Skip to content

Use Claude Code with your ChatGPT, Kimi, Cursor or Grok subscription via a local Anthropic-compatible proxy

Rust

echo Responses tool configuration

Raine Virta67d48a242341
diff --git a/src/openai_compat/mod.rs b/src/openai_compat/mod.rs
index 591b62a..40bb17c 100644
--- a/src/openai_compat/mod.rs
+++ b/src/openai_compat/mod.rs
@@ -7,7 +7,7 @@ use axum::{
http::StatusCode,
response::{IntoResponse, Response},
};
-use serde_json::json;
+use serde_json::{Value, json};
use crate::provider::{ProviderError, ProviderErrorKind};
@@ -30,6 +30,21 @@ impl OpenAiSurface {
}
}
+#[derive(Debug, Clone)]
+pub struct OpenAiResponseMetadata {
+ pub tools: Vec<Value>,
+ pub tool_choice: Value,
+}
+
+impl Default for OpenAiResponseMetadata {
+ fn default() -> Self {
+ Self {
+ tools: Vec::new(),
+ tool_choice: json!("auto"),
+ }
+ }
+}
+
#[derive(Debug, Clone)]
pub struct OpenAiError {
pub status: StatusCode,
diff --git a/src/openai_compat/request.rs b/src/openai_compat/request.rs
index 2d97eb9..659651b 100644
--- a/src/openai_compat/request.rs
+++ b/src/openai_compat/request.rs
@@ -7,7 +7,7 @@ use crate::{
registry::normalize_incoming_model,
};
-use super::{OpenAiError, OpenAiSurface};
+use super::{OpenAiError, OpenAiResponseMetadata, OpenAiSurface};
const MAX_MESSAGES: usize = 1_024;
const MAX_TOOLS: usize = 128;
@@ -48,6 +48,7 @@ pub struct ParsedOpenAiRequest {
pub normalized_model: String,
pub stream: bool,
pub include_usage: bool,
+ pub response_metadata: OpenAiResponseMetadata,
}
pub fn extract_model(body: &Value) -> Result<(String, String), OpenAiError> {
@@ -118,6 +119,22 @@ pub fn parse_request(
let max_tokens = parse_max_tokens(surface, object)?;
let tools = parse_tools(object.get("tools"), surface)?;
let tool_choice = parse_tool_choice(object.get("tool_choice"), &tools, surface)?;
+ let response_metadata = if surface == OpenAiSurface::Responses {
+ OpenAiResponseMetadata {
+ tools: object
+ .get("tools")
+ .and_then(Value::as_array)
+ .cloned()
+ .unwrap_or_default(),
+ tool_choice: object
+ .get("tool_choice")
+ .filter(|value| !value.is_null())
+ .cloned()
+ .unwrap_or_else(|| json!("auto")),
+ }
+ } else {
+ OpenAiResponseMetadata::default()
+ };
let effort = parse_effort(surface, object)?;
validate_cursor(provider, session_id, stream, &messages, &tools)?;
@@ -156,6 +173,7 @@ pub fn parse_request(
normalized_model,
stream,
include_usage,
+ response_metadata,
})
}
diff --git a/src/openai_compat/response.rs b/src/openai_compat/response.rs
index 1bee815..aef8bf7 100644
--- a/src/openai_compat/response.rs
+++ b/src/openai_compat/response.rs
@@ -1,6 +1,6 @@
use serde_json::{Value, json};
-use super::{OpenAiError, OpenAiSurface};
+use super::{OpenAiError, OpenAiResponseMetadata, OpenAiSurface};
fn upstream_invalid(message: impl Into<String>, _param: Option<impl Into<String>>) -> OpenAiError {
OpenAiError::upstream_protocol(message)
@@ -301,6 +301,7 @@ pub fn buffered_response(
response_id: &str,
model: &str,
created: u64,
+ response_metadata: &OpenAiResponseMetadata,
) -> Result<Value, OpenAiError> {
let mut state = AnthropicAccumulator::default();
for event in events {
@@ -314,7 +315,13 @@ pub fn buffered_response(
}
match surface {
OpenAiSurface::ChatCompletions => Ok(chat_response(&state, response_id, model, created)),
- OpenAiSurface::Responses => Ok(responses_response(&state, response_id, model, created)),
+ OpenAiSurface::Responses => Ok(responses_response(
+ &state,
+ response_id,
+ model,
+ created,
+ response_metadata,
+ )),
}
}
@@ -400,6 +407,7 @@ pub fn responses_response(
response_id: &str,
model: &str,
created: u64,
+ response_metadata: &OpenAiResponseMetadata,
) -> Value {
let mut output = Vec::new();
for block in state
@@ -455,6 +463,8 @@ pub fn responses_response(
"model":model,
"output":output,
"parallel_tool_calls":false,
+ "tool_choice":response_metadata.tool_choice,
+ "tools":response_metadata.tools,
"error":null,
"incomplete_details":if incomplete { json!({"reason":"max_output_tokens"}) } else { Value::Null },
"usage":state.usage.responses_value(),
@@ -641,6 +651,7 @@ mod tests {
"resp_test",
"grok-4.5",
1,
+ &OpenAiResponseMetadata::default(),
)
.unwrap();
assert!(
@@ -670,6 +681,7 @@ mod tests {
"chatcmpl_test",
"kimi-k2.6",
1,
+ &OpenAiResponseMetadata::default(),
)
.unwrap();
assert_eq!(response["choices"][0]["message"]["content"], "hello");
@@ -689,6 +701,7 @@ mod tests {
"resp_test",
"kimi-k2.6",
1,
+ &OpenAiResponseMetadata::default(),
)
.unwrap();
assert_eq!(response["object"], "response");
diff --git a/src/openai_compat/stream.rs b/src/openai_compat/stream.rs
index 6cceb6f..5599fe1 100644
--- a/src/openai_compat/stream.rs
+++ b/src/openai_compat/stream.rs
@@ -17,7 +17,8 @@ use crate::{
};
use super::{
- MAX_PROVIDER_STREAM_BYTES, MAX_SSE_EVENT_BYTES, OpenAiError, OpenAiSurface,
+ MAX_PROVIDER_STREAM_BYTES, MAX_SSE_EVENT_BYTES, OpenAiError, OpenAiResponseMetadata,
+ OpenAiSurface,
response::{
AnthropicAccumulator, BlockKind, SseEvent, buffered_response, chat_citation,
chat_finish_reason, hosted_search_action, normalized_arguments, responses_citation,
@@ -74,6 +75,7 @@ pub async fn openai_response(
generation: Generation,
stream: bool,
include_usage: bool,
+ response_metadata: OpenAiResponseMetadata,
traffic: Option<Arc<TrafficCapture>>,
) -> Result<Response, OpenAiError> {
let model = generation.resolved_model.clone();
@@ -84,18 +86,28 @@ pub async fn openai_response(
let created = current_seconds();
if stream {
return Ok(streaming_response(
- surface,
generation.body,
- include_usage,
- response_id,
- model,
- created,
+ Renderer::new(
+ surface,
+ include_usage,
+ response_id,
+ model,
+ created,
+ response_metadata,
+ ),
traffic,
));
}
let bytes = collect_generation(generation.body).await?;
let events = decode_all(&bytes)?;
- let value = buffered_response(surface, &events, &response_id, &model, created)?;
+ let value = buffered_response(
+ surface,
+ &events,
+ &response_id,
+ &model,
+ created,
+ &response_metadata,
+ )?;
if let Some(capture) = traffic.as_ref() {
capture.write_json("071-openai-downstream-response", &value);
}
@@ -146,12 +158,8 @@ fn decode_all(bytes: &[u8]) -> Result<Vec<SseEvent>, OpenAiError> {
}
fn streaming_response(
- surface: OpenAiSurface,
body: GenerationBody,
- include_usage: bool,
- response_id: String,
- model: String,
- created: u64,
+ renderer: Renderer,
traffic: Option<Arc<TrafficCapture>>,
) -> Response {
let outcome = NativeResponseOutcome::default();
@@ -161,7 +169,7 @@ fn streaming_response(
GenerationBody::LiveSse(body) => body,
},
decoder: SseDecoder::default(),
- renderer: Renderer::new(surface, include_usage, response_id, model, created),
+ renderer,
pending: VecDeque::new(),
finished: false,
bytes: 0,
@@ -321,6 +329,7 @@ struct Renderer {
created: u64,
sequence: u64,
state: AnthropicAccumulator,
+ response_metadata: OpenAiResponseMetadata,
}
impl Renderer {
@@ -330,6 +339,7 @@ impl Renderer {
response_id: String,
model: String,
created: u64,
+ response_metadata: OpenAiResponseMetadata,
) -> Self {
Self {
surface,
@@ -339,6 +349,7 @@ impl Renderer {
created,
sequence: 0,
state: AnthropicAccumulator::default(),
+ response_metadata,
}
}
@@ -437,8 +448,13 @@ impl Renderer {
let mut out = Vec::new();
match kind {
"message_start" => {
- let response =
- response_shell(&self.response_id, &self.model, self.created, "in_progress");
+ let response = response_shell(
+ &self.response_id,
+ &self.model,
+ self.created,
+ "in_progress",
+ &self.response_metadata,
+ );
out.push(self.responses_event("response.created", json!({"response":response})));
out.push(
self.responses_event("response.in_progress", json!({"response":response})),
@@ -600,8 +616,13 @@ impl Renderer {
}
}
"message_stop" => {
- let response =
- responses_response(&self.state, &self.response_id, &self.model, self.created);
+ let response = responses_response(
+ &self.state,
+ &self.response_id,
+ &self.model,
+ self.created,
+ &self.response_metadata,
+ );
let kind = if self.state.stop_reason.as_deref() == Some("max_tokens") {
"response.incomplete"
} else {
@@ -623,8 +644,13 @@ impl Renderer {
Bytes::from_static(b"data: [DONE]\n\n"),
],
OpenAiSurface::Responses => {
- let mut response =
- response_shell(&self.response_id, &self.model, self.created, "failed");
+ let mut response = response_shell(
+ &self.response_id,
+ &self.model,
+ self.created,
+ "failed",
+ &self.response_metadata,
+ );
response["error"] = json!({"code":error.code,"message":error.message});
vec![self.responses_event("response.failed", json!({"response":response}))]
}
@@ -680,7 +706,13 @@ fn named_sse(kind: &str, value: Value) -> Bytes {
Bytes::from(format!("event: {kind}\ndata: {value}\n\n"))
}
-fn response_shell(id: &str, model: &str, created: u64, status: &str) -> Value {
+fn response_shell(
+ id: &str,
+ model: &str,
+ created: u64,
+ status: &str,
+ response_metadata: &OpenAiResponseMetadata,
+) -> Value {
json!({
"id":id,
"object":"response",
@@ -689,6 +721,8 @@ fn response_shell(id: &str, model: &str, created: u64, status: &str) -> Value {
"model":model,
"output":[],
"parallel_tool_calls":false,
+ "tool_choice":response_metadata.tool_choice,
+ "tools":response_metadata.tools,
"error":null,
"incomplete_details":null,
"usage":null,
@@ -788,12 +822,15 @@ mod tests {
"event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n",
);
let response = streaming_response(
- OpenAiSurface::ChatCompletions,
GenerationBody::BufferedSse(Bytes::from_static(input.as_bytes())),
- true,
- "chatcmpl_test".into(),
- "kimi-k2.6".into(),
- 1,
+ Renderer::new(
+ OpenAiSurface::ChatCompletions,
+ true,
+ "chatcmpl_test".into(),
+ "kimi-k2.6".into(),
+ 1,
+ OpenAiResponseMetadata::default(),
+ ),
None,
);
let bytes = response.into_body().collect().await.unwrap().to_bytes();
@@ -815,12 +852,15 @@ mod tests {
"event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n",
);
let response = streaming_response(
- OpenAiSurface::Responses,
GenerationBody::BufferedSse(Bytes::from_static(input.as_bytes())),
- false,
- "resp_test".into(),
- "grok-4.5".into(),
- 1,
+ Renderer::new(
+ OpenAiSurface::Responses,
+ false,
+ "resp_test".into(),
+ "grok-4.5".into(),
+ 1,
+ OpenAiResponseMetadata::default(),
+ ),
None,
);
let bytes = response.into_body().collect().await.unwrap().to_bytes();
diff --git a/src/server.rs b/src/server.rs
index a4364dd..545ca0a 100644
--- a/src/server.rs
+++ b/src/server.rs
@@ -778,6 +778,7 @@ async fn handler_responses(State(state): State<Arc<AppState>>, req: Request<Body
generation,
parsed.stream,
parsed.include_usage,
+ parsed.response_metadata,
traffic.clone(),
)
.await
@@ -1029,6 +1030,7 @@ async fn handler_chat_completions(
generation,
parsed.stream,
parsed.include_usage,
+ parsed.response_metadata,
traffic.clone(),
)
.await