> Fetch clean Markdown by appending `.md` to any page URL under https://signalwire.com/docs or requesting it with the HTTP header `Accept: text/markdown`. The root index at https://signalwire.com/docs/llms.txt lists the available documentation indexes.

# set_tool_response

> Split the response into a factual result and a speaking instruction.

[functionresult]: /docs/server-sdks/reference/python/agents/function-result

[setresponse]: /docs/server-sdks/reference/python/agents/function-result/set-response

Set the structured form of the response, separating what the tool did from what
the model should say next. The response becomes an object with up to two keys:

```json
{
  "tool_result": "status: on hold",
  "tool_prompt": "Tell the caller you are placing them on hold."
}
```

Splitting them keeps the model from reading a status line aloud, and keeps the
spoken instruction from being mistaken for data. For a single instruction
string, use [`set_response()`][setresponse].

## Parameters

**`tool_result`** `str | None` — default: None

What the tool did: a factual status line for the model to reason from, such
as `"payment declined"` or `"3 seats left"`. Omit it when there is nothing
to report beyond the instruction.

---

**`tool_prompt`** `str | None` — default: None

What the model should now say, as an instruction in the second person. Omit
it for a silent, status-only result.

---

## Returns

[`FunctionResult`][functionresult] — self, for chaining.

## Example

```python {10-13}
from signalwire import AgentBase
from signalwire import FunctionResult

agent = AgentBase(name="my-agent", route="/agent")
agent.set_prompt_text("You are a helpful assistant.")

@agent.tool(name="check_seats", description="Check remaining seats for a showing")
def check_seats(args, raw_data):
    seats = 3  # look this up in your booking system
    return FunctionResult().set_tool_response(
        tool_result=f"seats remaining: {seats}",
        tool_prompt="Tell the caller how many seats are left and ask how many they want.",
    )

agent.serve()
```

> Split the response into a factual result and a speaking instruction.