Human-in-the-loop

The Human-in-the-Loop (HITL) middleware lets you add human oversight to agent tool calls. When a model proposes an action that might require review—for example, writing to a file or executing SQL—the middleware can pause execution and wait for a decision.

It does this by checking each tool call against a configurable policy. If intervention is needed, the middleware issues an interrupt that halts execution. The graph state is saved using LangGraph’s persistence layer, so execution can pause safely and resume later.

A human decision then determines what happens next: the action can be approved as-is (approve), modified before running (edit), rejected with feedback (reject), or responded to directly (respond) for “ask user” style tools.

Interrupt decision types

The middleware defines four built-in ways a human can respond to an interrupt:

Decision Type	Description	Example Use Case
✅ `approve`	Execute the tool with the original arguments as proposed by the agent.	Send an email draft exactly as written
✏️ `edit`	Modify the tool arguments before execution.	Change the recipient before sending an email
❌ `reject`	Skip executing this tool call entirely and return rejection feedback to the agent.	Deny file deletion and explain why
💬 `respond`	Return the human’s message directly as a synthetic tool result, skipping execution, for “ask user” style tools.	Answer an `"ask_user"` prompt with a direct reply

The available decision types for each tool depend on the policy you configure in interrupt_on.

When multiple tool calls are paused at the same time, each action requires a separate decision.

In the default batched mode, provide decisions in the same order as the actions appear in the interrupt request. In per-call mode, each tool call has its own interrupt and takes one decision.

Use reject when the human is denying the requested action. Use respond only when the human is acting as the tool, such as answering an ask_user prompt. Do not use respond to deny side-effecting tools, because its message is treated as a successful tool result.

Tip
When editing tool arguments, make changes conservatively. Significant modifications to the original arguments may cause the model to re-evaluate its approach and potentially execute the tool multiple times or take unexpected actions.

Configuring interrupts

To use HITL, add the middleware to the agent’s middleware list when creating the agent.

You configure it with a mapping of tool actions to the decision types that are allowed for each action. The middleware will interrupt execution when a tool call matches an action in the mapping.

This example uses the default batched mode. To pause once per tool call, see Set the interrupt mode.

from langchain.agents import create_agent
from langchain.agents.middleware import HumanInTheLoopMiddleware # [!code highlight]
from langgraph.checkpoint.memory import InMemorySaver # [!code highlight]


agent = create_agent(
    model="gpt-5.5",
    tools=[write_file, execute_sql, read_data],
    middleware=[
        HumanInTheLoopMiddleware( # [!code highlight]
            interrupt_on={
                "write_file": True,  # All decisions (approve, edit, reject, respond) allowed
                "execute_sql": {"allowed_decisions": ["approve", "reject"]},  # No editing allowed
                "read_data": False, # Safe operation, no approval needed
            },
            # Prefix for interrupt messages - combined with tool name and args to form the full message
            # e.g., "Tool execution pending approval: execute_sql with query='DELETE FROM...'"
            # Individual tools can override this by specifying a "description" in their interrupt config
            description_prefix="Tool execution pending approval",
        ),
    ],
    # Human-in-the-loop requires checkpointing to handle interrupts.
    # In production, use a persistent checkpointer like AsyncPostgresSaver or MongoDBSaver.
    checkpointer=InMemorySaver(),  # [!code highlight]
)

Info
You must configure a checkpointer to persist the graph state across interrupts. In production, use a persistent checkpointer like AsyncPostgresSaver or MongoDBSaver. For testing or prototyping, use InMemorySaver.
When invoking the agent, pass a config that includes the thread ID to associate execution with a conversation thread. See the LangGraph interrupts documentation for details.

Configuration options

`interrupt_on`

Type: dict

Mapping of tool names to approval configs. Values can be True (interrupt with default config), False (auto-approve), or an InterruptOnConfig object.

`description_prefix`

Type: string

Prefix for action request descriptions

`interrupt_mode`

Type: string

How the agent pauses for review. "batched" raises one interrupt per model turn, answered with an ordered list. "per_call" raises one interrupt per gated call, answered by interrupt ID. "per_call" requires langchain>=1.4.6. See Set the interrupt mode.

`edit_notice`

Type: string | None

Text added before the result of a call that a reviewer replaced with an edit decision, so the model knows the call changed. Defaults to a note that a human reviewer replaced the call. Pass None to add nothing.

InterruptOnConfig options:

`allowed_decisions`

Type: list[string]

List of allowed decisions: 'approve', 'edit', 'reject', or 'respond'

`description`

Type: string | callable

Static string or callable function for custom description

`when`

Type: callable

Optional predicate that receives a ToolCallRequest and returns True to interrupt or False to auto-approve. Use it to gate interrupts on a call’s arguments. Requires langchain>=1.3.3.

Conditional interrupts

By default, every tool call listed in interrupt_on pauses for review. To pause only some calls, add a when predicate to a tool’s InterruptOnConfig. The predicate receives a ToolCallRequest and returns True to interrupt or False to auto-approve, so you can gate on the tool’s arguments.

Note
Conditional interrupts require langchain>=1.3.3.

from langchain.agents import create_agent
from langchain.agents.middleware import HumanInTheLoopMiddleware, ToolCallRequest
from langgraph.checkpoint.memory import InMemorySaver


def writes_outside_workspace(request: ToolCallRequest) -> bool:
    """Pause writes to paths outside the workspace directory."""
    path = request.tool_call["args"].get("path", "")
    return not path.startswith("/workspace/")


def is_write_query(request: ToolCallRequest) -> bool:
    """Pause SQL that isn't a read-only SELECT."""
    query = request.tool_call["args"].get("query", "")
    return not query.lstrip().upper().startswith("SELECT")


agent = create_agent(
    model="gpt-5.5",
    tools=[write_file, execute_sql, read_data],
    middleware=[
        HumanInTheLoopMiddleware(
            interrupt_on={
                "write_file": {
                    "allowed_decisions": ["approve", "edit", "reject"],
                    "when": writes_outside_workspace,
                },
                "execute_sql": {
                    "allowed_decisions": ["approve", "reject"],
                    "when": is_write_query,
                },
            },
        ),
    ],
    checkpointer=InMemorySaver(),
)

When the when predicate returns False, the call runs without interrupting. When it returns True, or when you omit when, the call pauses as usual. Calls that evaluate to False never pause, so a reviewer only sees the actions that need a decision.

The predicate runs again when the agent resumes, so it must return the same result for the same call. If it returns False on resume, the tool runs and the reviewer’s answer is ignored, even a reject. In batched mode, if the result changes for only some of the calls, the middleware raises a ValueError.

Set the interrupt mode

The middleware pauses in one of two modes, set with interrupt_mode:

"per_call": Each tool call that needs review raises its own interrupt as it starts. Each interrupt carries a response_schema of the answers it accepts. You answer it with one decision, keyed by the interrupt ID.
"batched" (default): One interrupt per model turn lists every call that needs review. You answer with a list of decisions in the same order.

Use per-call mode for new agents. LangChain plans to make it the default in the next major release. For a side-by-side comparison and the problems per-call mode solves, see the human-in-the-loop overview.

Pause once per tool call

Note
Per-call interrupts require langchain>=1.4.6.

Set interrupt_mode="per_call" to review each tool call separately. This agent gates write_file and execute_sql, and lets read_data run without review:

Google

from langchain.agents import create_agent
from langchain.agents.middleware import HumanInTheLoopMiddleware
from langchain.tools import tool
from langgraph.checkpoint.memory import InMemorySaver


@tool
def write_file(path: str, content: str) -> str:
    """Write content to a file."""
    return f"Wrote {len(content)} characters to {path}"


@tool
def execute_sql(query: str) -> str:
    """Run a SQL query."""
    return "Deleted 42 rows"


@tool
def read_data(table: str) -> str:
    """Read rows from a table."""
    return f"3 rows from {table}"


agent = create_agent(
    model="google_genai:gemini-3.6-flash",
    tools=[write_file, execute_sql, read_data],
    middleware=[
        HumanInTheLoopMiddleware(
            interrupt_on={
                "write_file": True,
                "execute_sql": {"allowed_decisions": ["approve", "reject"]},
                "read_data": False,
            },
            interrupt_mode="per_call",  # [!code highlight]
        ),
    ],
    checkpointer=InMemorySaver(),
)

OpenAI

from langchain.agents import create_agent
from langchain.agents.middleware import HumanInTheLoopMiddleware
from langchain.tools import tool
from langgraph.checkpoint.memory import InMemorySaver


@tool
def write_file(path: str, content: str) -> str:
    """Write content to a file."""
    return f"Wrote {len(content)} characters to {path}"


@tool
def execute_sql(query: str) -> str:
    """Run a SQL query."""
    return "Deleted 42 rows"


@tool
def read_data(table: str) -> str:
    """Read rows from a table."""
    return f"3 rows from {table}"


agent = create_agent(
    model="openai:gpt-5.5",
    tools=[write_file, execute_sql, read_data],
    middleware=[
        HumanInTheLoopMiddleware(
            interrupt_on={
                "write_file": True,
                "execute_sql": {"allowed_decisions": ["approve", "reject"]},
                "read_data": False,
            },
            interrupt_mode="per_call",  # [!code highlight]
        ),
    ],
    checkpointer=InMemorySaver(),
)

Anthropic

from langchain.agents import create_agent
from langchain.agents.middleware import HumanInTheLoopMiddleware
from langchain.tools import tool
from langgraph.checkpoint.memory import InMemorySaver


@tool
def write_file(path: str, content: str) -> str:
    """Write content to a file."""
    return f"Wrote {len(content)} characters to {path}"


@tool
def execute_sql(query: str) -> str:
    """Run a SQL query."""
    return "Deleted 42 rows"


@tool
def read_data(table: str) -> str:
    """Read rows from a table."""
    return f"3 rows from {table}"


agent = create_agent(
    model="anthropic:claude-sonnet-5",
    tools=[write_file, execute_sql, read_data],
    middleware=[
        HumanInTheLoopMiddleware(
            interrupt_on={
                "write_file": True,
                "execute_sql": {"allowed_decisions": ["approve", "reject"]},
                "read_data": False,
            },
            interrupt_mode="per_call",  # [!code highlight]
        ),
    ],
    checkpointer=InMemorySaver(),
)

OpenRouter

from langchain.agents import create_agent
from langchain.agents.middleware import HumanInTheLoopMiddleware
from langchain.tools import tool
from langgraph.checkpoint.memory import InMemorySaver


@tool
def write_file(path: str, content: str) -> str:
    """Write content to a file."""
    return f"Wrote {len(content)} characters to {path}"


@tool
def execute_sql(query: str) -> str:
    """Run a SQL query."""
    return "Deleted 42 rows"


@tool
def read_data(table: str) -> str:
    """Read rows from a table."""
    return f"3 rows from {table}"


agent = create_agent(
    model="openrouter:z-ai/glm-5.2",
    tools=[write_file, execute_sql, read_data],
    middleware=[
        HumanInTheLoopMiddleware(
            interrupt_on={
                "write_file": True,
                "execute_sql": {"allowed_decisions": ["approve", "reject"]},
                "read_data": False,
            },
            interrupt_mode="per_call",  # [!code highlight]
        ),
    ],
    checkpointer=InMemorySaver(),
)

Fireworks

from langchain.agents import create_agent
from langchain.agents.middleware import HumanInTheLoopMiddleware
from langchain.tools import tool
from langgraph.checkpoint.memory import InMemorySaver


@tool
def write_file(path: str, content: str) -> str:
    """Write content to a file."""
    return f"Wrote {len(content)} characters to {path}"


@tool
def execute_sql(query: str) -> str:
    """Run a SQL query."""
    return "Deleted 42 rows"


@tool
def read_data(table: str) -> str:
    """Read rows from a table."""
    return f"3 rows from {table}"


agent = create_agent(
    model="fireworks:accounts/fireworks/models/glm-5p2",
    tools=[write_file, execute_sql, read_data],
    middleware=[
        HumanInTheLoopMiddleware(
            interrupt_on={
                "write_file": True,
                "execute_sql": {"allowed_decisions": ["approve", "reject"]},
                "read_data": False,
            },
            interrupt_mode="per_call",  # [!code highlight]
        ),
    ],
    checkpointer=InMemorySaver(),
)

Baseten

from langchain.agents import create_agent
from langchain.agents.middleware import HumanInTheLoopMiddleware
from langchain.tools import tool
from langgraph.checkpoint.memory import InMemorySaver


@tool
def write_file(path: str, content: str) -> str:
    """Write content to a file."""
    return f"Wrote {len(content)} characters to {path}"


@tool
def execute_sql(query: str) -> str:
    """Run a SQL query."""
    return "Deleted 42 rows"


@tool
def read_data(table: str) -> str:
    """Read rows from a table."""
    return f"3 rows from {table}"


agent = create_agent(
    model="baseten:zai-org/GLM-5.2",
    tools=[write_file, execute_sql, read_data],
    middleware=[
        HumanInTheLoopMiddleware(
            interrupt_on={
                "write_file": True,
                "execute_sql": {"allowed_decisions": ["approve", "reject"]},
                "read_data": False,
            },
            interrupt_mode="per_call",  # [!code highlight]
        ),
    ],
    checkpointer=InMemorySaver(),
)

Ollama

from langchain.agents import create_agent
from langchain.agents.middleware import HumanInTheLoopMiddleware
from langchain.tools import tool
from langgraph.checkpoint.memory import InMemorySaver


@tool
def write_file(path: str, content: str) -> str:
    """Write content to a file."""
    return f"Wrote {len(content)} characters to {path}"


@tool
def execute_sql(query: str) -> str:
    """Run a SQL query."""
    return "Deleted 42 rows"


@tool
def read_data(table: str) -> str:
    """Read rows from a table."""
    return f"3 rows from {table}"


agent = create_agent(
    model="ollama:north-mini-code-1.0",
    tools=[write_file, execute_sql, read_data],
    middleware=[
        HumanInTheLoopMiddleware(
            interrupt_on={
                "write_file": True,
                "execute_sql": {"allowed_decisions": ["approve", "reject"]},
                "read_data": False,
            },
            interrupt_mode="per_call",  # [!code highlight]
        ),
    ],
    checkpointer=InMemorySaver(),
)

Read the interrupts

When the model calls all three tools in one turn, read_data runs right away. The two gated calls each raise an interrupt:

config = {"configurable": {"thread_id": "1"}}
result = agent.invoke(
    {"messages": [{"role": "user", "content": "Archive old orders, then delete them."}]},
    config=config,
    version="v2",
)

# One interrupt per gated tool call, in no fixed order
for interrupt in result.interrupts:
    print(interrupt.value)
# > {'type': 'tool_approval', 'tool_call_id': 'call_write', 'name': 'write_file', 'args': {'path': 'orders_archive.csv', 'content': 'id,total\n1,20\n'}, 'description': "Tool execution requires approval\n\nTool: write_file\nArgs: {'path': 'orders_archive.csv', 'content': 'id,total\\n1,20\\n'}"}
# > {'type': 'tool_approval', 'tool_call_id': 'call_sql', 'name': 'execute_sql', 'args': {'query': "DELETE FROM orders WHERE created_at < '2025-01-01'"}, 'description': 'Tool execution requires approval\n\nTool: execute_sql\nArgs: {\'query\': "DELETE FROM orders WHERE created_at < \'2025-01-01\'"}'}

Each interrupt value has type: "tool_approval", so a client can tell it apart from other interrupts. The tool_call_id, name, and args fields identify the call under review.

Read the response schema

Each interrupt also carries a response_schema: a JSON schema with one branch per decision the tool allows, matched on the decision’s type. The execute_sql schema has approve and reject branches. The write_file schema has all four. For a tool that allows one decision, response_schema is that decision’s schema alone, with no oneOf.

Each tab shows one write_file branch, an answer that matches it, and the ToolMessage the model receives. For readability, the branches below expand $ref definitions in place and omit the title and description fields.

approve

Run the call as proposed.

{"properties": {"type": {"const": "approve", "type": "string"}}, "required": ["type"], "type": "object"}

{"type": "approve"}

Wrote 14 characters to orders_archive.csv

edit

Run the call with new arguments. edited_action.name must stay write_file, and edited_action.args must match write_file’s argument schema. Arguments the tool does not declare are ignored, as they are when the model sends them.

{
  "properties": {
    "type": {"const": "edit", "type": "string"},
    "edited_action": {
      "properties": {
        "name": {"const": "write_file", "type": "string"},
        "args": {
          "properties": {
            "path": {"type": "string"},
            "content": {"type": "string"}
          },
          "required": ["path", "content"],
          "type": "object"
        }
      },
      "required": ["name", "args"],
      "type": "object"
    }
  },
  "required": ["type", "edited_action"],
  "type": "object"
}

{
    "type": "edit",
    "edited_action": {
        "name": "write_file",
        "args": {"path": "archive/orders_2024.csv", "content": "id,total\n1,20\n"},
    },
}

The model receives the tool’s result after a note that a reviewer replaced the call. Set edit_notice on the middleware to change the note, or pass None to remove it.

Note: a human reviewer replaced this tool call before it ran. The call recorded in your message is the one you produced, not the one that executed. This was intentional and authorized. Do not re-issue your original call. Executed instead: write_file with arguments {"path": "archive/orders_2024.csv", "content": "id,total\n1,20\n"}.

Tool response:
Wrote 14 characters to archive/orders_2024.csv

reject

Skip the call. The message is optional and tells the model why.

{"properties": {"type": {"const": "reject", "type": "string"}, "message": {"type": "string"}}, "required": ["type"], "type": "object"}

{"type": "reject", "message": "Write to the archive/ folder instead."}

User rejected the tool call for `write_file` with reason: Write to the archive/ folder instead.

respond

Skip the call and answer in place of the tool. The message is required. The model receives it as a successful tool result, so use reject to deny a call.

{"properties": {"type": {"const": "respond", "type": "string"}, "message": {"type": "string"}}, "required": ["type", "message"], "type": "object"}

{"type": "respond", "message": "Saved orders_archive.csv to the shared drive myself."}

Saved orders_archive.csv to the shared drive myself.

Resume each call by ID

Answer each interrupt with one decision, keyed by its ID:

from langgraph.types import Command

decisions = {}
for interrupt in result.interrupts:
    if interrupt.value["name"] == "execute_sql":
        decisions[interrupt.id] = {
            "type": "reject",
            "message": "Archive the rows before deleting them.",
        }
    else:
        decisions[interrupt.id] = {"type": "approve"}

result = agent.invoke(Command(resume=decisions), config=config, version="v2")

You can also answer interrupts one at a time. An interrupt you do not answer stays pending, and its tool runs once it gets an answer.

Warning
If one of several answers sent together is invalid, the tools for the others can run while their interrupts still show as pending. Answering them again runs those tools twice. Check answers against response_schema before you send them together, or send one at a time.

Correct an invalid answer

LangGraph checks each answer against the interrupt’s response_schema before it saves the answer. An invalid answer raises a ValidationError and nothing is saved, so you can send a corrected answer to the same interrupt ID. Here, the edit leaves out content:

from pydantic import ValidationError

write_file_interrupt = next(
    interrupt for interrupt in result.interrupts if interrupt.value["name"] == "write_file"
)
edit = {
    "type": "edit",
    "edited_action": {"name": "write_file", "args": {"path": "archive/orders_2024.csv"}},
}

try:
    agent.invoke(Command(resume={write_file_interrupt.id: edit}), config=config, version="v2")
except ValidationError as error:
    for problem in error.errors():
        print(".".join(str(part) for part in problem["loc"]), problem["msg"])
# > edit.edited_action.args.content Field required

# Nothing was saved, so answer the same interrupt again
edit["edited_action"]["args"]["content"] = "id,total\n1,20\n"
result = agent.invoke(
    Command(resume={write_file_interrupt.id: edit}), config=config, version="v2"
)

Each error names the field that failed, so you can show it to the reviewer. For a tool whose arguments are a raw JSON schema instead of a Pydantic model, response_schema shows the arguments, but LangGraph does not check edits against them.

Order middleware for per-call review

In per-call mode, the review happens as the tool call starts, inside the tool-call middleware chain:

List HumanInTheLoopMiddleware before tool retry or error-handling middleware, such as ToolRetryMiddleware and ToolErrorMiddleware, so it wraps them. Otherwise a retried call asks the reviewer again, and an invalid answer is retried or returned to the model as a tool error.
When the agent resumes, middleware listed before HumanInTheLoopMiddleware runs again up to the pause, so it must not have side effects before it calls its handler. The tool runs only after the answer.
A when predicate runs as each call starts, so the ToolCallRequest it receives includes the tool.

Responding to interrupts

When you invoke the agent, it runs until it either completes or an interrupt is raised. An interrupt is triggered when a tool call matches the policy you configured in interrupt_on. With version="v2", the result is a GraphOutput with an interrupts attribute containing the actions that require review. You can then present those actions to a reviewer and resume execution once decisions are provided.

The examples in this section and the decision types below use the default batched mode. For per-call mode, see Pause once per tool call.

from langgraph.types import Command

# Human-in-the-loop leverages LangGraph's persistence layer.
# You must provide a thread ID to associate the execution with a conversation thread,
# so the conversation can be paused and resumed (as is needed for human review).
config = {"configurable": {"thread_id": "some_id"}} # [!code highlight]
# Run the graph until the interrupt is hit.
result = agent.invoke(
    {
        "messages": [
            {
                "role": "user",
                "content": "Delete old records from the database",
            }
        ]
    },
    config=config, # [!code highlight]
    version="v2", # [!code highlight]
)

# result is a GraphOutput with .value and .interrupts
print(result.interrupts)  # [!code highlight]
# > (
# >    Interrupt(
# >       value={
# >          'action_requests': [
# >             {
# >                'name': 'execute_sql',
# >                'args': {'query': 'DELETE FROM records WHERE created_at < NOW() - INTERVAL \'30 days\';'},
# >                'description': 'Tool execution pending approval\n\nTool: execute_sql\nArgs: {...}'
# >             }
# >          ],
# >          'review_configs': [
# >             {
# >                'action_name': 'execute_sql',
# >                'allowed_decisions': ['approve', 'reject']
# >             }
# >          ]
# >       }
# >    ),
# > )


# Resume with approval decision
agent.invoke(
    Command( # [!code highlight]
        resume={"decisions": [{"type": "approve"}]}  # or "reject" [!code highlight]
    ), # [!code highlight]
    config=config, # Same thread ID to resume the paused conversation
    version="v2",
)

Decision types

approve

Use approve to approve the tool call as-is and execute it without changes.

agent.invoke(
    Command(
        # Decisions are provided as a list, one per action under review.
        # The order of decisions must match the order of actions
        # in the interrupt request.
        resume={
            "decisions": [
                {
                    "type": "approve",
                }
            ]
        }
    ),
    config=config,  # Same thread ID to resume the paused conversation
    version="v2",
)

edit

Use `edit` to modify the tool call before execution.
Provide the edited action with the new tool name and arguments.

```python
agent.invoke(
    Command(
        # Decisions are provided as a list, one per action under review.
        # The order of decisions must match the order of actions
        # in the interrupt request.
        resume={
            "decisions": [
                {
                    "type": "edit",
                    # Edited action with tool name and args
                    "edited_action": {
                        # Tool name to call.
                        # Will usually be the same as the original action.
                        "name": "new_tool_name",
                        # Arguments to pass to the tool.
                        "args": {"key1": "new_value", "key2": "original_value"},
                    }
                }
            ]
        }
    ),
    config=config,  # Same thread ID to resume the paused conversation
    version="v2",
)
```

Tip
When editing tool arguments, make changes conservatively. Significant modifications to the original arguments may cause the model to re-evaluate its approach and potentially execute the tool multiple times or take unexpected actions.

reject

Use `reject` to deny the tool call and provide feedback instead of execution. The tool is not executed.

```python
agent.invoke(
    Command(
        # Decisions are provided as a list, one per action under review.
        # The order of decisions must match the order of actions
        # in the interrupt request.
        resume={
            "decisions": [
                {
                    "type": "reject",
                    # Optional: explain why the action was rejected
                    # and whether the agent should retry a different approach.
                    "message": "User rejected this action. Do not retry this tool call.",
                }
            ]
        }
    ),
    config=config,  # Same thread ID to resume the paused conversation
    version="v2",
)
```

The message is added to the conversation as feedback to help the agent understand why the action was rejected and what it should do instead. When you omit message, the middleware uses a default rejection message that tells the model the tool was not executed and not to retry the same tool call unless the user asks. For side-effecting tools, provide a domain-specific message that is explicit about whether the agent should abandon the action, ask a follow-up question, or try a safer alternative.

respond

Use `respond` for "ask user" style tools where the tool's real implementation is the human's reply. The `message` content is returned directly as the tool result; the tool itself is not executed.

```python
agent.invoke(
    Command(
        # Decisions are provided as a list, one per action under review.
        # The order of decisions must match the order of actions
        # in the interrupt request.
        resume={
            "decisions": [
                {
                    "type": "respond",
                    # The human's reply, returned directly as the tool result
                    "message": "Blue.",
                }
            ]
        }
    ),
    config=config,  # Same thread ID to resume the paused conversation
    version="v2",
)
```

The message is returned to the agent as a successful ToolMessage. Use respond when the tool is intentionally a placeholder for human input, for example, an ask_user tool that prompts for clarification. Do not use respond to deny a proposed action, because it tells the model that the tool completed successfully.

Multiple decisions

When multiple actions are under review, provide a decision for each action in the same order as they appear in the interrupt:

{
    "decisions": [
        {"type": "approve"},
        {
            "type": "edit",
            "edited_action": {
                "name": "tool_name",
                "args": {"param": "new_value"}
            }
        },
        {
            "type": "reject",
            "message": "This action is not allowed"
        }
    ]
}

Streaming with human-in-the-loop

You can stream real-time updates while the agent runs and handles interrupts using stream_events(). Use stream.messages to stream LLM tokens and stream.values to check agent state snapshots for interrupts.

from langgraph.types import Command

config = {"configurable": {"thread_id": "some_id"}}

# Stream agent progress and LLM tokens until interrupt
stream = agent.stream_events(
    {"messages": [{"role": "user", "content": "Delete old records from the database"}]},
    config=config,
    version="v3",  # [!code highlight]
)
for message in stream.messages:  # [!code highlight]
    for token in message.text:  # [!code highlight]
        print(token, end="", flush=True)

# Check whether the run paused for human input
if stream.interrupted:  # [!code highlight]
    print(f"\n\nInterrupt: {stream.interrupts}")  # [!code highlight]

# Resume with streaming after human decision
stream = agent.stream_events(
    Command(resume={"decisions": [{"type": "approve"}]}),
    config=config,
    version="v3",  # [!code highlight]
)
for message in stream.messages:  # [!code highlight]
    for token in message.text:
        print(token, end="", flush=True)

In per-call mode, stream.interrupts holds one interrupt per gated tool call. Resume with a dict keyed by interrupt ID:

decisions = {interrupt.id: {"type": "approve"} for interrupt in stream.interrupts}
stream = agent.stream_events(Command(resume=decisions), config=config, version="v3")
for message in stream.messages:
    for token in message.text:
        print(token, end="", flush=True)

See the Streaming guide for more details on stream modes.

Execution lifecycle

The middleware defines an after_model hook that runs after the model generates a response but before any tool calls are executed:

The agent invokes the model to generate a response.
The middleware inspects the response for tool calls.
If any calls require human input, the middleware builds a HITLRequest with action_requests and review_configs and calls interrupt.
The agent waits for human decisions.
Based on the HITLResponse decisions, the middleware executes approved or edited calls, synthesizes ToolMessage’s for rejected calls, returns human replies directly as ToolMessage’s for respond decisions, and resumes execution.

In per-call mode, the after_model hook does nothing. The middleware instead pauses each gated call as it runs, with one interrupt per call, and runs the tool once that call’s decision arrives.

Custom HITL logic

For more specialized workflows, you can build custom HITL logic directly using the interrupt primitive and middleware abstraction.

Review the execution lifecycle above to understand how to integrate interrupts into the agent’s operation.

Guardrails Install LangChain