1use anyhow::{Context as _, Result, anyhow};
2use futures::{
3 AsyncBufReadExt, AsyncReadExt, StreamExt,
4 io::BufReader,
5 stream::{self, BoxStream},
6};
7use http_client::{AsyncBody, HttpClient, Method, Request as HttpRequest};
8use serde::{Deserialize, Serialize};
9use serde_json::Value;
10use std::{
11 convert::TryFrom,
12 future::{self, Future},
13};
14use strum::EnumIter;
15
16pub const OPEN_AI_API_URL: &str = "https://api.openai.com/v1";
17
18fn is_none_or_empty<T: AsRef<[U]>, U>(opt: &Option<T>) -> bool {
19 opt.as_ref().map_or(true, |v| v.as_ref().is_empty())
20}
21
22#[derive(Clone, Copy, Serialize, Deserialize, Debug, Eq, PartialEq)]
23#[serde(rename_all = "lowercase")]
24pub enum Role {
25 User,
26 Assistant,
27 System,
28 Tool,
29}
30
31impl TryFrom<String> for Role {
32 type Error = anyhow::Error;
33
34 fn try_from(value: String) -> Result<Self> {
35 match value.as_str() {
36 "user" => Ok(Self::User),
37 "assistant" => Ok(Self::Assistant),
38 "system" => Ok(Self::System),
39 "tool" => Ok(Self::Tool),
40 _ => anyhow::bail!("invalid role '{value}'"),
41 }
42 }
43}
44
45impl From<Role> for String {
46 fn from(val: Role) -> Self {
47 match val {
48 Role::User => "user".to_owned(),
49 Role::Assistant => "assistant".to_owned(),
50 Role::System => "system".to_owned(),
51 Role::Tool => "tool".to_owned(),
52 }
53 }
54}
55
56#[cfg_attr(feature = "schemars", derive(schemars::JsonSchema))]
57#[derive(Clone, Debug, Default, Serialize, Deserialize, PartialEq, EnumIter)]
58pub enum Model {
59 #[serde(rename = "gpt-3.5-turbo")]
60 ThreePointFiveTurbo,
61 #[serde(rename = "gpt-4")]
62 Four,
63 #[serde(rename = "gpt-4-turbo")]
64 FourTurbo,
65 #[serde(rename = "gpt-4o")]
66 #[default]
67 FourOmni,
68 #[serde(rename = "gpt-4o-mini")]
69 FourOmniMini,
70 #[serde(rename = "gpt-4.1")]
71 FourPointOne,
72 #[serde(rename = "gpt-4.1-mini")]
73 FourPointOneMini,
74 #[serde(rename = "gpt-4.1-nano")]
75 FourPointOneNano,
76 #[serde(rename = "o1")]
77 O1,
78 #[serde(rename = "o1-preview")]
79 O1Preview,
80 #[serde(rename = "o1-mini")]
81 O1Mini,
82 #[serde(rename = "o3-mini")]
83 O3Mini,
84 #[serde(rename = "o3")]
85 O3,
86 #[serde(rename = "o4-mini")]
87 O4Mini,
88
89 #[serde(rename = "custom")]
90 Custom {
91 name: String,
92 /// The name displayed in the UI, such as in the assistant panel model dropdown menu.
93 display_name: Option<String>,
94 max_tokens: usize,
95 max_output_tokens: Option<u32>,
96 max_completion_tokens: Option<u32>,
97 },
98}
99
100impl Model {
101 pub fn default_fast() -> Self {
102 Self::FourPointOneMini
103 }
104
105 pub fn from_id(id: &str) -> Result<Self> {
106 match id {
107 "gpt-3.5-turbo" => Ok(Self::ThreePointFiveTurbo),
108 "gpt-4" => Ok(Self::Four),
109 "gpt-4-turbo-preview" => Ok(Self::FourTurbo),
110 "gpt-4o" => Ok(Self::FourOmni),
111 "gpt-4o-mini" => Ok(Self::FourOmniMini),
112 "gpt-4.1" => Ok(Self::FourPointOne),
113 "gpt-4.1-mini" => Ok(Self::FourPointOneMini),
114 "gpt-4.1-nano" => Ok(Self::FourPointOneNano),
115 "o1" => Ok(Self::O1),
116 "o1-preview" => Ok(Self::O1Preview),
117 "o1-mini" => Ok(Self::O1Mini),
118 "o3-mini" => Ok(Self::O3Mini),
119 "o3" => Ok(Self::O3),
120 "o4-mini" => Ok(Self::O4Mini),
121 invalid_id => anyhow::bail!("invalid model id '{invalid_id}'"),
122 }
123 }
124
125 pub fn id(&self) -> &str {
126 match self {
127 Self::ThreePointFiveTurbo => "gpt-3.5-turbo",
128 Self::Four => "gpt-4",
129 Self::FourTurbo => "gpt-4-turbo",
130 Self::FourOmni => "gpt-4o",
131 Self::FourOmniMini => "gpt-4o-mini",
132 Self::FourPointOne => "gpt-4.1",
133 Self::FourPointOneMini => "gpt-4.1-mini",
134 Self::FourPointOneNano => "gpt-4.1-nano",
135 Self::O1 => "o1",
136 Self::O1Preview => "o1-preview",
137 Self::O1Mini => "o1-mini",
138 Self::O3Mini => "o3-mini",
139 Self::O3 => "o3",
140 Self::O4Mini => "o4-mini",
141 Self::Custom { name, .. } => name,
142 }
143 }
144
145 pub fn display_name(&self) -> &str {
146 match self {
147 Self::ThreePointFiveTurbo => "gpt-3.5-turbo",
148 Self::Four => "gpt-4",
149 Self::FourTurbo => "gpt-4-turbo",
150 Self::FourOmni => "gpt-4o",
151 Self::FourOmniMini => "gpt-4o-mini",
152 Self::FourPointOne => "gpt-4.1",
153 Self::FourPointOneMini => "gpt-4.1-mini",
154 Self::FourPointOneNano => "gpt-4.1-nano",
155 Self::O1 => "o1",
156 Self::O1Preview => "o1-preview",
157 Self::O1Mini => "o1-mini",
158 Self::O3Mini => "o3-mini",
159 Self::O3 => "o3",
160 Self::O4Mini => "o4-mini",
161 Self::Custom {
162 name, display_name, ..
163 } => display_name.as_ref().unwrap_or(name),
164 }
165 }
166
167 pub fn max_token_count(&self) -> usize {
168 match self {
169 Self::ThreePointFiveTurbo => 16_385,
170 Self::Four => 8_192,
171 Self::FourTurbo => 128_000,
172 Self::FourOmni => 128_000,
173 Self::FourOmniMini => 128_000,
174 Self::FourPointOne => 1_047_576,
175 Self::FourPointOneMini => 1_047_576,
176 Self::FourPointOneNano => 1_047_576,
177 Self::O1 => 200_000,
178 Self::O1Preview => 128_000,
179 Self::O1Mini => 128_000,
180 Self::O3Mini => 200_000,
181 Self::O3 => 200_000,
182 Self::O4Mini => 200_000,
183 Self::Custom { max_tokens, .. } => *max_tokens,
184 }
185 }
186
187 pub fn max_output_tokens(&self) -> Option<u32> {
188 match self {
189 Self::Custom {
190 max_output_tokens, ..
191 } => *max_output_tokens,
192 Self::ThreePointFiveTurbo => Some(4_096),
193 Self::Four => Some(8_192),
194 Self::FourTurbo => Some(4_096),
195 Self::FourOmni => Some(16_384),
196 Self::FourOmniMini => Some(16_384),
197 Self::FourPointOne => Some(32_768),
198 Self::FourPointOneMini => Some(32_768),
199 Self::FourPointOneNano => Some(32_768),
200 Self::O1 => Some(100_000),
201 Self::O1Preview => Some(32_768),
202 Self::O1Mini => Some(65_536),
203 Self::O3Mini => Some(100_000),
204 Self::O3 => Some(100_000),
205 Self::O4Mini => Some(100_000),
206 }
207 }
208
209 /// Returns whether the given model supports the `parallel_tool_calls` parameter.
210 ///
211 /// If the model does not support the parameter, do not pass it up, or the API will return an error.
212 pub fn supports_parallel_tool_calls(&self) -> bool {
213 match self {
214 Self::ThreePointFiveTurbo
215 | Self::Four
216 | Self::FourTurbo
217 | Self::FourOmni
218 | Self::FourOmniMini
219 | Self::FourPointOne
220 | Self::FourPointOneMini
221 | Self::FourPointOneNano => true,
222 Self::O1
223 | Self::O1Preview
224 | Self::O1Mini
225 | Self::O3
226 | Self::O3Mini
227 | Self::O4Mini
228 | Model::Custom { .. } => false,
229 }
230 }
231}
232
233#[derive(Debug, Serialize, Deserialize)]
234pub struct Request {
235 pub model: String,
236 pub messages: Vec<RequestMessage>,
237 pub stream: bool,
238 #[serde(default, skip_serializing_if = "Option::is_none")]
239 pub max_tokens: Option<u32>,
240 #[serde(default, skip_serializing_if = "Vec::is_empty")]
241 pub stop: Vec<String>,
242 pub temperature: f32,
243 #[serde(default, skip_serializing_if = "Option::is_none")]
244 pub tool_choice: Option<ToolChoice>,
245 /// Whether to enable parallel function calling during tool use.
246 #[serde(default, skip_serializing_if = "Option::is_none")]
247 pub parallel_tool_calls: Option<bool>,
248 #[serde(default, skip_serializing_if = "Vec::is_empty")]
249 pub tools: Vec<ToolDefinition>,
250}
251
252#[derive(Debug, Serialize, Deserialize)]
253pub struct CompletionRequest {
254 pub model: String,
255 pub prompt: String,
256 pub max_tokens: u32,
257 pub temperature: f32,
258 #[serde(default, skip_serializing_if = "Option::is_none")]
259 pub prediction: Option<Prediction>,
260 #[serde(default, skip_serializing_if = "Option::is_none")]
261 pub rewrite_speculation: Option<bool>,
262}
263
264#[derive(Clone, Deserialize, Serialize, Debug)]
265#[serde(tag = "type", rename_all = "snake_case")]
266pub enum Prediction {
267 Content { content: String },
268}
269
270#[derive(Debug, Serialize, Deserialize)]
271#[serde(untagged)]
272pub enum ToolChoice {
273 Auto,
274 Required,
275 None,
276 Other(ToolDefinition),
277}
278
279#[derive(Clone, Deserialize, Serialize, Debug)]
280#[serde(tag = "type", rename_all = "snake_case")]
281pub enum ToolDefinition {
282 #[allow(dead_code)]
283 Function { function: FunctionDefinition },
284}
285
286#[derive(Clone, Debug, Serialize, Deserialize)]
287pub struct FunctionDefinition {
288 pub name: String,
289 pub description: Option<String>,
290 pub parameters: Option<Value>,
291}
292
293#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
294#[serde(tag = "role", rename_all = "lowercase")]
295pub enum RequestMessage {
296 Assistant {
297 content: Option<MessageContent>,
298 #[serde(default, skip_serializing_if = "Vec::is_empty")]
299 tool_calls: Vec<ToolCall>,
300 },
301 User {
302 content: MessageContent,
303 },
304 System {
305 content: MessageContent,
306 },
307 Tool {
308 content: MessageContent,
309 tool_call_id: String,
310 },
311}
312
313#[derive(Serialize, Deserialize, Clone, Debug, Eq, PartialEq)]
314#[serde(untagged)]
315pub enum MessageContent {
316 Plain(String),
317 Multipart(Vec<MessagePart>),
318}
319
320impl MessageContent {
321 pub fn empty() -> Self {
322 MessageContent::Multipart(vec![])
323 }
324
325 pub fn push_part(&mut self, part: MessagePart) {
326 match self {
327 MessageContent::Plain(text) => {
328 *self =
329 MessageContent::Multipart(vec![MessagePart::Text { text: text.clone() }, part]);
330 }
331 MessageContent::Multipart(parts) if parts.is_empty() => match part {
332 MessagePart::Text { text } => *self = MessageContent::Plain(text),
333 MessagePart::Image { .. } => *self = MessageContent::Multipart(vec![part]),
334 },
335 MessageContent::Multipart(parts) => parts.push(part),
336 }
337 }
338}
339
340impl From<Vec<MessagePart>> for MessageContent {
341 fn from(mut parts: Vec<MessagePart>) -> Self {
342 if let [MessagePart::Text { text }] = parts.as_mut_slice() {
343 MessageContent::Plain(std::mem::take(text))
344 } else {
345 MessageContent::Multipart(parts)
346 }
347 }
348}
349
350#[derive(Serialize, Deserialize, Clone, Debug, Eq, PartialEq)]
351#[serde(tag = "type")]
352pub enum MessagePart {
353 #[serde(rename = "text")]
354 Text { text: String },
355 #[serde(rename = "image_url")]
356 Image { image_url: ImageUrl },
357}
358
359#[derive(Serialize, Deserialize, Clone, Debug, Eq, PartialEq)]
360pub struct ImageUrl {
361 pub url: String,
362 #[serde(skip_serializing_if = "Option::is_none")]
363 pub detail: Option<String>,
364}
365
366#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
367pub struct ToolCall {
368 pub id: String,
369 #[serde(flatten)]
370 pub content: ToolCallContent,
371}
372
373#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
374#[serde(tag = "type", rename_all = "lowercase")]
375pub enum ToolCallContent {
376 Function { function: FunctionContent },
377}
378
379#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
380pub struct FunctionContent {
381 pub name: String,
382 pub arguments: String,
383}
384
385#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
386pub struct ResponseMessageDelta {
387 pub role: Option<Role>,
388 pub content: Option<String>,
389 #[serde(default, skip_serializing_if = "is_none_or_empty")]
390 pub tool_calls: Option<Vec<ToolCallChunk>>,
391}
392
393#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
394pub struct ToolCallChunk {
395 pub index: usize,
396 pub id: Option<String>,
397
398 // There is also an optional `type` field that would determine if a
399 // function is there. Sometimes this streams in with the `function` before
400 // it streams in the `type`
401 pub function: Option<FunctionChunk>,
402}
403
404#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
405pub struct FunctionChunk {
406 pub name: Option<String>,
407 pub arguments: Option<String>,
408}
409
410#[derive(Serialize, Deserialize, Debug)]
411pub struct Usage {
412 pub prompt_tokens: u32,
413 pub completion_tokens: u32,
414 pub total_tokens: u32,
415}
416
417#[derive(Serialize, Deserialize, Debug)]
418pub struct ChoiceDelta {
419 pub index: u32,
420 pub delta: ResponseMessageDelta,
421 pub finish_reason: Option<String>,
422}
423
424#[derive(Serialize, Deserialize, Debug)]
425#[serde(untagged)]
426pub enum ResponseStreamResult {
427 Ok(ResponseStreamEvent),
428 Err { error: String },
429}
430
431#[derive(Serialize, Deserialize, Debug)]
432pub struct ResponseStreamEvent {
433 pub model: String,
434 pub choices: Vec<ChoiceDelta>,
435 pub usage: Option<Usage>,
436}
437
438#[derive(Serialize, Deserialize, Debug)]
439pub struct CompletionResponse {
440 pub id: String,
441 pub object: String,
442 pub created: u64,
443 pub model: String,
444 pub choices: Vec<CompletionChoice>,
445 pub usage: Usage,
446}
447
448#[derive(Serialize, Deserialize, Debug)]
449pub struct CompletionChoice {
450 pub text: String,
451}
452
453#[derive(Serialize, Deserialize, Debug)]
454pub struct Response {
455 pub id: String,
456 pub object: String,
457 pub created: u64,
458 pub model: String,
459 pub choices: Vec<Choice>,
460 pub usage: Usage,
461}
462
463#[derive(Serialize, Deserialize, Debug)]
464pub struct Choice {
465 pub index: u32,
466 pub message: RequestMessage,
467 pub finish_reason: Option<String>,
468}
469
470pub async fn complete(
471 client: &dyn HttpClient,
472 api_url: &str,
473 api_key: &str,
474 request: Request,
475) -> Result<Response> {
476 let uri = format!("{api_url}/chat/completions");
477 let request_builder = HttpRequest::builder()
478 .method(Method::POST)
479 .uri(uri)
480 .header("Content-Type", "application/json")
481 .header("Authorization", format!("Bearer {}", api_key));
482
483 let mut request_body = request;
484 request_body.stream = false;
485
486 let request = request_builder.body(AsyncBody::from(serde_json::to_string(&request_body)?))?;
487 let mut response = client.send(request).await?;
488
489 if response.status().is_success() {
490 let mut body = String::new();
491 response.body_mut().read_to_string(&mut body).await?;
492 let response: Response = serde_json::from_str(&body)?;
493 Ok(response)
494 } else {
495 let mut body = String::new();
496 response.body_mut().read_to_string(&mut body).await?;
497
498 #[derive(Deserialize)]
499 struct OpenAiResponse {
500 error: OpenAiError,
501 }
502
503 #[derive(Deserialize)]
504 struct OpenAiError {
505 message: String,
506 }
507
508 match serde_json::from_str::<OpenAiResponse>(&body) {
509 Ok(response) if !response.error.message.is_empty() => anyhow::bail!(
510 "Failed to connect to OpenAI API: {}",
511 response.error.message,
512 ),
513 _ => anyhow::bail!(
514 "Failed to connect to OpenAI API: {} {}",
515 response.status(),
516 body,
517 ),
518 }
519 }
520}
521
522pub async fn complete_text(
523 client: &dyn HttpClient,
524 api_url: &str,
525 api_key: &str,
526 request: CompletionRequest,
527) -> Result<CompletionResponse> {
528 let uri = format!("{api_url}/completions");
529 let request_builder = HttpRequest::builder()
530 .method(Method::POST)
531 .uri(uri)
532 .header("Content-Type", "application/json")
533 .header("Authorization", format!("Bearer {}", api_key));
534
535 let request = request_builder.body(AsyncBody::from(serde_json::to_string(&request)?))?;
536 let mut response = client.send(request).await?;
537
538 if response.status().is_success() {
539 let mut body = String::new();
540 response.body_mut().read_to_string(&mut body).await?;
541 let response = serde_json::from_str(&body)?;
542 Ok(response)
543 } else {
544 let mut body = String::new();
545 response.body_mut().read_to_string(&mut body).await?;
546
547 #[derive(Deserialize)]
548 struct OpenAiResponse {
549 error: OpenAiError,
550 }
551
552 #[derive(Deserialize)]
553 struct OpenAiError {
554 message: String,
555 }
556
557 match serde_json::from_str::<OpenAiResponse>(&body) {
558 Ok(response) if !response.error.message.is_empty() => anyhow::bail!(
559 "Failed to connect to OpenAI API: {}",
560 response.error.message,
561 ),
562 _ => anyhow::bail!(
563 "Failed to connect to OpenAI API: {} {}",
564 response.status(),
565 body,
566 ),
567 }
568 }
569}
570
571fn adapt_response_to_stream(response: Response) -> ResponseStreamEvent {
572 ResponseStreamEvent {
573 model: response.model,
574 choices: response
575 .choices
576 .into_iter()
577 .map(|choice| {
578 let content = match &choice.message {
579 RequestMessage::Assistant { content, .. } => content.as_ref(),
580 RequestMessage::User { content } => Some(content),
581 RequestMessage::System { content } => Some(content),
582 RequestMessage::Tool { content, .. } => Some(content),
583 };
584
585 let mut text_content = String::new();
586 match content {
587 Some(MessageContent::Plain(text)) => text_content.push_str(&text),
588 Some(MessageContent::Multipart(parts)) => {
589 for part in parts {
590 match part {
591 MessagePart::Text { text } => text_content.push_str(&text),
592 MessagePart::Image { .. } => {}
593 }
594 }
595 }
596 None => {}
597 };
598
599 ChoiceDelta {
600 index: choice.index,
601 delta: ResponseMessageDelta {
602 role: Some(match choice.message {
603 RequestMessage::Assistant { .. } => Role::Assistant,
604 RequestMessage::User { .. } => Role::User,
605 RequestMessage::System { .. } => Role::System,
606 RequestMessage::Tool { .. } => Role::Tool,
607 }),
608 content: if text_content.is_empty() {
609 None
610 } else {
611 Some(text_content)
612 },
613 tool_calls: None,
614 },
615 finish_reason: choice.finish_reason,
616 }
617 })
618 .collect(),
619 usage: Some(response.usage),
620 }
621}
622
623pub async fn stream_completion(
624 client: &dyn HttpClient,
625 api_url: &str,
626 api_key: &str,
627 request: Request,
628) -> Result<BoxStream<'static, Result<ResponseStreamEvent>>> {
629 if request.model.starts_with("o1") {
630 let response = complete(client, api_url, api_key, request).await;
631 let response_stream_event = response.map(adapt_response_to_stream);
632 return Ok(stream::once(future::ready(response_stream_event)).boxed());
633 }
634
635 let uri = format!("{api_url}/chat/completions");
636 let request_builder = HttpRequest::builder()
637 .method(Method::POST)
638 .uri(uri)
639 .header("Content-Type", "application/json")
640 .header("Authorization", format!("Bearer {}", api_key));
641
642 let request = request_builder.body(AsyncBody::from(serde_json::to_string(&request)?))?;
643 let mut response = client.send(request).await?;
644 if response.status().is_success() {
645 let reader = BufReader::new(response.into_body());
646 Ok(reader
647 .lines()
648 .filter_map(|line| async move {
649 match line {
650 Ok(line) => {
651 let line = line.strip_prefix("data: ")?;
652 if line == "[DONE]" {
653 None
654 } else {
655 match serde_json::from_str(line) {
656 Ok(ResponseStreamResult::Ok(response)) => Some(Ok(response)),
657 Ok(ResponseStreamResult::Err { error }) => {
658 Some(Err(anyhow!(error)))
659 }
660 Err(error) => Some(Err(anyhow!(error))),
661 }
662 }
663 }
664 Err(error) => Some(Err(anyhow!(error))),
665 }
666 })
667 .boxed())
668 } else {
669 let mut body = String::new();
670 response.body_mut().read_to_string(&mut body).await?;
671
672 #[derive(Deserialize)]
673 struct OpenAiResponse {
674 error: OpenAiError,
675 }
676
677 #[derive(Deserialize)]
678 struct OpenAiError {
679 message: String,
680 }
681
682 match serde_json::from_str::<OpenAiResponse>(&body) {
683 Ok(response) if !response.error.message.is_empty() => Err(anyhow!(
684 "Failed to connect to OpenAI API: {}",
685 response.error.message,
686 )),
687
688 _ => anyhow::bail!(
689 "Failed to connect to OpenAI API: {} {}",
690 response.status(),
691 body,
692 ),
693 }
694 }
695}
696
697#[derive(Copy, Clone, Serialize, Deserialize)]
698pub enum OpenAiEmbeddingModel {
699 #[serde(rename = "text-embedding-3-small")]
700 TextEmbedding3Small,
701 #[serde(rename = "text-embedding-3-large")]
702 TextEmbedding3Large,
703}
704
705#[derive(Serialize)]
706struct OpenAiEmbeddingRequest<'a> {
707 model: OpenAiEmbeddingModel,
708 input: Vec<&'a str>,
709}
710
711#[derive(Deserialize)]
712pub struct OpenAiEmbeddingResponse {
713 pub data: Vec<OpenAiEmbedding>,
714}
715
716#[derive(Deserialize)]
717pub struct OpenAiEmbedding {
718 pub embedding: Vec<f32>,
719}
720
721pub fn embed<'a>(
722 client: &dyn HttpClient,
723 api_url: &str,
724 api_key: &str,
725 model: OpenAiEmbeddingModel,
726 texts: impl IntoIterator<Item = &'a str>,
727) -> impl 'static + Future<Output = Result<OpenAiEmbeddingResponse>> {
728 let uri = format!("{api_url}/embeddings");
729
730 let request = OpenAiEmbeddingRequest {
731 model,
732 input: texts.into_iter().collect(),
733 };
734 let body = AsyncBody::from(serde_json::to_string(&request).unwrap());
735 let request = HttpRequest::builder()
736 .method(Method::POST)
737 .uri(uri)
738 .header("Content-Type", "application/json")
739 .header("Authorization", format!("Bearer {}", api_key))
740 .body(body)
741 .map(|request| client.send(request));
742
743 async move {
744 let mut response = request?.await?;
745 let mut body = String::new();
746 response.body_mut().read_to_string(&mut body).await?;
747
748 anyhow::ensure!(
749 response.status().is_success(),
750 "error during embedding, status: {:?}, body: {:?}",
751 response.status(),
752 body
753 );
754 let response: OpenAiEmbeddingResponse =
755 serde_json::from_str(&body).context("failed to parse OpenAI embedding response")?;
756 Ok(response)
757 }
758}