Deep AgentsDeep Agents 0.7 · Python 3.11+
0%
1
Curious builder0 XP earned · 300 to level 2
0 daysFinish a lesson to begin
Badge collection0 of 6 unlocked
28 small wins to finish your pathNext lesson →

Edit decisions: changing a tool call before it runs

An edit decision resumes a paused tool call with arguments you changed, so the tool runs with your version instead of the model's.

Last updated: 29 Sep, 2026 · Deep Agents 0.7

Approve and reject, from Human-in-the-loop: approving a booking, are yes or no. Often the agent's call is nearly right: the right hotel for the wrong number of nights. An edit fixes it in place, without another round with the model.

The edit decision syntax

python
request = result["__interrupt__"][0].value["action_requests"][0]
edited = {"type": "edit",
          "edited_action": {"name": request["name"], "args": {**request["args"], "nights": 2}}}
agent.invoke(Command(resume={"decisions": [edited]}), config=config)

The same booking agent

Start trip.py with search_travel and book_hotel, as in Human-in-the-loop: approving a booking:

python
from langchain.tools import tool

CATALOG = {
    "paris": {
        "flight": ["Return flight Delhi to Paris: 42,000 rupees"],
        "hotel": ["Seine Budget Inn, Latin Quarter: 5,200 rupees a night",
                  "Hotel Lumiere, Montmartre: 7,500 rupees a night",
                  "Le Grand Opera Hotel: 16,000 rupees a night"],
        "sight": ["Eiffel Tower summit: 3,100 rupees", "Louvre Museum: 2,000 rupees",
                  "Seine river cruise: 1,500 rupees", "Versailles day trip: 2,600 rupees",
                  "Montmartre walking tour: free"],
        "food": ["Cafe breakfast and bistro dinner: 3,000 rupees a day"],
    },
}


@tool
def search_travel(city: str, kind: str) -> str:
    """Search the travel catalog. kind is "flight", "hotel", "sight" or "food". Prices are in rupees."""
    entries = CATALOG.get(city.lower(), {}).get(kind)
    return "\n".join(entries) if entries else f"The catalog has no {kind} entries for {city}."


@tool
def book_hotel(hotel: str, nights: int) -> str:
    """Book a hotel for a number of nights. This spends the traveller's money."""
    return f"Booked {hotel} for {nights} nights."
python
from deepagents import create_deep_agent
from langchain.chat_models import init_chat_model

model = init_chat_model("groq:openai/gpt-oss-120b", temperature=0, max_retries=6)
python
from langgraph.checkpoint.memory import InMemorySaver
from langgraph.types import Command

agent = create_deep_agent(
    model=model,
    tools=[search_travel, book_hotel],
    interrupt_on={"book_hotel": True},          # pause before every booking
    checkpointer=InMemorySaver(),               # a pause needs saved state
    system_prompt="You book trips. Look up prices with search_travel. Reply in one short sentence.",
)

Cutting a booking from 3 nights to 2

The traveller asks for 3 nights; the person reviewing knows the dates changed and books 2.

ExampleAPI keytrip.py, continued
config = {"configurable": {"thread_id": "edit"}}
result = agent.invoke({"messages": [{"role": "user", "content": "Book Hotel Lumiere, Montmartre for 3 nights."}]}, config=config)
request = result["__interrupt__"][0].value["action_requests"][0]
print("agent asked:", request["name"], request["args"])

edited = {"type": "edit", "edited_action": {"name": "book_hotel", "args": {**request["args"], "nights": 2}}}
result = agent.invoke(Command(resume={"decisions": [edited]}), config=config)
print("tool said:  ", result["messages"][-2].text)
print("agent said: ", result["messages"][-1].text)

What the edit changed

  • The agent asked for Hotel Lumiere with nights 3, as the traveller said.
  • The tool ran with 2 nights. The tool message starts with a note from the middleware telling the model that a reviewer replaced its call, so it does not send the original again, then the booking result.
  • The agent's reply reports the booking the tool made, not the one it had asked for.

Edit vs reject and ask again

editreject, then a new request
Model callsNone extraAt least one more
Who decides the final argumentsThe reviewerThe model, again
Good forSmall fixes: a number, a dateWrong tool or wrong idea

Where edits help

  • Fixing a quantity, a date or a recipient before an action runs.
  • Letting an assistant draft a message that a person corrects before it is sent.
  • Keeping a human in charge of the final values without rewriting the whole request.
Watch out. Keep edits small. The docs warn that a large change to the arguments can make the model rethink its approach and call the tool again.
Try it yourself
  • Edit the hotel name to "Seine Budget Inn, Latin Quarter" instead of the nights.
  • Allow only approve and reject for book_hotel, then try the edit and read the error.
  • Print result["messages"][-3].tool_calls to see what the model originally asked for.
PreviousHuman-in-the-loop: approving a booking

Little by little, you're building something great.