1pub mod batches;
2pub mod responses;
3
4use anyhow::{Context as _, Result, anyhow};
5use futures::{AsyncBufReadExt, AsyncReadExt, StreamExt, io::BufReader, stream::BoxStream};
6use http_client::{
7 AsyncBody, HttpClient, Method, Request as HttpRequest, StatusCode,
8 http::{HeaderMap, HeaderValue},
9};
10use serde::{Deserialize, Serialize};
11use serde_json::Value;
12pub use settings::OpenAiReasoningEffort as ReasoningEffort;
13use std::{convert::TryFrom, future::Future};
14use strum::EnumIter;
15use thiserror::Error;
16
17pub const OPEN_AI_API_URL: &str = "https://api.openai.com/v1";
18
19fn is_none_or_empty<T: AsRef<[U]>, U>(opt: &Option<T>) -> bool {
20 opt.as_ref().is_none_or(|v| v.as_ref().is_empty())
21}
22
23#[derive(Clone, Copy, Serialize, Deserialize, Debug, Eq, PartialEq)]
24#[serde(rename_all = "lowercase")]
25pub enum Role {
26 User,
27 Assistant,
28 System,
29 Tool,
30}
31
32impl TryFrom<String> for Role {
33 type Error = anyhow::Error;
34
35 fn try_from(value: String) -> Result<Self> {
36 match value.as_str() {
37 "user" => Ok(Self::User),
38 "assistant" => Ok(Self::Assistant),
39 "system" => Ok(Self::System),
40 "tool" => Ok(Self::Tool),
41 _ => anyhow::bail!("invalid role '{value}'"),
42 }
43 }
44}
45
46impl From<Role> for String {
47 fn from(val: Role) -> Self {
48 match val {
49 Role::User => "user".to_owned(),
50 Role::Assistant => "assistant".to_owned(),
51 Role::System => "system".to_owned(),
52 Role::Tool => "tool".to_owned(),
53 }
54 }
55}
56
57#[cfg_attr(feature = "schemars", derive(schemars::JsonSchema))]
58#[derive(Clone, Debug, Default, Serialize, Deserialize, PartialEq, EnumIter)]
59pub enum Model {
60 #[serde(rename = "gpt-3.5-turbo")]
61 ThreePointFiveTurbo,
62 #[serde(rename = "gpt-4")]
63 Four,
64 #[serde(rename = "gpt-4-turbo")]
65 FourTurbo,
66 #[serde(rename = "gpt-4o-mini")]
67 FourOmniMini,
68 #[serde(rename = "gpt-4.1-nano")]
69 FourPointOneNano,
70 #[serde(rename = "o1")]
71 O1,
72 #[serde(rename = "o3-mini")]
73 O3Mini,
74 #[serde(rename = "o3")]
75 O3,
76 #[serde(rename = "gpt-5")]
77 Five,
78 #[serde(rename = "gpt-5-codex")]
79 FiveCodex,
80 #[serde(rename = "gpt-5-mini")]
81 #[default]
82 FiveMini,
83 #[serde(rename = "gpt-5-nano")]
84 FiveNano,
85 #[serde(rename = "gpt-5.1")]
86 FivePointOne,
87 #[serde(rename = "gpt-5.2")]
88 FivePointTwo,
89 #[serde(rename = "gpt-5.2-codex")]
90 FivePointTwoCodex,
91 #[serde(rename = "gpt-5.3-codex")]
92 FivePointThreeCodex,
93 #[serde(rename = "gpt-5.4")]
94 FivePointFour,
95 #[serde(rename = "gpt-5.4-pro")]
96 FivePointFourPro,
97 #[serde(rename = "custom")]
98 Custom {
99 name: String,
100 /// The name displayed in the UI, such as in the assistant panel model dropdown menu.
101 display_name: Option<String>,
102 max_tokens: u64,
103 max_output_tokens: Option<u64>,
104 max_completion_tokens: Option<u64>,
105 reasoning_effort: Option<ReasoningEffort>,
106 #[serde(default = "default_supports_chat_completions")]
107 supports_chat_completions: bool,
108 },
109}
110
111const fn default_supports_chat_completions() -> bool {
112 true
113}
114
115impl Model {
116 pub fn default_fast() -> Self {
117 Self::FiveMini
118 }
119
120 pub fn from_id(id: &str) -> Result<Self> {
121 match id {
122 "gpt-3.5-turbo" => Ok(Self::ThreePointFiveTurbo),
123 "gpt-4" => Ok(Self::Four),
124 "gpt-4-turbo-preview" => Ok(Self::FourTurbo),
125 "gpt-4o-mini" => Ok(Self::FourOmniMini),
126 "gpt-4.1-nano" => Ok(Self::FourPointOneNano),
127 "o1" => Ok(Self::O1),
128 "o3-mini" => Ok(Self::O3Mini),
129 "o3" => Ok(Self::O3),
130 "gpt-5" => Ok(Self::Five),
131 "gpt-5-codex" => Ok(Self::FiveCodex),
132 "gpt-5-mini" => Ok(Self::FiveMini),
133 "gpt-5-nano" => Ok(Self::FiveNano),
134 "gpt-5.1" => Ok(Self::FivePointOne),
135 "gpt-5.2" => Ok(Self::FivePointTwo),
136 "gpt-5.2-codex" => Ok(Self::FivePointTwoCodex),
137 "gpt-5.3-codex" => Ok(Self::FivePointThreeCodex),
138 "gpt-5.4" => Ok(Self::FivePointFour),
139 "gpt-5.4-pro" => Ok(Self::FivePointFourPro),
140 invalid_id => anyhow::bail!("invalid model id '{invalid_id}'"),
141 }
142 }
143
144 pub fn id(&self) -> &str {
145 match self {
146 Self::ThreePointFiveTurbo => "gpt-3.5-turbo",
147 Self::Four => "gpt-4",
148 Self::FourTurbo => "gpt-4-turbo",
149 Self::FourOmniMini => "gpt-4o-mini",
150 Self::FourPointOneNano => "gpt-4.1-nano",
151 Self::O1 => "o1",
152 Self::O3Mini => "o3-mini",
153 Self::O3 => "o3",
154 Self::Five => "gpt-5",
155 Self::FiveCodex => "gpt-5-codex",
156 Self::FiveMini => "gpt-5-mini",
157 Self::FiveNano => "gpt-5-nano",
158 Self::FivePointOne => "gpt-5.1",
159 Self::FivePointTwo => "gpt-5.2",
160 Self::FivePointTwoCodex => "gpt-5.2-codex",
161 Self::FivePointThreeCodex => "gpt-5.3-codex",
162 Self::FivePointFour => "gpt-5.4",
163 Self::FivePointFourPro => "gpt-5.4-pro",
164 Self::Custom { name, .. } => name,
165 }
166 }
167
168 pub fn display_name(&self) -> &str {
169 match self {
170 Self::ThreePointFiveTurbo => "gpt-3.5-turbo",
171 Self::Four => "gpt-4",
172 Self::FourTurbo => "gpt-4-turbo",
173 Self::FourOmniMini => "gpt-4o-mini",
174 Self::FourPointOneNano => "gpt-4.1-nano",
175 Self::O1 => "o1",
176 Self::O3Mini => "o3-mini",
177 Self::O3 => "o3",
178 Self::Five => "gpt-5",
179 Self::FiveCodex => "gpt-5-codex",
180 Self::FiveMini => "gpt-5-mini",
181 Self::FiveNano => "gpt-5-nano",
182 Self::FivePointOne => "gpt-5.1",
183 Self::FivePointTwo => "gpt-5.2",
184 Self::FivePointTwoCodex => "gpt-5.2-codex",
185 Self::FivePointThreeCodex => "gpt-5.3-codex",
186 Self::FivePointFour => "gpt-5.4",
187 Self::FivePointFourPro => "gpt-5.4-pro",
188 Self::Custom { display_name, .. } => display_name.as_deref().unwrap_or(&self.id()),
189 }
190 }
191
192 pub fn max_token_count(&self) -> u64 {
193 match self {
194 Self::ThreePointFiveTurbo => 16_385,
195 Self::Four => 8_192,
196 Self::FourTurbo => 128_000,
197 Self::FourOmniMini => 128_000,
198 Self::FourPointOneNano => 1_047_576,
199 Self::O1 => 200_000,
200 Self::O3Mini => 200_000,
201 Self::O3 => 200_000,
202 Self::Five => 272_000,
203 Self::FiveCodex => 272_000,
204 Self::FiveMini => 400_000,
205 Self::FiveNano => 400_000,
206 Self::FivePointOne => 400_000,
207 Self::FivePointTwo => 400_000,
208 Self::FivePointTwoCodex => 400_000,
209 Self::FivePointThreeCodex => 400_000,
210 Self::FivePointFour => 1_050_000,
211 Self::FivePointFourPro => 1_050_000,
212 Self::Custom { max_tokens, .. } => *max_tokens,
213 }
214 }
215
216 pub fn max_output_tokens(&self) -> Option<u64> {
217 match self {
218 Self::Custom {
219 max_output_tokens, ..
220 } => *max_output_tokens,
221 Self::ThreePointFiveTurbo => Some(4_096),
222 Self::Four => Some(8_192),
223 Self::FourTurbo => Some(4_096),
224 Self::FourOmniMini => Some(16_384),
225 Self::FourPointOneNano => Some(32_768),
226 Self::O1 => Some(100_000),
227 Self::O3Mini => Some(100_000),
228 Self::O3 => Some(100_000),
229 Self::Five => Some(128_000),
230 Self::FiveCodex => Some(128_000),
231 Self::FiveMini => Some(128_000),
232 Self::FiveNano => Some(128_000),
233 Self::FivePointOne => Some(128_000),
234 Self::FivePointTwo => Some(128_000),
235 Self::FivePointTwoCodex => Some(128_000),
236 Self::FivePointThreeCodex => Some(128_000),
237 Self::FivePointFour => Some(128_000),
238 Self::FivePointFourPro => Some(128_000),
239 }
240 }
241
242 pub fn reasoning_effort(&self) -> Option<ReasoningEffort> {
243 match self {
244 Self::Custom {
245 reasoning_effort, ..
246 } => reasoning_effort.to_owned(),
247 Self::FivePointThreeCodex | Self::FivePointFourPro => Some(ReasoningEffort::Medium),
248 _ => None,
249 }
250 }
251
252 pub fn supports_chat_completions(&self) -> bool {
253 match self {
254 Self::Custom {
255 supports_chat_completions,
256 ..
257 } => *supports_chat_completions,
258 Self::FiveCodex
259 | Self::FivePointTwoCodex
260 | Self::FivePointThreeCodex
261 | Self::FivePointFourPro => false,
262 _ => true,
263 }
264 }
265
266 /// Returns whether the given model supports the `parallel_tool_calls` parameter.
267 ///
268 /// If the model does not support the parameter, do not pass it up, or the API will return an error.
269 pub fn supports_parallel_tool_calls(&self) -> bool {
270 match self {
271 Self::ThreePointFiveTurbo
272 | Self::Four
273 | Self::FourTurbo
274 | Self::FourOmniMini
275 | Self::FourPointOneNano
276 | Self::Five
277 | Self::FiveCodex
278 | Self::FiveMini
279 | Self::FivePointOne
280 | Self::FivePointTwo
281 | Self::FivePointTwoCodex
282 | Self::FivePointThreeCodex
283 | Self::FivePointFour
284 | Self::FivePointFourPro
285 | Self::FiveNano => true,
286 Self::O1 | Self::O3 | Self::O3Mini | Model::Custom { .. } => false,
287 }
288 }
289
290 /// Returns whether the given model supports the `prompt_cache_key` parameter.
291 ///
292 /// If the model does not support the parameter, do not pass it up.
293 pub fn supports_prompt_cache_key(&self) -> bool {
294 true
295 }
296}
297
298#[derive(Debug, Serialize, Deserialize)]
299pub struct Request {
300 pub model: String,
301 pub messages: Vec<RequestMessage>,
302 pub stream: bool,
303 #[serde(default, skip_serializing_if = "Option::is_none")]
304 pub max_completion_tokens: Option<u64>,
305 #[serde(default, skip_serializing_if = "Vec::is_empty")]
306 pub stop: Vec<String>,
307 #[serde(default, skip_serializing_if = "Option::is_none")]
308 pub temperature: Option<f32>,
309 #[serde(default, skip_serializing_if = "Option::is_none")]
310 pub tool_choice: Option<ToolChoice>,
311 /// Whether to enable parallel function calling during tool use.
312 #[serde(default, skip_serializing_if = "Option::is_none")]
313 pub parallel_tool_calls: Option<bool>,
314 #[serde(default, skip_serializing_if = "Vec::is_empty")]
315 pub tools: Vec<ToolDefinition>,
316 #[serde(default, skip_serializing_if = "Option::is_none")]
317 pub prompt_cache_key: Option<String>,
318 #[serde(default, skip_serializing_if = "Option::is_none")]
319 pub reasoning_effort: Option<ReasoningEffort>,
320}
321
322#[derive(Debug, Serialize, Deserialize)]
323#[serde(rename_all = "lowercase")]
324pub enum ToolChoice {
325 Auto,
326 Required,
327 None,
328 #[serde(untagged)]
329 Other(ToolDefinition),
330}
331
332#[derive(Clone, Deserialize, Serialize, Debug)]
333#[serde(tag = "type", rename_all = "snake_case")]
334pub enum ToolDefinition {
335 #[allow(dead_code)]
336 Function { function: FunctionDefinition },
337}
338
339#[derive(Clone, Debug, Serialize, Deserialize)]
340pub struct FunctionDefinition {
341 pub name: String,
342 pub description: Option<String>,
343 pub parameters: Option<Value>,
344}
345
346#[derive(Clone, Serialize, Deserialize, Debug, Eq, PartialEq)]
347#[serde(tag = "role", rename_all = "lowercase")]
348pub enum RequestMessage {
349 Assistant {
350 content: Option<MessageContent>,
351 #[serde(default, skip_serializing_if = "Vec::is_empty")]
352 tool_calls: Vec<ToolCall>,
353 },
354 User {
355 content: MessageContent,
356 },
357 System {
358 content: MessageContent,
359 },
360 Tool {
361 content: MessageContent,
362 tool_call_id: String,
363 },
364}
365
366#[derive(Serialize, Deserialize, Clone, Debug, Eq, PartialEq)]
367#[serde(untagged)]
368pub enum MessageContent {
369 Plain(String),
370 Multipart(Vec<MessagePart>),
371}
372
373impl MessageContent {
374 pub fn empty() -> Self {
375 MessageContent::Multipart(vec![])
376 }
377
378 pub fn push_part(&mut self, part: MessagePart) {
379 match self {
380 MessageContent::Plain(text) => {
381 *self =
382 MessageContent::Multipart(vec![MessagePart::Text { text: text.clone() }, part]);
383 }
384 MessageContent::Multipart(parts) if parts.is_empty() => match part {
385 MessagePart::Text { text } => *self = MessageContent::Plain(text),
386 MessagePart::Image { .. } => *self = MessageContent::Multipart(vec![part]),
387 },
388 MessageContent::Multipart(parts) => parts.push(part),
389 }
390 }
391}
392
393impl From<Vec<MessagePart>> for MessageContent {
394 fn from(mut parts: Vec<MessagePart>) -> Self {
395 if let [MessagePart::Text { text }] = parts.as_mut_slice() {
396 MessageContent::Plain(std::mem::take(text))
397 } else {
398 MessageContent::Multipart(parts)
399 }
400 }
401}
402
403#[derive(Serialize, Deserialize, Clone, Debug, Eq, PartialEq)]
404#[serde(tag = "type")]
405pub enum MessagePart {
406 #[serde(rename = "text")]
407 Text { text: String },
408 #[serde(rename = "image_url")]
409 Image { image_url: ImageUrl },
410}
411
412#[derive(Serialize, Deserialize, Clone, Debug, Eq, PartialEq)]
413pub struct ImageUrl {
414 pub url: String,
415 #[serde(skip_serializing_if = "Option::is_none")]
416 pub detail: Option<String>,
417}
418
419#[derive(Clone, Serialize, Deserialize, Debug, Eq, PartialEq)]
420pub struct ToolCall {
421 pub id: String,
422 #[serde(flatten)]
423 pub content: ToolCallContent,
424}
425
426#[derive(Clone, Serialize, Deserialize, Debug, Eq, PartialEq)]
427#[serde(tag = "type", rename_all = "lowercase")]
428pub enum ToolCallContent {
429 Function { function: FunctionContent },
430}
431
432#[derive(Clone, Serialize, Deserialize, Debug, Eq, PartialEq)]
433pub struct FunctionContent {
434 pub name: String,
435 pub arguments: String,
436}
437
438#[derive(Clone, Serialize, Deserialize, Debug)]
439pub struct Response {
440 pub id: String,
441 pub object: String,
442 pub created: u64,
443 pub model: String,
444 pub choices: Vec<Choice>,
445 pub usage: Usage,
446}
447
448#[derive(Clone, Serialize, Deserialize, Debug)]
449pub struct Choice {
450 pub index: u32,
451 pub message: RequestMessage,
452 pub finish_reason: Option<String>,
453}
454
455#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
456pub struct ResponseMessageDelta {
457 pub role: Option<Role>,
458 pub content: Option<String>,
459 #[serde(default, skip_serializing_if = "is_none_or_empty")]
460 pub tool_calls: Option<Vec<ToolCallChunk>>,
461 #[serde(default, skip_serializing_if = "is_none_or_empty")]
462 pub reasoning_content: Option<String>,
463}
464
465#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
466pub struct ToolCallChunk {
467 pub index: usize,
468 pub id: Option<String>,
469
470 // There is also an optional `type` field that would determine if a
471 // function is there. Sometimes this streams in with the `function` before
472 // it streams in the `type`
473 pub function: Option<FunctionChunk>,
474}
475
476#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
477pub struct FunctionChunk {
478 pub name: Option<String>,
479 pub arguments: Option<String>,
480}
481
482#[derive(Clone, Serialize, Deserialize, Debug)]
483pub struct Usage {
484 pub prompt_tokens: u64,
485 pub completion_tokens: u64,
486 pub total_tokens: u64,
487}
488
489#[derive(Serialize, Deserialize, Debug)]
490pub struct ChoiceDelta {
491 pub index: u32,
492 pub delta: Option<ResponseMessageDelta>,
493 pub finish_reason: Option<String>,
494}
495
496#[derive(Error, Debug)]
497pub enum RequestError {
498 #[error("HTTP response error from {provider}'s API: status {status_code} - {body:?}")]
499 HttpResponseError {
500 provider: String,
501 status_code: StatusCode,
502 body: String,
503 headers: HeaderMap<HeaderValue>,
504 },
505 #[error(transparent)]
506 Other(#[from] anyhow::Error),
507}
508
509#[derive(Serialize, Deserialize, Debug)]
510pub struct ResponseStreamError {
511 message: String,
512}
513
514#[derive(Serialize, Deserialize, Debug)]
515#[serde(untagged)]
516pub enum ResponseStreamResult {
517 Ok(ResponseStreamEvent),
518 Err { error: ResponseStreamError },
519}
520
521#[derive(Serialize, Deserialize, Debug)]
522pub struct ResponseStreamEvent {
523 pub choices: Vec<ChoiceDelta>,
524 pub usage: Option<Usage>,
525}
526
527pub async fn non_streaming_completion(
528 client: &dyn HttpClient,
529 api_url: &str,
530 api_key: &str,
531 request: Request,
532) -> Result<Response, RequestError> {
533 let uri = format!("{api_url}/chat/completions");
534 let request_builder = HttpRequest::builder()
535 .method(Method::POST)
536 .uri(uri)
537 .header("Content-Type", "application/json")
538 .header("Authorization", format!("Bearer {}", api_key.trim()));
539
540 let request = request_builder
541 .body(AsyncBody::from(
542 serde_json::to_string(&request).map_err(|e| RequestError::Other(e.into()))?,
543 ))
544 .map_err(|e| RequestError::Other(e.into()))?;
545
546 let mut response = client.send(request).await?;
547 if response.status().is_success() {
548 let mut body = String::new();
549 response
550 .body_mut()
551 .read_to_string(&mut body)
552 .await
553 .map_err(|e| RequestError::Other(e.into()))?;
554
555 serde_json::from_str(&body).map_err(|e| RequestError::Other(e.into()))
556 } else {
557 let mut body = String::new();
558 response
559 .body_mut()
560 .read_to_string(&mut body)
561 .await
562 .map_err(|e| RequestError::Other(e.into()))?;
563
564 Err(RequestError::HttpResponseError {
565 provider: "openai".to_owned(),
566 status_code: response.status(),
567 body,
568 headers: response.headers().clone(),
569 })
570 }
571}
572
573pub async fn stream_completion(
574 client: &dyn HttpClient,
575 provider_name: &str,
576 api_url: &str,
577 api_key: &str,
578 request: Request,
579) -> Result<BoxStream<'static, Result<ResponseStreamEvent>>, RequestError> {
580 let uri = format!("{api_url}/chat/completions");
581 let request_builder = HttpRequest::builder()
582 .method(Method::POST)
583 .uri(uri)
584 .header("Content-Type", "application/json")
585 .header("Authorization", format!("Bearer {}", api_key.trim()));
586
587 let request = request_builder
588 .body(AsyncBody::from(
589 serde_json::to_string(&request).map_err(|e| RequestError::Other(e.into()))?,
590 ))
591 .map_err(|e| RequestError::Other(e.into()))?;
592
593 let mut response = client.send(request).await?;
594 if response.status().is_success() {
595 let reader = BufReader::new(response.into_body());
596 Ok(reader
597 .lines()
598 .filter_map(|line| async move {
599 match line {
600 Ok(line) => {
601 let line = line.strip_prefix("data: ").or_else(|| line.strip_prefix("data:"))?;
602 if line == "[DONE]" {
603 None
604 } else {
605 match serde_json::from_str(line) {
606 Ok(ResponseStreamResult::Ok(response)) => Some(Ok(response)),
607 Ok(ResponseStreamResult::Err { error }) => {
608 Some(Err(anyhow!(error.message)))
609 }
610 Err(error) => {
611 log::error!(
612 "Failed to parse OpenAI response into ResponseStreamResult: `{}`\n\
613 Response: `{}`",
614 error,
615 line,
616 );
617 Some(Err(anyhow!(error)))
618 }
619 }
620 }
621 }
622 Err(error) => Some(Err(anyhow!(error))),
623 }
624 })
625 .boxed())
626 } else {
627 let mut body = String::new();
628 response
629 .body_mut()
630 .read_to_string(&mut body)
631 .await
632 .map_err(|e| RequestError::Other(e.into()))?;
633
634 Err(RequestError::HttpResponseError {
635 provider: provider_name.to_owned(),
636 status_code: response.status(),
637 body,
638 headers: response.headers().clone(),
639 })
640 }
641}
642
643#[derive(Copy, Clone, Serialize, Deserialize)]
644pub enum OpenAiEmbeddingModel {
645 #[serde(rename = "text-embedding-3-small")]
646 TextEmbedding3Small,
647 #[serde(rename = "text-embedding-3-large")]
648 TextEmbedding3Large,
649}
650
651#[derive(Serialize)]
652struct OpenAiEmbeddingRequest<'a> {
653 model: OpenAiEmbeddingModel,
654 input: Vec<&'a str>,
655}
656
657#[derive(Deserialize)]
658pub struct OpenAiEmbeddingResponse {
659 pub data: Vec<OpenAiEmbedding>,
660}
661
662#[derive(Deserialize)]
663pub struct OpenAiEmbedding {
664 pub embedding: Vec<f32>,
665}
666
667pub fn embed<'a>(
668 client: &dyn HttpClient,
669 api_url: &str,
670 api_key: &str,
671 model: OpenAiEmbeddingModel,
672 texts: impl IntoIterator<Item = &'a str>,
673) -> impl 'static + Future<Output = Result<OpenAiEmbeddingResponse>> {
674 let uri = format!("{api_url}/embeddings");
675
676 let request = OpenAiEmbeddingRequest {
677 model,
678 input: texts.into_iter().collect(),
679 };
680 let body = AsyncBody::from(serde_json::to_string(&request).unwrap());
681 let request = HttpRequest::builder()
682 .method(Method::POST)
683 .uri(uri)
684 .header("Content-Type", "application/json")
685 .header("Authorization", format!("Bearer {}", api_key.trim()))
686 .body(body)
687 .map(|request| client.send(request));
688
689 async move {
690 let mut response = request?.await?;
691 let mut body = String::new();
692 response.body_mut().read_to_string(&mut body).await?;
693
694 anyhow::ensure!(
695 response.status().is_success(),
696 "error during embedding, status: {:?}, body: {:?}",
697 response.status(),
698 body
699 );
700 let response: OpenAiEmbeddingResponse =
701 serde_json::from_str(&body).context("failed to parse OpenAI embedding response")?;
702 Ok(response)
703 }
704}