Message class for AI assistant responses.

See: Message API

SystemMessage

Message class for system instructions.

See: Message API

ToolMessage

Message class for tool execution results.

See: Message API

InMemoryStore

In-memory key-value store for agent data.

See: Storage API

ToolStrategy

Strategy class for tool-based structured outputs.

See: Structured Output API

ProviderStrategy

Strategy class for provider-native structured outputs.

See: Structured Output API

Document

Document container for text processing.

See: Document API

Types

CreateAgentParams

Configuration object for createAgent().

See: Agent API | Quick Reference

UserInput

Input type for agent invocation with messages and state.

See: Agent API - Input Types

State

Agent state returned from invoke/stream with messages and custom state.

See: Agent API - State Types

InvokeConfiguration

Configuration for agent invocation including context and thread_id.

See: Agent API - Configuration Types

StreamConfiguration

Configuration for streaming with streamMode option.

See: Agent API - Configuration Types

ToolCall

Information about a tool call made by the model.

See: Agent API - Tool Types | Tool API

ToolResult

Result from tool execution.

See: Agent API - Tool Types

ToolConfig

Configuration passed to tools during execution.

See: Tool API

ResponseFormat

Union type for structured output formats (Zod, JSON schema, strategies).

See: Structured Output API - Response Format Types

AgentMiddleware

Middleware interface with hooks and schemas.

See: Middleware Overview - Base Interface

MiddlewareConfig

Configuration for creating custom middleware.

See: Middleware Overview - Creating Custom Middleware

Runtime

Runtime context available to middleware hooks.

See: Agent API - Runtime Types | Middleware Overview

Type Inference

InferAgentState

Extracts full merged state type from agent (includes built-in state).

See: Type Inference - Agent Type Inference | Agent API

InferAgentResponse

Extracts structured response type from agent.

See: Type Inference - Agent Type Inference

InferAgentContext

Extracts full merged context type from agent.

See: Type Inference - Agent Type Inference

InferAgentTools

Extracts tools array type from agent.

See: Type Inference - Agent Type Inference

InferAgentMiddleware

Extracts middleware array type from agent.

See: Type Inference - Agent Type Inference

InferMiddlewareState

Extracts state type from single middleware.

See: Type Inference - Middleware Type Inference

InferMiddlewareStates

Merges states from middleware array.

See: Type Inference - Middleware Type Inference

InferMiddlewareContext

Extracts context type from single middleware.

See: Type Inference - Middleware Type Inference

InferMiddlewareContexts

Merges contexts from middleware array.

See: Type Inference - Middleware Type Inference

InferMiddlewareTools

Extracts tools from middleware array.

See: Type Inference - Tool Type Inference

InferSchemaInput

Infers input type from Zod schema or object.

See: Type Inference - Schema Type Inference

ToolsToMessageToolSet

Converts tools array to mapped type by name.

See: Type Inference - Tool Type Inference

Error Classes

MultipleToolsBoundError

Thrown when attempting to bind tools/structured output to a model that already has tools bound.

See: Agent API - Error Classes | Error Handling Guide

MultipleStructuredOutputsError

Thrown when multiple structured outputs are returned but only one was expected.

See: Agent API - Error Classes

StructuredOutputParsingError

Thrown when structured output parsing fails.

See: Agent API - Error Classes | Structured Output API

ToolInvocationError

Thrown when tool execution fails.

See: Agent API - Error Classes

ToolCallLimitExceededError

Thrown when tool call limit is exceeded.

See: Built-in Middleware - Tool Call Limit

PIIDetectionError

Thrown when PII detection fails.

See: Built-in Middleware - PII Detection

Built-in Middleware

humanInTheLoopMiddleware

Pauses agent execution for human review and approval.

See: Human-in-the-Loop Guide | Built-in Catalog

summarizationMiddleware

Automatically summarizes and compresses message history.

See: Built-in Catalog

piiMiddleware

Detects and handles personally identifiable information.

See: Built-in Catalog

piiRedactionMiddleware

Simple PII redaction without detection configuration.

See: Built-in Catalog

dynamicSystemPromptMiddleware

Dynamically updates system prompts based on state/context.

See: Built-in Catalog

llmToolSelectorMiddleware

Uses an LLM to intelligently select tools based on queries.

See: Built-in Catalog

contextEditingMiddleware

Edits or clears context and messages during execution.

See: Built-in Catalog

toolCallLimitMiddleware

Limits the number of tool calls per invocation.

See: Built-in Catalog

modelCallLimitMiddleware

Limits the number of model invocations.

See: Built-in Catalog

todoListMiddleware

Adds todo list management capabilities to agents.

See: Built-in Catalog

modelRetryMiddleware

Retries failed model calls with configurable backoff.

See: Built-in Catalog

modelFallbackMiddleware

Automatically falls back to alternative models on failure.

See: Built-in Catalog

toolRetryMiddleware

Retries failed tool calls with configurable backoff.

See: Built-in Catalog

toolEmulatorMiddleware

Emulates tool execution for testing without calling real tools.

See: Built-in Catalog

openAIModerationMiddleware

Uses OpenAI's moderation API to check for policy violations.

See: Built-in Catalog

anthropicPromptCachingMiddleware

Enables Anthropic's prompt caching for improved performance.

See: Built-in Catalog

Model Providers

Quick lookup for supported model provider formats:

• OpenAI: openai:gpt-4o, openai:gpt-4o-mini
• Anthropic: anthropic:claude-3-5-sonnet-20241022
• Google: google:gemini-1.5-pro
• Ollama: ollama:llama3.1
• Cohere: cohere:command-r-plus
• Groq: groq:mixtral-8x7b-32768
• Together: together:meta-llama/Llama-3-70b-chat-hf

See: Model Guide | Quick Reference - Model Providers

Common Patterns

Agent with Tools

Creating an agent with tool-calling capabilities.

See: Quick Reference - Agent with Tools | Agent Guide

Agent with Structured Output

Getting type-safe structured responses from agents.

See: Quick Reference - Agent with Structured Output | Structured Output Guide

Agent with Custom State

Adding custom state that persists across invocations.

See: Quick Reference - Agent with Custom State | Agent Guide

Agent with Persistence

Using checkpointers for conversation memory.

See: Quick Reference - Agent with Persistence | Agent Guide

Streaming Responses

Streaming agent responses in real-time.

See: Quick Reference - Streaming Responses | Agent Guide

Batch Processing

Processing multiple inputs in parallel.

See: Agent Guide - Batch Processing

Human-in-the-Loop

Adding human review and approval to agent workflows.

See: Human-in-the-Loop Guide | Agent Guide

Environment Variables

• OPENAI_API_KEY - OpenAI API key
• ANTHROPIC_API_KEY - Anthropic API key
• GOOGLE_APPLICATION_CREDENTIALS - Google credentials path
• COHERE_API_KEY - Cohere API key
• GROQ_API_KEY - Groq API key
• MISTRAL_API_KEY - Mistral AI API key
• TOGETHER_API_KEY - Together AI API key
• FIREWORKS_API_KEY - Fireworks AI API key

See: Quick Reference - Environment Variables

Navigation

By Topic

• Agents: Guide | API
• Tools: Guide | API
• Models: Guide | API
• Messages: Guide | API
• Structured Outputs: Guide | API
• Middleware: Guide | Overview | Catalog
• Storage: Guide | API
• Type Inference: Guide
• Error Handling: Guide

Quick Access

• Quick Reference - Fast API lookups
• API Reference Index - Complete API index
• Main Documentation - Start here

Search Tips

1. By name: Use Ctrl/Cmd+F on this page to find specific APIs or concepts
2. By category: Navigate to specific sections above
3. By use case: Check the "Common Patterns" section
4. By API: See Quick Reference or API Reference Index
5. By type: See Type Inference Guide