LangChain Tools

This module provides tool utilities for LangChain agents.

Represents an action with a name and arguments.

Represents an action request with a name, arguments, and description.

Base middleware interface.

Type bag that encapsulates all agent type parameters.

This interface bundles all the generic type parameters used throughout the agent system into a single configuration object. This pattern simplifi

Response when a human approves the action.

Configuration for clearing tool outputs when token limits are exceeded.

LangChain Messages

Protocol describing a context editing strategy.

Implement this interface to create custom strategies for managing conversation context size. The apply method should modify the messages array in-pla

Configuration for the Context Editing Middleware.

Default type configuration for agents. Used when no explicit type parameters are provided.

LangChain Documents

Response when a human edits the action.

Information about a tool call that has been executed.

Request for human feedback on a sequence of actions requested by a model.

Response payload for a HITLRequest.

Represents information about an interrupt.

LangChain Testing Utilities

Type bag that encapsulates all middleware type parameters.

This interface bundles all the generic type parameters used throughout the middleware system into a single configuration object. This patter

Configuration for modifying a model call at runtime. All fields are optional and only provided fields will override defaults.

Options for configuring the OpenAI Moderation middleware.

Represents a detected PII match in content

ReactAgent is a production-ready ReAct (Reasoning + Acting) agent that combines language models with tools and middleware.

The agent is parameterized by a single type bag Types that encapsulates al

Configuration for a redaction rule

Response when a human rejects the action.

Resolved redaction rule with a concrete detector function

Policy for reviewing a HITL request.

A nested named-agent execution surfaced on run.subagents.

A subagent is a nested createAgent run whose lc_agent_name differs from its parent's — e.g. a createAgent({ name }) invoked inside a

Information about a tool call that has been executed.

Represents a tool call request for the wrapToolCall hook. Contains the tool call information along with the agent's current state and runtime.

Options for configuring the Tool Emulator middleware.

Configuration for toolErrorMiddleware.

Information about a tool result from a tool execution.

Branded type for ToolStrategy arrays that preserves type information

Content block to represent an invalid tool call

Represents a request to call a tool.

Content block to represent partial data of a tool call

Annotation for citing data from a document.

Provider-specific content block.

This is used to represent content blocks that are not part of the standard LangChain content model. If a provider's non-standard output includes reasoning and tool ca

Reasoning output from a LLM.

Text output from a LLM.

This typically represents the main text content of a message, such as the response from a language model or the text of a user message.

Represents an action with a name and arguments.

Represents an action request with a name, arguments, and description.

Base middleware interface.

Type bag that encapsulates all agent type parameters.

This interface bundles all the generic type parameters used throughout the agent system into a single configuration object. This pattern simplifi

Response when a human approves the action.

Configuration for clearing tool outputs when token limits are exceeded.

LangChain Messages

Protocol describing a context editing strategy.

Implement this interface to create custom strategies for managing conversation context size. The apply method should modify the messages array in-pla

Configuration for the Context Editing Middleware.

Default type configuration for agents. Used when no explicit type parameters are provided.

LangChain Documents

Response when a human edits the action.

Information about a tool call that has been executed.

Request for human feedback on a sequence of actions requested by a model.

Response payload for a HITLRequest.

Represents information about an interrupt.

LangChain Testing Utilities

Type bag that encapsulates all middleware type parameters.

This interface bundles all the generic type parameters used throughout the middleware system into a single configuration object. This patter

Configuration for modifying a model call at runtime. All fields are optional and only provided fields will override defaults.

Options for configuring the OpenAI Moderation middleware.

Represents a detected PII match in content

ReactAgent is a production-ready ReAct (Reasoning + Acting) agent that combines language models with tools and middleware.

The agent is parameterized by a single type bag Types that encapsulates al

Configuration for a redaction rule

Response when a human rejects the action.

Resolved redaction rule with a concrete detector function

Policy for reviewing a HITL request.

A nested named-agent execution surfaced on run.subagents.

A subagent is a nested createAgent run whose lc_agent_name differs from its parent's — e.g. a createAgent({ name }) invoked inside a

Information about a tool call that has been executed.

Represents a tool call request for the wrapToolCall hook. Contains the tool call information along with the agent's current state and runtime.

Options for configuring the Tool Emulator middleware.

Configuration for toolErrorMiddleware.

Information about a tool result from a tool execution.

Branded type for ToolStrategy arrays that preserves type information

Interface for objects that can be serialized. This is a duck-typed interface to avoid circular imports.

Hook type for the afterAgent lifecycle event. Can be either a handler function or an object with a handler and optional jump targets. This hook is called once at the end of the agent invocation.

Hook type for the afterModel lifecycle event. Can be either a handler function or an object with a handler and optional jump targets. This hook is called after each model invocation.

A GraphRunStream with native agent-level projections assigned directly on the instance by createGraphRunStream (via __native transformers).

This is a pure type overlay — no runtime subclass exist

Widened middleware type for agent configuration arrays. Accepts middleware instances with any registered tools or stream transformers while preserving precise inference when middleware arrays are pass

Hook type for the beforeAgent lifecycle event. Can be either a handler function or an object with a handler and optional jump targets. This hook is called once at the start of the agent invocation.

Hook type for the beforeModel lifecycle event. Can be either a handler function or an object with a handler and optional jump targets. This hook is called before each model invocation.

Built-in PII types

Helper type to combine agent stream transformers with middleware stream transformers.

Helper type to combine agent tools with middleware tools into a single readonly array.

Union of all possible decision types.

Default type configuration for middleware. Used when no explicit type parameters are provided.

Function type that dynamically generates a description for a tool call approval request.

Type helper to extract union type from an array of Zod schemas

A headless tool that always interrupts agent execution on the server.

The implementation is provided separately on the client via useStream({ tools: [...] }) using .implement().

Configuration fields for creating a headless tool.

A tool implementation that pairs a headless tool with its execution function.

Created by calling .implement() on a HeadlessTool. Pass to useStream({ tools: [...] }) on the client side.

Helper type to infer the full merged context from an agent, including:

• The agent's own context schema (if provided via contextSchema)
• All middleware context schemas

This matches the context typ

Shorthand helper to extract the raw Context schema type from an AgentTypeConfig or ReactAgent. This returns just the contextSchema type passed to createAgent, not merged with middleware.

For the

Shorthand helper to extract the Middleware type from an AgentTypeConfig or ReactAgent.

Shorthand helper to extract the Response type from an AgentTypeConfig or ReactAgent.

Helper type to infer the full merged state from an agent, including:

• The agent's own state schema (if provided via stateSchema)
• All middleware states

This matches the state type returned by `in

Shorthand helper to extract the raw State schema type from an AgentTypeConfig or ReactAgent. This returns just the stateSchema type passed to createAgent, not merged with middleware.

For the comp

Shorthand helper to extract the StreamTransformers type (the tuple of transformer factories) from an AgentTypeConfig or ReactAgent.

Shorthand helper to extract the Tools type from an AgentTypeConfig or ReactAgent.

Helper type to extract any property from an AgentTypeConfig or ReactAgent.

Helper type to extract input type from context schema (with optional defaults)

Helper type to infer merged input state from an array of middleware (includes built-in state)

Helper type to infer merged state from an array of middleware (includes built-in state)

Helper type to infer the context schema type from a middleware

Helper type to infer the input context schema type from a middleware (with optional defaults)

Helper type to infer merged input context from an array of middleware (with optional defaults)

Helper type to infer merged context from an array of middleware

Shorthand helper to extract the ContextSchema type from a MiddlewareTypeConfig or AgentMiddleware.

Shorthand helper to extract the FullContext type from a MiddlewareTypeConfig or AgentMiddleware.

Helper type to infer the input state schema type from a middleware (all properties optional) This filters out private properties (those starting with underscore) Supports both Zod schemas (InteropZodO

Helper type to infer merged input state from an array of middleware (with optional defaults)

Shorthand helper to extract the Schema type from a MiddlewareTypeConfig or AgentMiddleware.

Helper type to infer the state schema type from a middleware This filters out private properties (those starting with underscore) Supports both Zod schemas (InteropZodObject) and StateSchema from Lang

Helper type to infer merged state from an array of middleware (just the middleware states)

Helper type to infer stream transformers from a single middleware instance.

Helper type to infer and merge stream transformers from an array of middleware.

Shorthand helper to extract the StreamTransformers type from a MiddlewareTypeConfig or AgentMiddleware.

Helper type to infer tools from a single middleware instance. Extracts the TTools type parameter from AgentMiddleware.

Helper type to infer and merge tools from an array of middleware. Recursively extracts tools from each middleware and combines them into a single tuple.

Shorthand helper to extract the Tools type from a MiddlewareTypeConfig or AgentMiddleware.

Helper type to extract any property from a MiddlewareTypeConfig or AgentMiddleware.

jump targets (internal)

Result type for middleware functions.

Custom detector function that takes content and returns matches

Strategy for handling detected PII

Helper type to resolve an AgentTypeConfig from either:

• An AgentTypeConfig directly
• A ReactAgent instance (using typeof agent)

Helper type to resolve a MiddlewareTypeConfig from either:

• A MiddlewareTypeConfig directly
• An AgentMiddleware instance (using typeof middleware)

Special type to indicate that no response format is provided. When this type is used, the structuredResponse property should not be present in the result.

Runtime information available to middleware (readonly).

Handler function type for wrapping tool calls. Takes a tool call request and returns the tool result or a command.

Discriminated union of ToolCallStream variants, one per tool in TTools. Enables TypeScript to narrow .input and .output when the consumer checks call.name === "someToolName".

Falls back to `

Handler called when tool execution throws.

Return content to surface the error to the model as an error ToolMessage. Return nothing to propagate the original error.

Runtime context automatically injected into tools.

When a tool function has a parameter named tool_runtime with type hint ToolRuntime, the tool execution system will automatically inject an insta

Helper type to convert an array of tools (ClientTool | ServerTool)[] to a MessageToolSet. This maps each tool's name (as a literal type) to its MessageToolDefinition containing the input and output ty

Base input type for .invoke and .stream methods.

Predicate controlling whether a given tool call triggers an interrupt.

Receives a ToolCallRequest and returns true to interrupt or false to auto-approve the tool call.

The request is constructed

Handler function type for wrapping model calls. Takes a model request and returns the AI message response.

Wrapper function type for the wrapModelCall hook. Allows middleware to intercept and modify model execution. This enables you to:

• Modify the request before calling the model (e.g., change system pro

Wrapper function type for the wrapToolCall hook. Allows middleware to intercept and modify tool execution.

Content block for audio data

Content block for multimodal data

Content block for file data

Content block for image data

Content block for plain text data

Content block for video data

Hook type for the afterAgent lifecycle event. Can be either a handler function or an object with a handler and optional jump targets. This hook is called once at the end of the agent invocation.

Hook type for the afterModel lifecycle event. Can be either a handler function or an object with a handler and optional jump targets. This hook is called after each model invocation.

A GraphRunStream with native agent-level projections assigned directly on the instance by createGraphRunStream (via __native transformers).

This is a pure type overlay — no runtime subclass exist

Widened middleware type for agent configuration arrays. Accepts middleware instances with any registered tools or stream transformers while preserving precise inference when middleware arrays are pass

Hook type for the beforeAgent lifecycle event. Can be either a handler function or an object with a handler and optional jump targets. This hook is called once at the start of the agent invocation.

Hook type for the beforeModel lifecycle event. Can be either a handler function or an object with a handler and optional jump targets. This hook is called before each model invocation.

Built-in PII types

Helper type to combine agent stream transformers with middleware stream transformers.

Helper type to combine agent tools with middleware tools into a single readonly array.

Union of all possible decision types.

Default type configuration for middleware. Used when no explicit type parameters are provided.

Function type that dynamically generates a description for a tool call approval request.

Type helper to extract union type from an array of Zod schemas

A headless tool that always interrupts agent execution on the server.

The implementation is provided separately on the client via useStream({ tools: [...] }) using .implement().

Configuration fields for creating a headless tool.

A tool implementation that pairs a headless tool with its execution function.

Created by calling .implement() on a HeadlessTool. Pass to useStream({ tools: [...] }) on the client side.

Helper type to infer the full merged context from an agent, including:

• The agent's own context schema (if provided via contextSchema)
• All middleware context schemas

This matches the context typ

Shorthand helper to extract the raw Context schema type from an AgentTypeConfig or ReactAgent. This returns just the contextSchema type passed to createAgent, not merged with middleware.

For the

Shorthand helper to extract the Middleware type from an AgentTypeConfig or ReactAgent.

Shorthand helper to extract the Response type from an AgentTypeConfig or ReactAgent.

Helper type to infer the full merged state from an agent, including:

• The agent's own state schema (if provided via stateSchema)
• All middleware states

This matches the state type returned by `in

Shorthand helper to extract the raw State schema type from an AgentTypeConfig or ReactAgent. This returns just the stateSchema type passed to createAgent, not merged with middleware.

For the comp

Shorthand helper to extract the StreamTransformers type (the tuple of transformer factories) from an AgentTypeConfig or ReactAgent.

Shorthand helper to extract the Tools type from an AgentTypeConfig or ReactAgent.

Helper type to extract any property from an AgentTypeConfig or ReactAgent.

Helper type to extract input type from context schema (with optional defaults)

Helper type to infer merged input state from an array of middleware (includes built-in state)

Helper type to infer merged state from an array of middleware (includes built-in state)

Helper type to infer the context schema type from a middleware

Helper type to infer the input context schema type from a middleware (with optional defaults)

Helper type to infer merged input context from an array of middleware (with optional defaults)

Helper type to infer merged context from an array of middleware

Shorthand helper to extract the ContextSchema type from a MiddlewareTypeConfig or AgentMiddleware.

Shorthand helper to extract the FullContext type from a MiddlewareTypeConfig or AgentMiddleware.

Helper type to infer the input state schema type from a middleware (all properties optional) This filters out private properties (those starting with underscore) Supports both Zod schemas (InteropZodO

Helper type to infer merged input state from an array of middleware (with optional defaults)

Shorthand helper to extract the Schema type from a MiddlewareTypeConfig or AgentMiddleware.

Helper type to infer the state schema type from a middleware This filters out private properties (those starting with underscore) Supports both Zod schemas (InteropZodObject) and StateSchema from Lang

Helper type to infer merged state from an array of middleware (just the middleware states)

Helper type to infer stream transformers from a single middleware instance.

Helper type to infer and merge stream transformers from an array of middleware.

Shorthand helper to extract the StreamTransformers type from a MiddlewareTypeConfig or AgentMiddleware.

Helper type to infer tools from a single middleware instance. Extracts the TTools type parameter from AgentMiddleware.

Helper type to infer and merge tools from an array of middleware. Recursively extracts tools from each middleware and combines them into a single tuple.

Shorthand helper to extract the Tools type from a MiddlewareTypeConfig or AgentMiddleware.

Helper type to extract any property from a MiddlewareTypeConfig or AgentMiddleware.

jump targets (internal)

Result type for middleware functions.

Custom detector function that takes content and returns matches

Strategy for handling detected PII

Helper type to resolve an AgentTypeConfig from either:

• An AgentTypeConfig directly
• A ReactAgent instance (using typeof agent)

Helper type to resolve a MiddlewareTypeConfig from either:

• A MiddlewareTypeConfig directly
• An AgentMiddleware instance (using typeof middleware)

Special type to indicate that no response format is provided. When this type is used, the structuredResponse property should not be present in the result.

Runtime information available to middleware (readonly).

Handler function type for wrapping tool calls. Takes a tool call request and returns the tool result or a command.

Discriminated union of ToolCallStream variants, one per tool in TTools. Enables TypeScript to narrow .input and .output when the consumer checks call.name === "someToolName".

Falls back to `

Handler called when tool execution throws.

Return content to surface the error to the model as an error ToolMessage. Return nothing to propagate the original error.

Runtime context automatically injected into tools.

When a tool function has a parameter named tool_runtime with type hint ToolRuntime, the tool execution system will automatically inject an insta

Helper type to convert an array of tools (ClientTool | ServerTool)[] to a MessageToolSet. This maps each tool's name (as a literal type) to its MessageToolDefinition containing the input and output ty

Base input type for .invoke and .stream methods.

Predicate controlling whether a given tool call triggers an interrupt.

Receives a ToolCallRequest and returns true to interrupt or false to auto-approve the tool call.

The request is constructed

Handler function type for wrapping model calls. Takes a model request and returns the AI message response.

Wrapper function type for the wrapModelCall hook. Allows middleware to intercept and modify model execution. This enables you to:

• Modify the request before calling the model (e.g., change system pro

Wrapper function type for the wrapToolCall hook. Allows middleware to intercept and modify tool execution.

A headless tool that always interrupts agent execution on the server.

The implementation is provided separately on the client via useStream({ tools: [...] }) using .implement().

Configuration fields for creating a headless tool.

A tool implementation that pairs a headless tool with its execution function.

Created by calling .implement() on a HeadlessTool. Pass to useStream({ tools: [...] }) on the client side.