Skip to content

Tool calls and image support in the Responses API. #519

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
janimo merged 1 commit into from
May 14, 2026
Merged

Tool calls and image support in the Responses API. #519

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
316 changes: 251 additions & 65 deletions catgrad-llm/src/types/openai.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -389,12 +389,45 @@ pub mod responses {
}

/// A single item in a structured Responses API input array.
#[derive(Debug, Clone, Serialize, Deserialize, TypedBuilder, PartialEq)]
pub struct ResponseInputItem {
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
#[serde(untagged)]
pub enum ResponseInputItem {
Message(ResponseInputMessageItem),
FunctionCallOutput(ResponseInputFunctionCallOutputItem),
}

#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
pub struct ResponseInputMessageItem {
#[serde(rename = "type", default = "default_response_input_message_type")]
pub item_type: ResponseInputMessageItemType,
pub role: String,
pub content: ResponseInputMessageContent,
}

#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
pub struct ResponseInputFunctionCallOutputItem {
#[serde(rename = "type")]
pub item_type: ResponseInputFunctionCallOutputItemType,
pub call_id: String,
pub output: String,
}

#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
#[serde(rename_all = "snake_case")]
pub enum ResponseInputMessageItemType {
Message,
}

#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
#[serde(rename_all = "snake_case")]
pub enum ResponseInputFunctionCallOutputItemType {
FunctionCallOutput,
}

fn default_response_input_message_type() -> ResponseInputMessageItemType {
ResponseInputMessageItemType::Message
}

/// Content payload for a structured input message.
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
#[serde(untagged)]
Expand All @@ -404,11 +437,12 @@ pub mod responses {
}

/// A supported content part in structured response input.
#[derive(Debug, Clone, Serialize, Deserialize, TypedBuilder, PartialEq, Eq)]
pub struct ResponseInputContentPart {
#[serde(rename = "type")]
pub content_type: String,
pub text: String,
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
#[serde(tag = "type", rename_all = "snake_case")]
pub enum ResponseInputContentPart {
InputText { text: String },
OutputText { text: String },
InputImage { image_url: super::ImageUrl },
}

impl ResponseRequest {
Expand Down Expand Up @@ -448,10 +482,23 @@ pub mod responses {
type Error = LLMError;

fn try_from(value: ResponseInputItem) -> Result<Self, Self::Error> {
Ok(super::ChatMessage::builder()
.role(value.role)
.content(Some(value.content.try_into()?))
.build())
match value {
ResponseInputItem::Message(ResponseInputMessageItem { role, content, .. }) => {
Ok(super::ChatMessage::builder()
.role(role)
.content(Some(content.try_into()?))
.build())
}
ResponseInputItem::FunctionCallOutput(ResponseInputFunctionCallOutputItem {
call_id,
output,
..
}) => Ok(super::ChatMessage::builder()
.role("tool".to_string())
.content(Some(super::MessageContent::Text(output)))
.tool_call_id(Some(call_id))
.build()),
}
}
}

Expand All @@ -475,11 +522,12 @@ pub mod responses {
type Error = LLMError;

fn try_from(value: ResponseInputContentPart) -> Result<Self, Self::Error> {
match value.content_type.as_str() {
"input_text" | "output_text" => Ok(Self::Text { text: value.text }),
other => Err(LLMError::UnsupportedWireConversion(format!(
"Unsupported responses input content type `{other}`"
))),
match value {
ResponseInputContentPart::InputText { text }
| ResponseInputContentPart::OutputText { text } => Ok(Self::Text { text }),
ResponseInputContentPart::InputImage { image_url } => {
Ok(Self::ImageUrl { image_url })
}
}
}
}
Expand Down Expand Up @@ -619,7 +667,7 @@ pub mod responses {
mod tests {
use super::*;
use crate::types::Message;
use crate::types::openai::{ChatMessage, ContentPart, MessageContent};
use crate::types::openai::{ChatMessage, ContentPart, ImageUrl, MessageContent};
use serde_json::json;

#[test]
Expand Down Expand Up @@ -679,28 +727,27 @@ pub mod responses {
ResponseRequest::builder()
.model("gpt-4.1-mini".to_string())
.input(ResponseInput::Items(vec![
ResponseInputItem::builder()
.role("developer".to_string())
.content(ResponseInputMessageContent::Parts(vec![
ResponseInputContentPart::builder()
.content_type("input_text".to_string())
.text("You are helpful.".to_string())
.build(),
ResponseInputContentPart::builder()
.content_type("input_text".to_string())
.text("Answer briefly.".to_string())
.build(),
]))
.build(),
ResponseInputItem::builder()
.role("user".to_string())
.content(ResponseInputMessageContent::Parts(vec![
ResponseInputContentPart::builder()
.content_type("input_text".to_string())
.text("Tell a story about a unicorn.".to_string())
.build(),
]))
.build(),
ResponseInputItem::Message(ResponseInputMessageItem {
item_type: ResponseInputMessageItemType::Message,
role: "developer".to_string(),
content: ResponseInputMessageContent::Parts(vec![
ResponseInputContentPart::InputText {
text: "You are helpful.".to_string(),
},
ResponseInputContentPart::InputText {
text: "Answer briefly.".to_string(),
},
]),
}),
ResponseInputItem::Message(ResponseInputMessageItem {
item_type: ResponseInputMessageItemType::Message,
role: "user".to_string(),
content: ResponseInputMessageContent::Parts(vec![
ResponseInputContentPart::InputText {
text: "Tell a story about a unicorn.".to_string(),
},
]),
}),
]))
.instructions(Some("You are a nice LLM".to_string()))
.tools(Some(vec![json!({
Expand Down Expand Up @@ -952,17 +999,17 @@ pub mod responses {
fn response_request_accepts_output_text_input_parts() {
let request = ResponseRequest::builder()
.model("gpt-4.1-mini".to_string())
.input(ResponseInput::Items(vec![
ResponseInputItem::builder()
.role("assistant".to_string())
.content(ResponseInputMessageContent::Parts(vec![
ResponseInputContentPart::builder()
.content_type("output_text".to_string())
.text("Previous answer".to_string())
.build(),
]))
.build(),
]))
.input(ResponseInput::Items(vec![ResponseInputItem::Message(
ResponseInputMessageItem {
item_type: ResponseInputMessageItemType::Message,
role: "assistant".to_string(),
content: ResponseInputMessageContent::Parts(vec![
ResponseInputContentPart::OutputText {
text: "Previous answer".to_string(),
},
]),
},
)]))
.build();

let messages = request.to_messages().unwrap();
Expand All @@ -979,25 +1026,118 @@ pub mod responses {
);
}

#[test]
fn response_request_accepts_message_items_without_type() {
let parsed: ResponseRequest = serde_json::from_value(json!({
"model": "gpt-4.1-mini",
"input": [
{
"role": "user",
"content": [
{"type": "input_text", "text": "Hello"}
]
}
]
}))
.unwrap();

assert_eq!(
parsed,
ResponseRequest::builder()
.model("gpt-4.1-mini".to_string())
.input(ResponseInput::Items(vec![ResponseInputItem::Message(
ResponseInputMessageItem {
item_type: ResponseInputMessageItemType::Message,
role: "user".to_string(),
content: ResponseInputMessageContent::Parts(vec![
ResponseInputContentPart::InputText {
text: "Hello".to_string(),
},
]),
},
)]))
.build()
);
}

#[test]
fn response_request_accepts_input_image_parts() {
let parsed: ResponseRequest = serde_json::from_value(json!({
"model": "gpt-4.1-mini",
"input": [
{
"role": "user",
"content": [
{"type": "input_text", "text": "Describe this image"},
{"type": "input_image", "image_url": {"url": "https://example.com/image.png"}}
]
}
]
}))
.unwrap();

assert_eq!(
parsed,
ResponseRequest::builder()
.model("gpt-4.1-mini".to_string())
.input(ResponseInput::Items(vec![ResponseInputItem::Message(
ResponseInputMessageItem {
item_type: ResponseInputMessageItemType::Message,
role: "user".to_string(),
content: ResponseInputMessageContent::Parts(vec![
ResponseInputContentPart::InputText {
text: "Describe this image".to_string(),
},
ResponseInputContentPart::InputImage {
image_url: ImageUrl {
url: "https://example.com/image.png".to_string(),
},
},
]),
},
)]))
.build()
);

let messages = parsed.to_messages().unwrap();
assert_eq!(
messages,
vec![Message::openai(
ChatMessage::builder()
.role("user".to_string())
.content(Some(MessageContent::Parts(vec![
ContentPart::Text {
text: "Describe this image".to_string(),
},
ContentPart::ImageUrl {
image_url: ImageUrl {
url: "https://example.com/image.png".to_string(),
},
},
])))
.build(),
)]
);
}

#[test]
fn response_request_converts_to_chat_messages() {
let request = ResponseRequest::builder()
.model("gpt-4.1-mini".to_string())
.input(ResponseInput::Items(vec![
ResponseInputItem::builder()
.role("user".to_string())
.content(ResponseInputMessageContent::Parts(vec![
ResponseInputContentPart::builder()
.content_type("input_text".to_string())
.text("Hello".to_string())
.build(),
ResponseInputContentPart::builder()
.content_type("input_text".to_string())
.text(" world".to_string())
.build(),
]))
.build(),
]))
.input(ResponseInput::Items(vec![ResponseInputItem::Message(
ResponseInputMessageItem {
item_type: ResponseInputMessageItemType::Message,
role: "user".to_string(),
content: ResponseInputMessageContent::Parts(vec![
ResponseInputContentPart::InputText {
text: "Hello".to_string(),
},
ResponseInputContentPart::InputText {
text: " world".to_string(),
},
]),
},
)]))
.instructions(Some("Be concise.".to_string()))
.build();

Expand All @@ -1022,5 +1162,51 @@ pub mod responses {
]
);
}

#[test]
fn response_request_supports_function_call_output_items() {
let parsed: ResponseRequest = serde_json::from_value(json!({
"model": "gpt-4.1-mini",
"input": [
{
"type": "function_call_output",
"call_id": "call_123",
"output": "{\"result\":\"ok\"}"
}
]
}))
.unwrap();

assert_eq!(
parsed,
ResponseRequest::builder()
.model("gpt-4.1-mini".to_string())
.input(ResponseInput::Items(vec![
ResponseInputItem::FunctionCallOutput(
ResponseInputFunctionCallOutputItem {
item_type:
ResponseInputFunctionCallOutputItemType::FunctionCallOutput,
call_id: "call_123".to_string(),
output: "{\"result\":\"ok\"}".to_string(),
}
)
]))
.build()
);

let messages = parsed.to_messages().unwrap();
assert_eq!(
messages,
vec![Message::openai(
ChatMessage::builder()
.role("tool".to_string())
.content(Some(MessageContent::Text(
"{\"result\":\"ok\"}".to_string()
)))
.tool_call_id(Some("call_123".to_string()))
.build(),
)]
);
}
}
}