Interrupts

The interrupt system enables human-in-the-loop workflows by allowing users to pause agent execution and request human input before continuing. When an interrupt is raised, the agent stops its loop and returns control to the user. The user in turn provides a response to the agent. The agent then continues its execution starting from the point of interruption. Users can raise interrupts from either hook callbacks or tool definitions. The general flow looks as follows:

flowchart TD
    A[Invoke Agent] --> B[Execute Hook/Tool]
    B --> C{Interrupts Raised?}
    C -->|No| D[Continue Agent Loop]
    C -->|Yes| E[Stop Agent Loop]
    E --> F[Return Interrupts]
    F --> G[Respond to Interrupts]
    G --> H[Execute Hook/Tool with Responses]
    H --> I{New Interrupts?}
    I -->|Yes| E
    I -->|No| D

Hooks

Users can raise interrupts within their hook callbacks to pause agent execution at specific life-cycle events in the agentic loop.

Python

Both BeforeToolCallEvent and BeforeToolsEvent are interruptible. Interrupting on a BeforeToolCallEvent allows users to intercept an individual tool call before execution, while interrupting on BeforeToolsEvent pauses the entire batch of tool calls before any of them execute — useful for approving a whole set of tool calls at once.

BeforeToolCallEvent

import json
from typing import Any

from strands import Agent, tool
from strands.hooks import BeforeToolCallEvent, HookProvider, HookRegistry


@tool
def delete_files(paths: list[str]) -> bool:
    # Implementation here
    pass


@tool
def inspect_files(paths: list[str]) -> dict[str, Any]:
    # Implementation here
    pass


class ApprovalHook(HookProvider):
    def __init__(self, app_name: str) -> None:
        self.app_name = app_name

    def register_hooks(self, registry: HookRegistry, **kwargs: Any) -> None:
        registry.add_callback(BeforeToolCallEvent, self.approve)

    def approve(self, event: BeforeToolCallEvent) -> None:
        if event.tool_use["name"] != "delete_files":
            return

        approval = event.interrupt(f"{self.app_name}-approval", reason={"paths": event.tool_use["input"]["paths"]})
        if approval.lower() != "y":
            event.cancel_tool = "User denied permission to delete files"


agent = Agent(
    hooks=[ApprovalHook("myapp")],
    system_prompt="You delete files older than 5 days",
    tools=[delete_files, inspect_files],
    callback_handler=None,
)

paths = ["a/b/c.txt", "d/e/f.txt"]
result = agent(f"paths=<{paths}>")

while True:
    if result.stop_reason != "interrupt":
        break

    responses = []
    for interrupt in result.interrupts:
        if interrupt.name == "myapp-approval":
            user_input = input(f"Do you want to delete {interrupt.reason['paths']} (y/N): ")
            responses.append({
                "interruptResponse": {
                    "interruptId": interrupt.id,
                    "response": user_input
                }
            })

    result = agent(responses)

print(f"MESSAGE: {json.dumps(result.message)}")

BeforeToolsEvent

from typing import Any

from strands import Agent, tool
from strands.hooks import BeforeToolsEvent, HookProvider, HookRegistry


@tool
def delete_files(paths: list[str]) -> bool:
    # Implementation here
    pass


class BatchApprovalHook(HookProvider):
    def __init__(self, app_name: str) -> None:
        self.app_name = app_name

    def register_hooks(self, registry: HookRegistry, **kwargs: Any) -> None:
        registry.add_callback(BeforeToolsEvent, self.approve)

    def approve(self, event: BeforeToolsEvent) -> None:
        dangerous_tools = [
            content["toolUse"]["name"]
            for content in event.message["content"]
            if "toolUse" in content and content["toolUse"]["name"] == "delete_files"
        ]
        if not dangerous_tools:
            return

        approval = event.interrupt(f"{self.app_name}-batch-approval", reason={"tools": dangerous_tools})
        if approval.lower() != "y":
            event.cancel = "Batch cancelled by user"


agent = Agent(
    hooks=[BatchApprovalHook("myapp")],
    system_prompt="You delete files older than 5 days",
    tools=[delete_files],
    callback_handler=None,
)

result = agent("Delete a/b/c.txt and d/e/f.txt")

while result.stop_reason == "interrupt":
    responses = []
    for interrupt in result.interrupts:
        if interrupt.name == "myapp-batch-approval":
            user_input = input(f"Approve running {interrupt.reason['tools']}? (y/N): ")
            responses.append({"interruptResponse": {"interruptId": interrupt.id, "response": user_input}})

    result = agent(responses)

Setting event.cancel (to True for a default message, or a string for a custom one) produces an error tool result for every tool in the batch and skips execution entirely, so no per-tool BeforeToolCallEvent fires.

TypeScript

Both BeforeToolCallEvent and BeforeToolsEvent are interruptible. Interrupting on a BeforeToolCallEvent allows users to intercept individual tool calls before execution, while BeforeToolsEvent allows intercepting the entire batch of tool calls before any execute.

BeforeToolCallEvent

import { Agent, tool, BeforeToolCallEvent } from '@strands-agents/sdk'
import { z } from 'zod'

const deleteFiles = tool({
  name: 'delete_files',
  description: 'Delete files at the given paths',
  inputSchema: z.object({ paths: z.array(z.string()) }),
  callback: (input) => {
    // Implementation here
    return true
  },
})

const inspectFiles = tool({
  name: 'inspect_files',
  description: 'Inspect files at the given paths',
  inputSchema: z.object({ paths: z.array(z.string()) }),
  callback: (input) => {
    // Implementation here
    return {}
  },
})

const agent = new Agent({
  systemPrompt: 'You delete files older than 5 days',
  tools: [deleteFiles, inspectFiles],
})

agent.addHook(BeforeToolCallEvent, (event) => {
  if (event.toolUse.name !== 'delete_files') return

  const approval = event.interrupt<string>({
    name: 'myapp-approval',
    reason: { paths: (event.toolUse.input as { paths: string[] }).paths },
  })
  if (approval.toLowerCase() !== 'y') {
    event.cancel = 'User denied permission to delete files'
  }
})

const paths = ['a/b/c.txt', 'd/e/f.txt']
let result = await agent.invoke(`paths=<${JSON.stringify(paths)}>`)

while (result.stopReason === 'interrupt') {
  const responses = result.interrupts!.map((interrupt) => ({
    interruptResponse: {
      interruptId: interrupt.id,
      // In a real app, collect user input here
      response: 'y',
    },
  }))

  result = await agent.invoke(responses)
}

console.log('MESSAGE:', JSON.stringify(result.lastMessage))

BeforeToolsEvent

import { Agent, BeforeToolsEvent } from '@strands-agents/sdk'

const agent = new Agent({
  tools: [
    /* ... */
  ],
})

agent.addHook(BeforeToolsEvent, (event) => {
  const dangerousTools = event.message.content
    .filter((block) => block.type === 'toolUseBlock')
    .filter((block) => ['delete_files'].includes(block.name))

  if (dangerousTools.length > 0) {
    const response = event.interrupt<{ approved: boolean }>({
      name: 'batch_approval',
      reason: `Approve ${dangerousTools.length} dangerous tool calls?`,
    })
    if (!response.approved) {
      event.cancel = 'Batch cancelled by user'
    }
  }
})

Components

Interrupts in Strands are comprised of the following components:

Python

• event.interrupt - Raises an interrupt with a unique name and optional reason
  • The name must be unique across all interrupt calls configured on the same event (BeforeToolCallEvent or BeforeToolsEvent). In the example above, we demonstrate using app_name to namespace the interrupt call. This is particularly helpful if you plan to vend your hooks to other users.
  • You can assign additional context for raising the interrupt to the reason field. Note, the reason must be JSON-serializable.
• result.stop_reason - Check if agent stopped due to “interrupt”
• result.interrupts - List of interrupts that were raised
  • Each interrupt contains the user provided name and reason, along with an instance id.
• interruptResponse - Content block type for configuring the interrupt responses.
  • Each response is uniquely identified by their interrupt’s id and will be returned from the associated interrupt call when invoked the second time around. Note, the response must be JSON-serializable.
• event.cancel_tool (BeforeToolCallEvent) - Cancel a single tool call based on the interrupt response
  • You can either set cancel_tool to True or provide a custom cancellation message.
• event.cancel (BeforeToolsEvent) - Cancel every tool call in the batch based on the interrupt response
  • You can either set cancel to True or provide a custom cancellation message.

For additional details on each of these components, refer to the Python API Reference.

TypeScript

• BeforeToolCallEvent / BeforeToolsEvent: hook events that expose the ability to interrupt via the interrupt method
  • event.interrupt({ name, reason? }): halts the agent loop. name is a string identifier and reason is an optional JSON-serializable value providing context for why the interrupt was raised.
  • The name must be unique across all interrupt calls configured on the same event. In the example above, we demonstrate using a namespace prefix for the interrupt call. This is particularly helpful if you plan to vend your hooks to other users.
  • event.cancel: cancel tool execution based on the interrupt response. Set to true for a default message or provide a custom cancellation message string.
• AgentResult: returned by invoke() / stream(), contains interrupt information when the agent pauses
  • result.stopReason: check if agent stopped due to 'interrupt'
  • result.interrupts: array of Interrupt objects, each containing the user-provided name and reason, along with a unique id
• InterruptResponseContent: content block type for resuming from an interrupt
  • Pass an array of these to agent.invoke() to resume. Each response is keyed by the interrupt’s id and will be returned from the associated interrupt() call when the tool/hook re-executes. The response must be JSON-serializable.

Rules

Strands enforces the following rules for interrupts:

Python

• All hooks configured on the interrupted event will execute
• All hooks configured on the interrupted event are allowed to raise an interrupt
• A single hook can raise multiple interrupts but only one at a time
  • In other words, within a single hook, you can interrupt, respond to that interrupt, and then proceed to interrupt again.
• All tools running concurrently are interruptible
• All tools running concurrently that are not interrupted will execute
• When an interrupt fires from BeforeToolCallEvent, AfterToolCallEvent does not fire for that tool, but AfterToolsEvent still fires
• AfterToolsEvent fires once per event-loop cycle rather than once per logical batch. A per-tool interrupt splits a batch across cycles, so it fires on the interrupt cycle (carrying only the results collected so far) and again on resume (carrying the results produced that cycle) — a hook with side effects there can run more than once for one assistant message
• When an interrupt fires from BeforeToolsEvent, no tool in the batch executes (and no BeforeToolCallEvent fires) until the interrupt is answered; on resume the batch hook re-runs and, if not interrupted again, the tools execute

TypeScript

• All hooks configured on the interrupted event will execute
• All hooks configured on the interrupted event are allowed to raise an interrupt
• A single hook can raise multiple interrupts but only one at a time
  • In other words, within a single hook, you can interrupt, respond to that interrupt, and then proceed to interrupt again.
• When an interrupt fires from BeforeToolCallEvent, AfterToolCallEvent does not fire for that tool, but AfterToolsEvent still fires
• AfterToolsEvent fires once per event-loop cycle rather than once per logical batch. A per-tool interrupt splits a batch across cycles, so it fires on the interrupt cycle (carrying only the results collected so far) and again on resume (carrying the results produced that cycle) — a hook with side effects there can run more than once for one assistant message
• When an interrupt fires mid-batch, completed tool results are preserved so the agent skips the model call on resume and only executes remaining tools
• Both assistant and tool result messages are appended only after tool execution completes, preventing dangling toolUse blocks without matching results

Tools

Users can also raise interrupts from their tool definitions.

Python

from typing import Any

from strands import Agent, tool
from strands.types.tools import ToolContext


class DeleteTool:
    def __init__(self, app_name: str) -> None:
        self.app_name = app_name

    @tool(context=True)
    def delete_files(self, tool_context: ToolContext, paths: list[str]) -> bool:
        approval = tool_context.interrupt(f"{self.app_name}-approval", reason={"paths": paths})
        if approval.lower() != "y":
            return False

        # Implementation here

        return True


@tool
def inspect_files(paths: list[str]) -> dict[str, Any]:
    # Implementation here
    pass


agent = Agent(
    system_prompt="You delete files older than 5 days",
    tools=[DeleteTool("myapp").delete_files, inspect_files],
    callback_handler=None,
)

...

Interrupts are not supported in direct tool calls (i.e., calls such as agent.tool.my_tool()).

TypeScript

The tool callback receives a context parameter (the second argument) which provides the interrupt method.

import { Agent, tool } from '@strands-agents/sdk'
import { z } from 'zod'

const deleteFiles = tool({
  name: 'delete_files',
  description: 'Delete files at the given paths',
  inputSchema: z.object({ paths: z.array(z.string()) }),
  callback: (input, context) => {
    const approval = context.interrupt<string>({
      name: 'myapp-approval',
      reason: { paths: input.paths },
    })
    if (approval.toLowerCase() !== 'y') return false

    // Implementation here

    return true
  },
})

const inspectFiles = tool({
  name: 'inspect_files',
  description: 'Inspect files at the given paths',
  inputSchema: z.object({ paths: z.array(z.string()) }),
  callback: (input) => {
    // Implementation here
    return {}
  },
})

const agent = new Agent({
  systemPrompt: 'You delete files older than 5 days',
  tools: [deleteFiles, inspectFiles],
})

// ...

Components

Tool interrupts work like hook interrupts, with two differences: tools receive context instead of event, and interrupt names need only be unique within a tool definition rather than across all hooks on an event. For more on tool context, see ToolContext.

Python

• tool_context - Strands object that defines the interrupt call
• tool_context.interrupt - Raises an interrupt with a unique name and optional reason
  • The name must be unique only among interrupt calls configured in the same tool definition. It is still advisable however to namespace your interrupts so as to more easily distinguish the calls when constructing responses outside the agent.

TypeScript

• ToolContext: the second argument passed to the tool callback, providing access to the interrupt method
  • context.interrupt({ name, reason? }): halts the agent loop. name is a string identifier and reason is an optional JSON-serializable value.
  • The name must be unique only among interrupt calls configured in the same tool definition. It is still advisable however to namespace your interrupts so as to more easily distinguish the calls when constructing responses outside the agent.

Rules

Strands enforces the following rules for tool interrupts:

Python

• All tools running concurrently will execute
• All tools running concurrently are interruptible
• A single tool can raise multiple interrupts but only one at a time
  • In other words, within a single tool, you can interrupt, respond to that interrupt, and then proceed to interrupt again.

TypeScript

• A single tool can raise multiple interrupts but only one at a time
  • In other words, within a single tool, you can interrupt, respond to that interrupt, and then proceed to interrupt again.
• When an interrupt fires mid-batch, completed tool results are preserved so the agent skips the model call on resume and only executes remaining tools

Session Management

Users can session manage their interrupts and respond back at a later time under a new agent session. Additionally, users can session manage the responses to avoid repeated interrupts on subsequent tool calls.

Python

##### server.py #####

import json
from typing import Any

from strands import Agent, tool
from strands.agent import AgentResult
from strands.hooks import BeforeToolCallEvent, HookProvider, HookRegistry
from strands.session import FileSessionManager
from strands.types.agent import AgentInput


@tool
def delete_files(paths: list[str]) -> bool:
    # Implementation here
    pass


@tool
def inspect_files(paths: list[str]) -> dict[str, Any]:
    # Implementation here
    pass


class ApprovalHook(HookProvider):
    def __init__(self, app_name: str) -> None:
        self.app_name = app_name

    def register_hooks(self, registry: HookRegistry, **kwargs: Any) -> None:
        registry.add_callback(BeforeToolCallEvent, self.approve)

    def approve(self, event: BeforeToolCallEvent) -> None:
        if event.tool_use["name"] != "delete_files":
            return

        if event.agent.state.get(f"{self.app_name}-approval") == "t":  # (t)rust
            return

        approval = event.interrupt(f"{self.app_name}-approval", reason={"paths": event.tool_use["input"]["paths"]})
        if approval.lower() not in ["y", "t"]:
            event.cancel_tool = "User denied permission to delete files"

        event.agent.state.set(f"{self.app_name}-approval", approval.lower())


def server(prompt: AgentInput) -> AgentResult:
    agent = Agent(
        hooks=[ApprovalHook("myapp")],
        session_manager=FileSessionManager(session_id="myapp", storage_dir="/path/to/storage"),
        system_prompt="You delete files older than 5 days",
        tools=[delete_files, inspect_files],
        callback_handler=None,
    )
    return agent(prompt)

##### client.py #####

def client(paths: list[str]) -> AgentResult:
    result = server(f"paths=<{paths}>")

    while True:
        if result.stop_reason != "interrupt":
            break

        responses = []
        for interrupt in result.interrupts:
            if interrupt.name == "myapp-approval":
                user_input = input(f"Do you want to delete {interrupt.reason['paths']} (t/y/N): ")
                responses.append({
                    "interruptResponse": {
                        "interruptId": interrupt.id,
                        "response": user_input
                    }
                })

        result = server(responses)

    return result


paths = ["a/b/c.txt", "d/e/f.txt"]
result = client(paths)
print(f"MESSAGE: {json.dumps(result.message)}")

TypeScript

import {
  Agent,
  tool,
  SessionManager,
  FileStorage,
  BeforeToolCallEvent,
} from '@strands-agents/sdk'
import { z } from 'zod'

const deleteFiles = tool({
  name: 'delete_files',
  description: 'Delete files at the given paths',
  inputSchema: z.object({ paths: z.array(z.string()) }),
  callback: (input) => {
    // Implementation here
    return true
  },
})

const inspectFiles = tool({
  name: 'inspect_files',
  description: 'Inspect files at the given paths',
  inputSchema: z.object({ paths: z.array(z.string()) }),
  callback: (input) => {
    // Implementation here
    return {}
  },
})

// Server function — creates a fresh agent with session management each call
async function server(
  prompt: string | { interruptResponse: { interruptId: string; response: unknown } }[]
) {
  const agent = new Agent({
    systemPrompt: 'You delete files older than 5 days',
    tools: [deleteFiles, inspectFiles],
    sessionManager: new SessionManager({
      sessionId: 'myapp',
      storage: { snapshot: new FileStorage('/path/to/storage') },
    }),
  })

  agent.addHook(BeforeToolCallEvent, (event) => {
    if (event.toolUse.name !== 'delete_files') return

    // Check if user already trusted this approval
    if (event.agent.appState.get('myapp-approval') === 't') return

    const approval = event.interrupt<string>({
      name: 'myapp-approval',
      reason: { paths: (event.toolUse.input as { paths: string[] }).paths },
    })
    if (!['y', 't'].includes(approval.toLowerCase())) {
      event.cancel = 'User denied permission to delete files'
    }

    event.agent.appState.set('myapp-approval', approval.toLowerCase())
  })

  return agent.invoke(prompt)
}

// Client function
async function client(paths: string[]) {
  let result = await server(`paths=<${JSON.stringify(paths)}>`)

  while (result.stopReason === 'interrupt') {
    const responses = result.interrupts!.map((interrupt) => ({
      interruptResponse: {
        interruptId: interrupt.id,
        // In a real app, collect user input here
        response: 'y',
      },
    }))

    result = await server(responses)
  }

  return result
}

const paths = ['a/b/c.txt', 'd/e/f.txt']
const result = await client(paths)
console.log('MESSAGE:', JSON.stringify(result.lastMessage))

Components

Session managing interrupts involves the following key components:

Python

• session_manager - Automatically persists the agent interrupt state between tear down and start up
  • See Session Management for more.
• agent.state - General purpose key-value store that can be used to persist interrupt responses
  • On subsequent tool calls, you can reference the responses stored in agent.state to decide whether another interrupt is necessary. See Agent State for more.

TypeScript

• sessionManager - Automatically persists the agent interrupt state between tear down and start up
  • See Session Management for more.
• agent.appState - General purpose key-value store that can be used to persist interrupt responses
  • On subsequent tool calls, you can reference the responses stored in appState to decide whether another interrupt is necessary. See Agent State for more.

MCP Elicitation

To collect additional information from a user during an MCP tool call, use elicitation. An MCP server sends an elicitation request to the connecting client, which is handled by an elicitation callback. See MCP Elicitation for details.

Multi-Agents

Interrupts are supported in multi-agent patterns, enabling human-in-the-loop workflows across agent orchestration systems. The interfaces mirror those used for single-agent interrupts. You can raise interrupts from BeforeNodeCallEvent hooks executed before each node or from within the nodes themselves. Session management is also supported, allowing you to persist and resume your interrupted multi-agents.

Swarm

A Swarm is a collaborative agent orchestration system where multiple agents work together as a team to solve complex tasks. The following example demonstrates interrupting your swarm invocation through a BeforeNodeCallEvent hook.

Python

import json

from strands import Agent
from strands.hooks import BeforeNodeCallEvent, HookProvider, HookRegistry
from strands.multiagent import Swarm, Status


class ApprovalHook(HookProvider):
    def __init__(self, app_name: str) -> None:
        self.app_name = app_name

    def register_hooks(self, registry: HookRegistry) -> None:
        registry.add_callback(BeforeNodeCallEvent, self.approve)

    def approve(self, event: BeforeNodeCallEvent) -> None:
        if event.node_id != "cleanup":
            return

        approval = event.interrupt(f"{self.app_name}-approval", reason={"resources": "example"})
        if approval.lower() != "y":
            event.cancel_node = "User denied permission to cleanup resources"


swarm = Swarm(
    [
        Agent(name="cleanup", system_prompt="You clean up resources older than 5 days.", callback_handler=None),
    ],
    hooks=[ApprovalHook("myapp")],
)

result = swarm("Clean up my resources")
while result.status == Status.INTERRUPTED:
    responses = []
    for interrupt in result.interrupts:
        if interrupt.name == "myapp-approval":
            user_input = input(f"Do you want to cleanup {interrupt.reason['resources']} (y/N): ")
            responses.append({
                "interruptResponse": {
                    "interruptId": interrupt.id,
                    "response": user_input,
                },
            })

    result = swarm(responses)

print(f"MESSAGE: {json.dumps(result.results['cleanup'].result.message, indent=2)}")

TypeScript

import { Agent, Swarm, Status, BeforeNodeCallEvent } from '@strands-agents/sdk'

const cleanupAgent = new Agent({
  id: 'cleanup',
  systemPrompt: 'You clean up resources older than 5 days.',
})

const swarm = new Swarm({ nodes: [cleanupAgent], start: 'cleanup' })

swarm.addHook(BeforeNodeCallEvent, (event) => {
  if (event.nodeId !== 'cleanup') return

  const approval = event.interrupt<string>({
    name: 'myapp-approval',
    reason: { resources: 'example' },
  })
  if (approval.toLowerCase() !== 'y') {
    event.cancel = 'User denied permission to cleanup resources'
  }
})

let result = await swarm.invoke('Clean up my resources')

while (result.status === Status.INTERRUPTED) {
  const responses = result.interrupts!.map((interrupt) => ({
    interruptResponse: {
      interruptId: interrupt.id,
      // In a real app, collect user input here
      response: 'y',
    },
  }))

  result = await swarm.invoke(responses)
}

console.log('MESSAGE:', JSON.stringify(result.results, null, 2))

Swarms also support interrupts raised from within the nodes themselves following any of the single-agent interrupt patterns outlined above.

Components

Python

• event.interrupt - Raises an interrupt with a unique name and optional reason
  • The name must be unique across all interrupt calls configured on the BeforeNodeCallEvent. In the example above, we demonstrate using app_name to namespace the interrupt call. This is particularly helpful if you plan to vend your hooks to other users.
  • You can assign additional context for raising the interrupt to the reason field. Note, the reason must be JSON-serializable.
• result.status - Check if the swarm stopped due to Status.INTERRUPTED
• result.interrupts - List of interrupts that were raised
  • Each interrupt contains the user provided name and reason, along with an instance id.
• interruptResponse - Content block type for configuring the interrupt responses.
  • Each response is uniquely identified by their interrupt’s id and will be returned from the associated interrupt call when invoked the second time around. Note, the response must be JSON-serializable.
• event.cancel_node - Cancel node execution based on interrupt response
  • You can either set cancel_node to True or provide a custom cancellation message.

TypeScript

• BeforeNodeCallEvent: orchestrator hook event that exposes the ability to interrupt before a node runs
  • event.interrupt({ name, reason? }): halts the orchestrator. name is a string identifier and reason is an optional JSON-serializable value providing context for why the interrupt was raised.
  • The name must be unique across all interrupt calls configured on the same event. In the example above, we demonstrate using a namespace prefix for the interrupt call. This is particularly helpful if you plan to vend your hooks to other users.
  • event.cancel: cancel node execution based on the interrupt response. Set to true for a default message or provide a custom cancellation message string.
• MultiAgentResult: returned by invoke() / stream(), contains interrupt information when the orchestrator pauses
  • result.status: check if the swarm stopped due to Status.INTERRUPTED
  • result.interrupts: array of Interrupt objects, each with name, reason, and a unique id. Each interrupt’s source field is 'multiagent-hook' when raised from BeforeNodeCallEvent.
• InterruptResponseContent: content block type for resuming from an interrupt
  • Pass an array of these to swarm.invoke() to resume. The orchestrator routes each response to the node that raised the matching interrupt.

Rules

Strands enforces the following rules for interrupts in swarm:

• All hooks configured on the interrupted event will execute
• All hooks configured on the interrupted event are allowed to raise an interrupt
• A single hook can raise multiple interrupts but only one at a time
  • In other words, within a single hook, you can interrupt, respond to that interrupt, and then proceed to interrupt again.
• A single node can raise multiple interrupts following any of the single-agent interrupt patterns outlined above.

Graph

A Graph is a deterministic agent orchestration system based on a directed graph, where agents are nodes executed according to edge dependencies. The following example demonstrates interrupting your graph invocation through a BeforeNodeCallEvent hook.

Python

import json

from strands import Agent
from strands.hooks import BeforeNodeCallEvent, HookProvider, HookRegistry
from strands.multiagent import GraphBuilder, Status


class ApprovalHook(HookProvider):
    def __init__(self, app_name: str) -> None:
        self.app_name = app_name

    def register_hooks(self, registry: HookRegistry) -> None:
        registry.add_callback(BeforeNodeCallEvent, self.approve)

    def approve(self, event: BeforeNodeCallEvent) -> None:
        if event.node_id != "cleanup":
            return

        approval = event.interrupt(f"{self.app_name}-approval", reason={"resources": "example"})
        if approval.lower() != "y":
            event.cancel_node = "User denied permission to cleanup resources"


inspector_agent = Agent(name="inspector", system_prompt="You inspect resources.", callback_handler=None)
cleanup_agent = Agent(name="cleanup", system_prompt="You clean up resources older than 5 days.", callback_handler=None)

builder = GraphBuilder()
builder.add_node(inspector_agent, "inspector")
builder.add_node(cleanup_agent, "cleanup")
builder.add_edge("inspector", "cleanup")
builder.set_entry_point("inspector")
builder.set_hook_providers([ApprovalHook("myapp")])
graph = builder.build()

result = graph("Inspect and clean up my resources")
while result.status == Status.INTERRUPTED:
    responses = []
    for interrupt in result.interrupts:
        if interrupt.name == "myapp-approval":
            user_input = input(f"Do you want to cleanup {interrupt.reason['resources']} (y/N): ")
            responses.append({
                "interruptResponse": {
                    "interruptId": interrupt.id,
                    "response": user_input,
                },
            })

    result = graph(responses)

print(f"MESSAGE: {json.dumps(result.results['cleanup'].result.message, indent=2)}")

TypeScript

import { Agent, Graph, Status, BeforeNodeCallEvent } from '@strands-agents/sdk'

const inspectorAgent = new Agent({
  id: 'inspector',
  systemPrompt: 'You inspect resources.',
})
const cleanupAgent = new Agent({
  id: 'cleanup',
  systemPrompt: 'You clean up resources older than 5 days.',
})

const graph = new Graph({
  nodes: [inspectorAgent, cleanupAgent],
  edges: [['inspector', 'cleanup']],
})

graph.addHook(BeforeNodeCallEvent, (event) => {
  if (event.nodeId !== 'cleanup') return

  const approval = event.interrupt<string>({
    name: 'myapp-approval',
    reason: { resources: 'example' },
  })
  if (approval.toLowerCase() !== 'y') {
    event.cancel = 'User denied permission to cleanup resources'
  }
})

let result = await graph.invoke('Inspect and clean up my resources')

while (result.status === Status.INTERRUPTED) {
  const responses = result.interrupts!.map((interrupt) => ({
    interruptResponse: {
      interruptId: interrupt.id,
      // In a real app, collect user input here
      response: 'y',
    },
  }))

  result = await graph.invoke(responses)
}

console.log('MESSAGE:', JSON.stringify(result.results, null, 2))

Graphs also support interrupts raised from within the nodes themselves following any of the single-agent interrupt patterns outlined above.

Components

Python

• event.interrupt - Raises an interrupt with a unique name and optional reason
  • The name must be unique across all interrupt calls configured on the BeforeNodeCallEvent. In the example above, we demonstrate using app_name to namespace the interrupt call. This is particularly helpful if you plan to vend your hooks to other users.
  • You can assign additional context for raising the interrupt to the reason field. Note, the reason must be JSON-serializable.
• result.status - Check if the graph stopped due to Status.INTERRUPTED
• result.interrupts - List of interrupts that were raised
  • Each interrupt contains the user provided name and reason, along with an instance id.
• interruptResponse - Content block type for configuring the interrupt responses
  • Each response is uniquely identified by their interrupt’s id and will be returned from the associated interrupt call when invoked the second time around. Note, the response must be JSON-serializable.
• event.cancel_node - Cancel node execution based on interrupt response
  • You can either set cancel_node to True or provide a custom cancellation message.

TypeScript

• BeforeNodeCallEvent: orchestrator hook event that exposes the ability to interrupt before a node runs
  • event.interrupt({ name, reason? }): halts the orchestrator. name is a string identifier and reason is an optional JSON-serializable value providing context for why the interrupt was raised.
  • The name must be unique across all interrupt calls configured on the same event. In the example above, we demonstrate using a namespace prefix for the interrupt call. This is particularly helpful if you plan to vend your hooks to other users.
  • event.cancel: cancel node execution based on the interrupt response. Set to true for a default message or provide a custom cancellation message string.
• MultiAgentResult: returned by invoke() / stream(), contains interrupt information when the orchestrator pauses
  • result.status: check if the graph stopped due to Status.INTERRUPTED
  • result.interrupts: array of Interrupt objects, each with name, reason, and a unique id. Each interrupt’s source field is 'multiagent-hook' when raised from BeforeNodeCallEvent.
• InterruptResponseContent: content block type for resuming from an interrupt
  • Pass an array of these to graph.invoke() to resume. The orchestrator routes each response to the node that raised the matching interrupt; concurrent nodes already in flight run to completion.

Rules

Strands enforces the following rules for interrupts in graph:

• All hooks configured on the interrupted event will execute
• All hooks configured on the interrupted event are allowed to raise an interrupt
• A single hook can raise multiple interrupts but only one at a time
  • In other words, within a single hook, you can interrupt, respond to that interrupt, and then proceed to interrupt again.
• A single node can raise multiple interrupts following any of the single-agent interrupt patterns outlined above
• All nodes running concurrently will execute
• All nodes running concurrently are interruptible

Tags

event-loop8hooks9tool-execution12error-handling10