1use ai_onboarding::YoungAccountBanner;
2use anthropic::AnthropicModelMode;
3use anyhow::{Context as _, Result, anyhow};
4use chrono::{DateTime, Utc};
5use client::{Client, UserStore, zed_urls};
6use cloud_api_types::Plan;
7use cloud_llm_client::{
8 CLIENT_SUPPORTS_STATUS_MESSAGES_HEADER_NAME, CLIENT_SUPPORTS_X_AI_HEADER_NAME, CompletionBody,
9 CompletionEvent, CompletionRequestStatus, CountTokensBody, CountTokensResponse,
10 ListModelsResponse, SERVER_SUPPORTS_STATUS_MESSAGES_HEADER_NAME, ZED_VERSION_HEADER_NAME,
11};
12use futures::{
13 AsyncBufReadExt, FutureExt, Stream, StreamExt,
14 future::BoxFuture,
15 stream::{self, BoxStream},
16};
17use google_ai::GoogleModelMode;
18use gpui::{AnyElement, AnyView, App, AsyncApp, Context, Entity, Subscription, Task};
19use http_client::http::{HeaderMap, HeaderValue};
20use http_client::{AsyncBody, HttpClient, HttpRequestExt, Method, Response, StatusCode};
21use language_model::{
22 AuthenticateError, IconOrSvg, LanguageModel, LanguageModelCacheConfiguration,
23 LanguageModelCompletionError, LanguageModelCompletionEvent, LanguageModelEffortLevel,
24 LanguageModelId, LanguageModelName, LanguageModelProvider, LanguageModelProviderId,
25 LanguageModelProviderName, LanguageModelProviderState, LanguageModelRequest,
26 LanguageModelToolChoice, LanguageModelToolSchemaFormat, LlmApiToken, NeedsLlmTokenRefresh,
27 PaymentRequiredError, RateLimiter, RefreshLlmTokenListener,
28};
29use release_channel::AppVersion;
30use schemars::JsonSchema;
31use semver::Version;
32use serde::{Deserialize, Serialize, de::DeserializeOwned};
33use settings::SettingsStore;
34pub use settings::ZedDotDevAvailableModel as AvailableModel;
35pub use settings::ZedDotDevAvailableProvider as AvailableProvider;
36use smol::io::{AsyncReadExt, BufReader};
37use std::collections::VecDeque;
38use std::pin::Pin;
39use std::str::FromStr;
40use std::sync::Arc;
41use std::task::Poll;
42use std::time::Duration;
43use thiserror::Error;
44use ui::{TintColor, prelude::*};
45use util::{ResultExt as _, maybe};
46
47use crate::provider::anthropic::{
48 AnthropicEventMapper, count_anthropic_tokens_with_tiktoken, into_anthropic,
49};
50use crate::provider::google::{GoogleEventMapper, into_google};
51use crate::provider::open_ai::{
52 OpenAiEventMapper, OpenAiResponseEventMapper, count_open_ai_tokens, into_open_ai,
53 into_open_ai_response,
54};
55use crate::provider::x_ai::count_xai_tokens;
56
57const PROVIDER_ID: LanguageModelProviderId = language_model::ZED_CLOUD_PROVIDER_ID;
58const PROVIDER_NAME: LanguageModelProviderName = language_model::ZED_CLOUD_PROVIDER_NAME;
59
60#[derive(Default, Clone, Debug, PartialEq)]
61pub struct ZedDotDevSettings {
62 pub available_models: Vec<AvailableModel>,
63}
64#[derive(Default, Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]
65#[serde(tag = "type", rename_all = "lowercase")]
66pub enum ModelMode {
67 #[default]
68 Default,
69 Thinking {
70 /// The maximum number of tokens to use for reasoning. Must be lower than the model's `max_output_tokens`.
71 budget_tokens: Option<u32>,
72 },
73}
74
75impl From<ModelMode> for AnthropicModelMode {
76 fn from(value: ModelMode) -> Self {
77 match value {
78 ModelMode::Default => AnthropicModelMode::Default,
79 ModelMode::Thinking { budget_tokens } => AnthropicModelMode::Thinking { budget_tokens },
80 }
81 }
82}
83
84pub struct CloudLanguageModelProvider {
85 client: Arc<Client>,
86 state: Entity<State>,
87 _maintain_client_status: Task<()>,
88}
89
90pub struct State {
91 client: Arc<Client>,
92 llm_api_token: LlmApiToken,
93 user_store: Entity<UserStore>,
94 status: client::Status,
95 models: Vec<Arc<cloud_llm_client::LanguageModel>>,
96 default_model: Option<Arc<cloud_llm_client::LanguageModel>>,
97 default_fast_model: Option<Arc<cloud_llm_client::LanguageModel>>,
98 recommended_models: Vec<Arc<cloud_llm_client::LanguageModel>>,
99 _fetch_models_task: Task<()>,
100 _settings_subscription: Subscription,
101 _llm_token_subscription: Subscription,
102}
103
104impl State {
105 fn new(
106 client: Arc<Client>,
107 user_store: Entity<UserStore>,
108 status: client::Status,
109 cx: &mut Context<Self>,
110 ) -> Self {
111 let refresh_llm_token_listener = RefreshLlmTokenListener::global(cx);
112 let mut current_user = user_store.read(cx).watch_current_user();
113 Self {
114 client: client.clone(),
115 llm_api_token: LlmApiToken::default(),
116 user_store,
117 status,
118 models: Vec::new(),
119 default_model: None,
120 default_fast_model: None,
121 recommended_models: Vec::new(),
122 _fetch_models_task: cx.spawn(async move |this, cx| {
123 maybe!(async move {
124 let (client, llm_api_token) = this
125 .read_with(cx, |this, _cx| (client.clone(), this.llm_api_token.clone()))?;
126
127 while current_user.borrow().is_none() {
128 current_user.next().await;
129 }
130
131 let response =
132 Self::fetch_models(client.clone(), llm_api_token.clone()).await?;
133 this.update(cx, |this, cx| this.update_models(response, cx))?;
134 anyhow::Ok(())
135 })
136 .await
137 .context("failed to fetch Zed models")
138 .log_err();
139 }),
140 _settings_subscription: cx.observe_global::<SettingsStore>(|_, cx| {
141 cx.notify();
142 }),
143 _llm_token_subscription: cx.subscribe(
144 &refresh_llm_token_listener,
145 move |this, _listener, _event, cx| {
146 let client = this.client.clone();
147 let llm_api_token = this.llm_api_token.clone();
148 cx.spawn(async move |this, cx| {
149 llm_api_token.refresh(&client).await?;
150 let response = Self::fetch_models(client, llm_api_token).await?;
151 this.update(cx, |this, cx| {
152 this.update_models(response, cx);
153 })
154 })
155 .detach_and_log_err(cx);
156 },
157 ),
158 }
159 }
160
161 fn is_signed_out(&self, cx: &App) -> bool {
162 self.user_store.read(cx).current_user().is_none()
163 }
164
165 fn authenticate(&self, cx: &mut Context<Self>) -> Task<Result<()>> {
166 let client = self.client.clone();
167 cx.spawn(async move |state, cx| {
168 client.sign_in_with_optional_connect(true, cx).await?;
169 state.update(cx, |_, cx| cx.notify())
170 })
171 }
172
173 fn update_models(&mut self, response: ListModelsResponse, cx: &mut Context<Self>) {
174 let mut models = Vec::new();
175
176 for model in response.models {
177 models.push(Arc::new(model.clone()));
178 }
179
180 self.default_model = models
181 .iter()
182 .find(|model| {
183 response
184 .default_model
185 .as_ref()
186 .is_some_and(|default_model_id| &model.id == default_model_id)
187 })
188 .cloned();
189 self.default_fast_model = models
190 .iter()
191 .find(|model| {
192 response
193 .default_fast_model
194 .as_ref()
195 .is_some_and(|default_fast_model_id| &model.id == default_fast_model_id)
196 })
197 .cloned();
198 self.recommended_models = response
199 .recommended_models
200 .iter()
201 .filter_map(|id| models.iter().find(|model| &model.id == id))
202 .cloned()
203 .collect();
204 self.models = models;
205 cx.notify();
206 }
207
208 async fn fetch_models(
209 client: Arc<Client>,
210 llm_api_token: LlmApiToken,
211 ) -> Result<ListModelsResponse> {
212 let http_client = &client.http_client();
213 let token = llm_api_token.acquire(&client).await?;
214
215 let request = http_client::Request::builder()
216 .method(Method::GET)
217 .header(CLIENT_SUPPORTS_X_AI_HEADER_NAME, "true")
218 .uri(http_client.build_zed_llm_url("/models", &[])?.as_ref())
219 .header("Authorization", format!("Bearer {token}"))
220 .body(AsyncBody::empty())?;
221 let mut response = http_client
222 .send(request)
223 .await
224 .context("failed to send list models request")?;
225
226 if response.status().is_success() {
227 let mut body = String::new();
228 response.body_mut().read_to_string(&mut body).await?;
229 Ok(serde_json::from_str(&body)?)
230 } else {
231 let mut body = String::new();
232 response.body_mut().read_to_string(&mut body).await?;
233 anyhow::bail!(
234 "error listing models.\nStatus: {:?}\nBody: {body}",
235 response.status(),
236 );
237 }
238 }
239}
240
241impl CloudLanguageModelProvider {
242 pub fn new(user_store: Entity<UserStore>, client: Arc<Client>, cx: &mut App) -> Self {
243 let mut status_rx = client.status();
244 let status = *status_rx.borrow();
245
246 let state = cx.new(|cx| State::new(client.clone(), user_store.clone(), status, cx));
247
248 let state_ref = state.downgrade();
249 let maintain_client_status = cx.spawn(async move |cx| {
250 while let Some(status) = status_rx.next().await {
251 if let Some(this) = state_ref.upgrade() {
252 _ = this.update(cx, |this, cx| {
253 if this.status != status {
254 this.status = status;
255 cx.notify();
256 }
257 });
258 } else {
259 break;
260 }
261 }
262 });
263
264 Self {
265 client,
266 state,
267 _maintain_client_status: maintain_client_status,
268 }
269 }
270
271 fn create_language_model(
272 &self,
273 model: Arc<cloud_llm_client::LanguageModel>,
274 llm_api_token: LlmApiToken,
275 ) -> Arc<dyn LanguageModel> {
276 Arc::new(CloudLanguageModel {
277 id: LanguageModelId(SharedString::from(model.id.0.clone())),
278 model,
279 llm_api_token,
280 client: self.client.clone(),
281 request_limiter: RateLimiter::new(4),
282 })
283 }
284}
285
286impl LanguageModelProviderState for CloudLanguageModelProvider {
287 type ObservableEntity = State;
288
289 fn observable_entity(&self) -> Option<Entity<Self::ObservableEntity>> {
290 Some(self.state.clone())
291 }
292}
293
294impl LanguageModelProvider for CloudLanguageModelProvider {
295 fn id(&self) -> LanguageModelProviderId {
296 PROVIDER_ID
297 }
298
299 fn name(&self) -> LanguageModelProviderName {
300 PROVIDER_NAME
301 }
302
303 fn icon(&self) -> IconOrSvg {
304 IconOrSvg::Icon(IconName::AiZed)
305 }
306
307 fn default_model(&self, cx: &App) -> Option<Arc<dyn LanguageModel>> {
308 let default_model = self.state.read(cx).default_model.clone()?;
309 let llm_api_token = self.state.read(cx).llm_api_token.clone();
310 Some(self.create_language_model(default_model, llm_api_token))
311 }
312
313 fn default_fast_model(&self, cx: &App) -> Option<Arc<dyn LanguageModel>> {
314 let default_fast_model = self.state.read(cx).default_fast_model.clone()?;
315 let llm_api_token = self.state.read(cx).llm_api_token.clone();
316 Some(self.create_language_model(default_fast_model, llm_api_token))
317 }
318
319 fn recommended_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {
320 let llm_api_token = self.state.read(cx).llm_api_token.clone();
321 self.state
322 .read(cx)
323 .recommended_models
324 .iter()
325 .cloned()
326 .map(|model| self.create_language_model(model, llm_api_token.clone()))
327 .collect()
328 }
329
330 fn provided_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {
331 let llm_api_token = self.state.read(cx).llm_api_token.clone();
332 self.state
333 .read(cx)
334 .models
335 .iter()
336 .cloned()
337 .map(|model| self.create_language_model(model, llm_api_token.clone()))
338 .collect()
339 }
340
341 fn is_authenticated(&self, cx: &App) -> bool {
342 let state = self.state.read(cx);
343 !state.is_signed_out(cx)
344 }
345
346 fn authenticate(&self, _cx: &mut App) -> Task<Result<(), AuthenticateError>> {
347 Task::ready(Ok(()))
348 }
349
350 fn configuration_view(
351 &self,
352 _target_agent: language_model::ConfigurationViewTargetAgent,
353 _: &mut Window,
354 cx: &mut App,
355 ) -> AnyView {
356 cx.new(|_| ConfigurationView::new(self.state.clone()))
357 .into()
358 }
359
360 fn reset_credentials(&self, _cx: &mut App) -> Task<Result<()>> {
361 Task::ready(Ok(()))
362 }
363}
364
365pub struct CloudLanguageModel {
366 id: LanguageModelId,
367 model: Arc<cloud_llm_client::LanguageModel>,
368 llm_api_token: LlmApiToken,
369 client: Arc<Client>,
370 request_limiter: RateLimiter,
371}
372
373struct PerformLlmCompletionResponse {
374 response: Response<AsyncBody>,
375 includes_status_messages: bool,
376}
377
378impl CloudLanguageModel {
379 async fn perform_llm_completion(
380 client: Arc<Client>,
381 llm_api_token: LlmApiToken,
382 app_version: Option<Version>,
383 body: CompletionBody,
384 ) -> Result<PerformLlmCompletionResponse> {
385 let http_client = &client.http_client();
386
387 let mut token = llm_api_token.acquire(&client).await?;
388 let mut refreshed_token = false;
389
390 loop {
391 let request = http_client::Request::builder()
392 .method(Method::POST)
393 .uri(http_client.build_zed_llm_url("/completions", &[])?.as_ref())
394 .when_some(app_version.as_ref(), |builder, app_version| {
395 builder.header(ZED_VERSION_HEADER_NAME, app_version.to_string())
396 })
397 .header("Content-Type", "application/json")
398 .header("Authorization", format!("Bearer {token}"))
399 .header(CLIENT_SUPPORTS_STATUS_MESSAGES_HEADER_NAME, "true")
400 // TODO: Uncomment once the cloud-side StreamEnded support PR is merged.
401 // .header(CLIENT_SUPPORTS_STATUS_STREAM_ENDED_HEADER_NAME, "true")
402 .body(serde_json::to_string(&body)?.into())?;
403
404 let mut response = http_client.send(request).await?;
405 let status = response.status();
406 if status.is_success() {
407 let includes_status_messages = response
408 .headers()
409 .get(SERVER_SUPPORTS_STATUS_MESSAGES_HEADER_NAME)
410 .is_some();
411
412 return Ok(PerformLlmCompletionResponse {
413 response,
414 includes_status_messages,
415 });
416 }
417
418 if !refreshed_token && response.needs_llm_token_refresh() {
419 token = llm_api_token.refresh(&client).await?;
420 refreshed_token = true;
421 continue;
422 }
423
424 if status == StatusCode::PAYMENT_REQUIRED {
425 return Err(anyhow!(PaymentRequiredError));
426 }
427
428 let mut body = String::new();
429 let headers = response.headers().clone();
430 response.body_mut().read_to_string(&mut body).await?;
431 return Err(anyhow!(ApiError {
432 status,
433 body,
434 headers
435 }));
436 }
437 }
438}
439
440#[derive(Debug, Error)]
441#[error("cloud language model request failed with status {status}: {body}")]
442struct ApiError {
443 status: StatusCode,
444 body: String,
445 headers: HeaderMap<HeaderValue>,
446}
447
448/// Represents error responses from Zed's cloud API.
449///
450/// Example JSON for an upstream HTTP error:
451/// ```json
452/// {
453/// "code": "upstream_http_error",
454/// "message": "Received an error from the Anthropic API: upstream connect error or disconnect/reset before headers, reset reason: connection timeout",
455/// "upstream_status": 503
456/// }
457/// ```
458#[derive(Debug, serde::Deserialize)]
459struct CloudApiError {
460 code: String,
461 message: String,
462 #[serde(default)]
463 #[serde(deserialize_with = "deserialize_optional_status_code")]
464 upstream_status: Option<StatusCode>,
465 #[serde(default)]
466 retry_after: Option<f64>,
467}
468
469fn deserialize_optional_status_code<'de, D>(deserializer: D) -> Result<Option<StatusCode>, D::Error>
470where
471 D: serde::Deserializer<'de>,
472{
473 let opt: Option<u16> = Option::deserialize(deserializer)?;
474 Ok(opt.and_then(|code| StatusCode::from_u16(code).ok()))
475}
476
477impl From<ApiError> for LanguageModelCompletionError {
478 fn from(error: ApiError) -> Self {
479 if let Ok(cloud_error) = serde_json::from_str::<CloudApiError>(&error.body) {
480 if cloud_error.code.starts_with("upstream_http_") {
481 let status = if let Some(status) = cloud_error.upstream_status {
482 status
483 } else if cloud_error.code.ends_with("_error") {
484 error.status
485 } else {
486 // If there's a status code in the code string (e.g. "upstream_http_429")
487 // then use that; otherwise, see if the JSON contains a status code.
488 cloud_error
489 .code
490 .strip_prefix("upstream_http_")
491 .and_then(|code_str| code_str.parse::<u16>().ok())
492 .and_then(|code| StatusCode::from_u16(code).ok())
493 .unwrap_or(error.status)
494 };
495
496 return LanguageModelCompletionError::UpstreamProviderError {
497 message: cloud_error.message,
498 status,
499 retry_after: cloud_error.retry_after.map(Duration::from_secs_f64),
500 };
501 }
502
503 return LanguageModelCompletionError::from_http_status(
504 PROVIDER_NAME,
505 error.status,
506 cloud_error.message,
507 None,
508 );
509 }
510
511 let retry_after = None;
512 LanguageModelCompletionError::from_http_status(
513 PROVIDER_NAME,
514 error.status,
515 error.body,
516 retry_after,
517 )
518 }
519}
520
521impl LanguageModel for CloudLanguageModel {
522 fn id(&self) -> LanguageModelId {
523 self.id.clone()
524 }
525
526 fn name(&self) -> LanguageModelName {
527 LanguageModelName::from(self.model.display_name.clone())
528 }
529
530 fn provider_id(&self) -> LanguageModelProviderId {
531 PROVIDER_ID
532 }
533
534 fn provider_name(&self) -> LanguageModelProviderName {
535 PROVIDER_NAME
536 }
537
538 fn upstream_provider_id(&self) -> LanguageModelProviderId {
539 use cloud_llm_client::LanguageModelProvider::*;
540 match self.model.provider {
541 Anthropic => language_model::ANTHROPIC_PROVIDER_ID,
542 OpenAi => language_model::OPEN_AI_PROVIDER_ID,
543 Google => language_model::GOOGLE_PROVIDER_ID,
544 XAi => language_model::X_AI_PROVIDER_ID,
545 }
546 }
547
548 fn upstream_provider_name(&self) -> LanguageModelProviderName {
549 use cloud_llm_client::LanguageModelProvider::*;
550 match self.model.provider {
551 Anthropic => language_model::ANTHROPIC_PROVIDER_NAME,
552 OpenAi => language_model::OPEN_AI_PROVIDER_NAME,
553 Google => language_model::GOOGLE_PROVIDER_NAME,
554 XAi => language_model::X_AI_PROVIDER_NAME,
555 }
556 }
557
558 fn is_latest(&self) -> bool {
559 self.model.is_latest
560 }
561
562 fn supports_tools(&self) -> bool {
563 self.model.supports_tools
564 }
565
566 fn supports_images(&self) -> bool {
567 self.model.supports_images
568 }
569
570 fn supports_thinking(&self) -> bool {
571 self.model.supports_thinking
572 }
573
574 fn supported_effort_levels(&self) -> Vec<LanguageModelEffortLevel> {
575 self.model
576 .supported_effort_levels
577 .iter()
578 .map(|effort_level| LanguageModelEffortLevel {
579 name: effort_level.name.clone().into(),
580 value: effort_level.value.clone().into(),
581 is_default: effort_level.is_default.unwrap_or(false),
582 })
583 .collect()
584 }
585
586 fn supports_streaming_tools(&self) -> bool {
587 self.model.supports_streaming_tools
588 }
589
590 fn supports_tool_choice(&self, choice: LanguageModelToolChoice) -> bool {
591 match choice {
592 LanguageModelToolChoice::Auto
593 | LanguageModelToolChoice::Any
594 | LanguageModelToolChoice::None => true,
595 }
596 }
597
598 fn supports_split_token_display(&self) -> bool {
599 use cloud_llm_client::LanguageModelProvider::*;
600 matches!(self.model.provider, OpenAi)
601 }
602
603 fn telemetry_id(&self) -> String {
604 format!("zed.dev/{}", self.model.id)
605 }
606
607 fn tool_input_format(&self) -> LanguageModelToolSchemaFormat {
608 match self.model.provider {
609 cloud_llm_client::LanguageModelProvider::Anthropic
610 | cloud_llm_client::LanguageModelProvider::OpenAi
611 | cloud_llm_client::LanguageModelProvider::XAi => {
612 LanguageModelToolSchemaFormat::JsonSchema
613 }
614 cloud_llm_client::LanguageModelProvider::Google => {
615 LanguageModelToolSchemaFormat::JsonSchemaSubset
616 }
617 }
618 }
619
620 fn max_token_count(&self) -> u64 {
621 self.model.max_token_count as u64
622 }
623
624 fn max_output_tokens(&self) -> Option<u64> {
625 Some(self.model.max_output_tokens as u64)
626 }
627
628 fn cache_configuration(&self) -> Option<LanguageModelCacheConfiguration> {
629 match &self.model.provider {
630 cloud_llm_client::LanguageModelProvider::Anthropic => {
631 Some(LanguageModelCacheConfiguration {
632 min_total_token: 2_048,
633 should_speculate: true,
634 max_cache_anchors: 4,
635 })
636 }
637 cloud_llm_client::LanguageModelProvider::OpenAi
638 | cloud_llm_client::LanguageModelProvider::XAi
639 | cloud_llm_client::LanguageModelProvider::Google => None,
640 }
641 }
642
643 fn count_tokens(
644 &self,
645 request: LanguageModelRequest,
646 cx: &App,
647 ) -> BoxFuture<'static, Result<u64>> {
648 match self.model.provider {
649 cloud_llm_client::LanguageModelProvider::Anthropic => cx
650 .background_spawn(async move { count_anthropic_tokens_with_tiktoken(request) })
651 .boxed(),
652 cloud_llm_client::LanguageModelProvider::OpenAi => {
653 let model = match open_ai::Model::from_id(&self.model.id.0) {
654 Ok(model) => model,
655 Err(err) => return async move { Err(anyhow!(err)) }.boxed(),
656 };
657 count_open_ai_tokens(request, model, cx)
658 }
659 cloud_llm_client::LanguageModelProvider::XAi => {
660 let model = match x_ai::Model::from_id(&self.model.id.0) {
661 Ok(model) => model,
662 Err(err) => return async move { Err(anyhow!(err)) }.boxed(),
663 };
664 count_xai_tokens(request, model, cx)
665 }
666 cloud_llm_client::LanguageModelProvider::Google => {
667 let client = self.client.clone();
668 let llm_api_token = self.llm_api_token.clone();
669 let model_id = self.model.id.to_string();
670 let generate_content_request =
671 into_google(request, model_id.clone(), GoogleModelMode::Default);
672 async move {
673 let http_client = &client.http_client();
674 let token = llm_api_token.acquire(&client).await?;
675
676 let request_body = CountTokensBody {
677 provider: cloud_llm_client::LanguageModelProvider::Google,
678 model: model_id,
679 provider_request: serde_json::to_value(&google_ai::CountTokensRequest {
680 generate_content_request,
681 })?,
682 };
683 let request = http_client::Request::builder()
684 .method(Method::POST)
685 .uri(
686 http_client
687 .build_zed_llm_url("/count_tokens", &[])?
688 .as_ref(),
689 )
690 .header("Content-Type", "application/json")
691 .header("Authorization", format!("Bearer {token}"))
692 .body(serde_json::to_string(&request_body)?.into())?;
693 let mut response = http_client.send(request).await?;
694 let status = response.status();
695 let headers = response.headers().clone();
696 let mut response_body = String::new();
697 response
698 .body_mut()
699 .read_to_string(&mut response_body)
700 .await?;
701
702 if status.is_success() {
703 let response_body: CountTokensResponse =
704 serde_json::from_str(&response_body)?;
705
706 Ok(response_body.tokens as u64)
707 } else {
708 Err(anyhow!(ApiError {
709 status,
710 body: response_body,
711 headers
712 }))
713 }
714 }
715 .boxed()
716 }
717 }
718 }
719
720 fn stream_completion(
721 &self,
722 request: LanguageModelRequest,
723 cx: &AsyncApp,
724 ) -> BoxFuture<
725 'static,
726 Result<
727 BoxStream<'static, Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>,
728 LanguageModelCompletionError,
729 >,
730 > {
731 let thread_id = request.thread_id.clone();
732 let prompt_id = request.prompt_id.clone();
733 let intent = request.intent;
734 let app_version = Some(cx.update(|cx| AppVersion::global(cx)));
735 let thinking_allowed = request.thinking_allowed;
736 let enable_thinking = thinking_allowed && self.model.supports_thinking;
737 let provider_name = provider_name(&self.model.provider);
738 match self.model.provider {
739 cloud_llm_client::LanguageModelProvider::Anthropic => {
740 let effort = request
741 .thinking_effort
742 .as_ref()
743 .and_then(|effort| anthropic::Effort::from_str(effort).ok());
744
745 let mut request = into_anthropic(
746 request,
747 self.model.id.to_string(),
748 1.0,
749 self.model.max_output_tokens as u64,
750 if enable_thinking {
751 AnthropicModelMode::Thinking {
752 budget_tokens: Some(4_096),
753 }
754 } else {
755 AnthropicModelMode::Default
756 },
757 );
758
759 if enable_thinking && effort.is_some() {
760 request.thinking = Some(anthropic::Thinking::Adaptive);
761 request.output_config = Some(anthropic::OutputConfig { effort });
762 }
763
764 let client = self.client.clone();
765 let llm_api_token = self.llm_api_token.clone();
766 let future = self.request_limiter.stream(async move {
767 let PerformLlmCompletionResponse {
768 response,
769 includes_status_messages,
770 } = Self::perform_llm_completion(
771 client.clone(),
772 llm_api_token,
773 app_version,
774 CompletionBody {
775 thread_id,
776 prompt_id,
777 intent,
778 provider: cloud_llm_client::LanguageModelProvider::Anthropic,
779 model: request.model.clone(),
780 provider_request: serde_json::to_value(&request)
781 .map_err(|e| anyhow!(e))?,
782 },
783 )
784 .await
785 .map_err(|err| match err.downcast::<ApiError>() {
786 Ok(api_err) => anyhow!(LanguageModelCompletionError::from(api_err)),
787 Err(err) => anyhow!(err),
788 })?;
789
790 let mut mapper = AnthropicEventMapper::new();
791 Ok(map_cloud_completion_events(
792 Box::pin(response_lines(response, includes_status_messages)),
793 &provider_name,
794 move |event| mapper.map_event(event),
795 ))
796 });
797 async move { Ok(future.await?.boxed()) }.boxed()
798 }
799 cloud_llm_client::LanguageModelProvider::OpenAi => {
800 let client = self.client.clone();
801 let llm_api_token = self.llm_api_token.clone();
802 let effort = request
803 .thinking_effort
804 .as_ref()
805 .and_then(|effort| open_ai::ReasoningEffort::from_str(effort).ok());
806
807 let mut request = into_open_ai_response(
808 request,
809 &self.model.id.0,
810 self.model.supports_parallel_tool_calls,
811 true,
812 None,
813 None,
814 );
815
816 if enable_thinking && let Some(effort) = effort {
817 request.reasoning = Some(open_ai::responses::ReasoningConfig { effort });
818 }
819
820 let future = self.request_limiter.stream(async move {
821 let PerformLlmCompletionResponse {
822 response,
823 includes_status_messages,
824 } = Self::perform_llm_completion(
825 client.clone(),
826 llm_api_token,
827 app_version,
828 CompletionBody {
829 thread_id,
830 prompt_id,
831 intent,
832 provider: cloud_llm_client::LanguageModelProvider::OpenAi,
833 model: request.model.clone(),
834 provider_request: serde_json::to_value(&request)
835 .map_err(|e| anyhow!(e))?,
836 },
837 )
838 .await?;
839
840 let mut mapper = OpenAiResponseEventMapper::new();
841 Ok(map_cloud_completion_events(
842 Box::pin(response_lines(response, includes_status_messages)),
843 &provider_name,
844 move |event| mapper.map_event(event),
845 ))
846 });
847 async move { Ok(future.await?.boxed()) }.boxed()
848 }
849 cloud_llm_client::LanguageModelProvider::XAi => {
850 let client = self.client.clone();
851 let request = into_open_ai(
852 request,
853 &self.model.id.0,
854 self.model.supports_parallel_tool_calls,
855 false,
856 None,
857 None,
858 );
859 let llm_api_token = self.llm_api_token.clone();
860 let future = self.request_limiter.stream(async move {
861 let PerformLlmCompletionResponse {
862 response,
863 includes_status_messages,
864 } = Self::perform_llm_completion(
865 client.clone(),
866 llm_api_token,
867 app_version,
868 CompletionBody {
869 thread_id,
870 prompt_id,
871 intent,
872 provider: cloud_llm_client::LanguageModelProvider::XAi,
873 model: request.model.clone(),
874 provider_request: serde_json::to_value(&request)
875 .map_err(|e| anyhow!(e))?,
876 },
877 )
878 .await?;
879
880 let mut mapper = OpenAiEventMapper::new();
881 Ok(map_cloud_completion_events(
882 Box::pin(response_lines(response, includes_status_messages)),
883 &provider_name,
884 move |event| mapper.map_event(event),
885 ))
886 });
887 async move { Ok(future.await?.boxed()) }.boxed()
888 }
889 cloud_llm_client::LanguageModelProvider::Google => {
890 let client = self.client.clone();
891 let request =
892 into_google(request, self.model.id.to_string(), GoogleModelMode::Default);
893 let llm_api_token = self.llm_api_token.clone();
894 let future = self.request_limiter.stream(async move {
895 let PerformLlmCompletionResponse {
896 response,
897 includes_status_messages,
898 } = Self::perform_llm_completion(
899 client.clone(),
900 llm_api_token,
901 app_version,
902 CompletionBody {
903 thread_id,
904 prompt_id,
905 intent,
906 provider: cloud_llm_client::LanguageModelProvider::Google,
907 model: request.model.model_id.clone(),
908 provider_request: serde_json::to_value(&request)
909 .map_err(|e| anyhow!(e))?,
910 },
911 )
912 .await?;
913
914 let mut mapper = GoogleEventMapper::new();
915 Ok(map_cloud_completion_events(
916 Box::pin(response_lines(response, includes_status_messages)),
917 &provider_name,
918 move |event| mapper.map_event(event),
919 ))
920 });
921 async move { Ok(future.await?.boxed()) }.boxed()
922 }
923 }
924 }
925}
926
927fn map_cloud_completion_events<T, F>(
928 stream: Pin<Box<dyn Stream<Item = Result<CompletionEvent<T>>> + Send>>,
929 provider: &LanguageModelProviderName,
930 mut map_callback: F,
931) -> BoxStream<'static, Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
932where
933 T: DeserializeOwned + 'static,
934 F: FnMut(T) -> Vec<Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
935 + Send
936 + 'static,
937{
938 let provider = provider.clone();
939 let mut stream = stream.fuse();
940
941 // TODO: Uncomment once the cloud-side StreamEnded support PR is merged.
942 // let mut saw_stream_ended = false;
943
944 let mut done = false;
945 let mut pending = VecDeque::new();
946
947 stream::poll_fn(move |cx| {
948 loop {
949 if let Some(item) = pending.pop_front() {
950 return Poll::Ready(Some(item));
951 }
952
953 if done {
954 return Poll::Ready(None);
955 }
956
957 match stream.poll_next_unpin(cx) {
958 Poll::Ready(Some(event)) => {
959 let items = match event {
960 Err(error) => {
961 vec![Err(LanguageModelCompletionError::from(error))]
962 }
963 Ok(CompletionEvent::Status(CompletionRequestStatus::StreamEnded)) => {
964 // TODO: Uncomment once the cloud-side StreamEnded support PR is merged.
965 // let mut saw_stream_ended = false;
966 //
967 // saw_stream_ended = true;
968 vec![]
969 }
970 Ok(CompletionEvent::Status(status)) => {
971 LanguageModelCompletionEvent::from_completion_request_status(
972 status,
973 provider.clone(),
974 )
975 .transpose()
976 .map(|event| vec![event])
977 .unwrap_or_default()
978 }
979 Ok(CompletionEvent::Event(event)) => map_callback(event),
980 };
981 pending.extend(items);
982 }
983 Poll::Ready(None) => {
984 done = true;
985
986 // TODO: Uncomment once the cloud-side StreamEnded support PR is merged.
987 //
988 // if !saw_stream_ended {
989 // return Poll::Ready(Some(Err(
990 // LanguageModelCompletionError::StreamEndedUnexpectedly {
991 // provider: provider.clone(),
992 // },
993 // )));
994 // }
995 }
996 Poll::Pending => return Poll::Pending,
997 }
998 }
999 })
1000 .boxed()
1001}
1002
1003fn provider_name(provider: &cloud_llm_client::LanguageModelProvider) -> LanguageModelProviderName {
1004 match provider {
1005 cloud_llm_client::LanguageModelProvider::Anthropic => {
1006 language_model::ANTHROPIC_PROVIDER_NAME
1007 }
1008 cloud_llm_client::LanguageModelProvider::OpenAi => language_model::OPEN_AI_PROVIDER_NAME,
1009 cloud_llm_client::LanguageModelProvider::Google => language_model::GOOGLE_PROVIDER_NAME,
1010 cloud_llm_client::LanguageModelProvider::XAi => language_model::X_AI_PROVIDER_NAME,
1011 }
1012}
1013
1014fn response_lines<T: DeserializeOwned>(
1015 response: Response<AsyncBody>,
1016 includes_status_messages: bool,
1017) -> impl Stream<Item = Result<CompletionEvent<T>>> {
1018 futures::stream::try_unfold(
1019 (String::new(), BufReader::new(response.into_body())),
1020 move |(mut line, mut body)| async move {
1021 match body.read_line(&mut line).await {
1022 Ok(0) => Ok(None),
1023 Ok(_) => {
1024 let event = if includes_status_messages {
1025 serde_json::from_str::<CompletionEvent<T>>(&line)?
1026 } else {
1027 CompletionEvent::Event(serde_json::from_str::<T>(&line)?)
1028 };
1029
1030 line.clear();
1031 Ok(Some((event, (line, body))))
1032 }
1033 Err(e) => Err(e.into()),
1034 }
1035 },
1036 )
1037}
1038
1039#[derive(IntoElement, RegisterComponent)]
1040struct ZedAiConfiguration {
1041 is_connected: bool,
1042 plan: Option<Plan>,
1043 subscription_period: Option<(DateTime<Utc>, DateTime<Utc>)>,
1044 eligible_for_trial: bool,
1045 account_too_young: bool,
1046 sign_in_callback: Arc<dyn Fn(&mut Window, &mut App) + Send + Sync>,
1047}
1048
1049impl RenderOnce for ZedAiConfiguration {
1050 fn render(self, _window: &mut Window, _cx: &mut App) -> impl IntoElement {
1051 let is_pro = self.plan.is_some_and(|plan| plan == Plan::ZedPro);
1052 let subscription_text = match (self.plan, self.subscription_period) {
1053 (Some(Plan::ZedPro), Some(_)) => {
1054 "You have access to Zed's hosted models through your Pro subscription."
1055 }
1056 (Some(Plan::ZedProTrial), Some(_)) => {
1057 "You have access to Zed's hosted models through your Pro trial."
1058 }
1059 (Some(Plan::ZedFree), Some(_)) => {
1060 if self.eligible_for_trial {
1061 "Subscribe for access to Zed's hosted models. Start with a 14 day free trial."
1062 } else {
1063 "Subscribe for access to Zed's hosted models."
1064 }
1065 }
1066 _ => {
1067 if self.eligible_for_trial {
1068 "Subscribe for access to Zed's hosted models. Start with a 14 day free trial."
1069 } else {
1070 "Subscribe for access to Zed's hosted models."
1071 }
1072 }
1073 };
1074
1075 let manage_subscription_buttons = if is_pro {
1076 Button::new("manage_settings", "Manage Subscription")
1077 .full_width()
1078 .style(ButtonStyle::Tinted(TintColor::Accent))
1079 .on_click(|_, _, cx| cx.open_url(&zed_urls::account_url(cx)))
1080 .into_any_element()
1081 } else if self.plan.is_none() || self.eligible_for_trial {
1082 Button::new("start_trial", "Start 14-day Free Pro Trial")
1083 .full_width()
1084 .style(ui::ButtonStyle::Tinted(ui::TintColor::Accent))
1085 .on_click(|_, _, cx| cx.open_url(&zed_urls::start_trial_url(cx)))
1086 .into_any_element()
1087 } else {
1088 Button::new("upgrade", "Upgrade to Pro")
1089 .full_width()
1090 .style(ui::ButtonStyle::Tinted(ui::TintColor::Accent))
1091 .on_click(|_, _, cx| cx.open_url(&zed_urls::upgrade_to_zed_pro_url(cx)))
1092 .into_any_element()
1093 };
1094
1095 if !self.is_connected {
1096 return v_flex()
1097 .gap_2()
1098 .child(Label::new("Sign in to have access to Zed's complete agentic experience with hosted models."))
1099 .child(
1100 Button::new("sign_in", "Sign In to use Zed AI")
1101 .icon_color(Color::Muted)
1102 .icon(IconName::Github)
1103 .icon_size(IconSize::Small)
1104 .icon_position(IconPosition::Start)
1105 .full_width()
1106 .on_click({
1107 let callback = self.sign_in_callback.clone();
1108 move |_, window, cx| (callback)(window, cx)
1109 }),
1110 );
1111 }
1112
1113 v_flex().gap_2().w_full().map(|this| {
1114 if self.account_too_young {
1115 this.child(YoungAccountBanner).child(
1116 Button::new("upgrade", "Upgrade to Pro")
1117 .style(ui::ButtonStyle::Tinted(ui::TintColor::Accent))
1118 .full_width()
1119 .on_click(|_, _, cx| cx.open_url(&zed_urls::upgrade_to_zed_pro_url(cx))),
1120 )
1121 } else {
1122 this.text_sm()
1123 .child(subscription_text)
1124 .child(manage_subscription_buttons)
1125 }
1126 })
1127 }
1128}
1129
1130struct ConfigurationView {
1131 state: Entity<State>,
1132 sign_in_callback: Arc<dyn Fn(&mut Window, &mut App) + Send + Sync>,
1133}
1134
1135impl ConfigurationView {
1136 fn new(state: Entity<State>) -> Self {
1137 let sign_in_callback = Arc::new({
1138 let state = state.clone();
1139 move |_window: &mut Window, cx: &mut App| {
1140 state.update(cx, |state, cx| {
1141 state.authenticate(cx).detach_and_log_err(cx);
1142 });
1143 }
1144 });
1145
1146 Self {
1147 state,
1148 sign_in_callback,
1149 }
1150 }
1151}
1152
1153impl Render for ConfigurationView {
1154 fn render(&mut self, _: &mut Window, cx: &mut Context<Self>) -> impl IntoElement {
1155 let state = self.state.read(cx);
1156 let user_store = state.user_store.read(cx);
1157
1158 ZedAiConfiguration {
1159 is_connected: !state.is_signed_out(cx),
1160 plan: user_store.plan(),
1161 subscription_period: user_store.subscription_period(),
1162 eligible_for_trial: user_store.trial_started_at().is_none(),
1163 account_too_young: user_store.account_too_young(),
1164 sign_in_callback: self.sign_in_callback.clone(),
1165 }
1166 }
1167}
1168
1169impl Component for ZedAiConfiguration {
1170 fn name() -> &'static str {
1171 "AI Configuration Content"
1172 }
1173
1174 fn sort_name() -> &'static str {
1175 "AI Configuration Content"
1176 }
1177
1178 fn scope() -> ComponentScope {
1179 ComponentScope::Onboarding
1180 }
1181
1182 fn preview(_window: &mut Window, _cx: &mut App) -> Option<AnyElement> {
1183 fn configuration(
1184 is_connected: bool,
1185 plan: Option<Plan>,
1186 eligible_for_trial: bool,
1187 account_too_young: bool,
1188 ) -> AnyElement {
1189 ZedAiConfiguration {
1190 is_connected,
1191 plan,
1192 subscription_period: plan
1193 .is_some()
1194 .then(|| (Utc::now(), Utc::now() + chrono::Duration::days(7))),
1195 eligible_for_trial,
1196 account_too_young,
1197 sign_in_callback: Arc::new(|_, _| {}),
1198 }
1199 .into_any_element()
1200 }
1201
1202 Some(
1203 v_flex()
1204 .p_4()
1205 .gap_4()
1206 .children(vec![
1207 single_example("Not connected", configuration(false, None, false, false)),
1208 single_example(
1209 "Accept Terms of Service",
1210 configuration(true, None, true, false),
1211 ),
1212 single_example(
1213 "No Plan - Not eligible for trial",
1214 configuration(true, None, false, false),
1215 ),
1216 single_example(
1217 "No Plan - Eligible for trial",
1218 configuration(true, None, true, false),
1219 ),
1220 single_example(
1221 "Free Plan",
1222 configuration(true, Some(Plan::ZedFree), true, false),
1223 ),
1224 single_example(
1225 "Zed Pro Trial Plan",
1226 configuration(true, Some(Plan::ZedProTrial), true, false),
1227 ),
1228 single_example(
1229 "Zed Pro Plan",
1230 configuration(true, Some(Plan::ZedPro), true, false),
1231 ),
1232 ])
1233 .into_any_element(),
1234 )
1235 }
1236}
1237
1238#[cfg(test)]
1239mod tests {
1240 use super::*;
1241 use http_client::http::{HeaderMap, StatusCode};
1242 use language_model::LanguageModelCompletionError;
1243
1244 #[test]
1245 fn test_api_error_conversion_with_upstream_http_error() {
1246 // upstream_http_error with 503 status should become ServerOverloaded
1247 let error_body = r#"{"code":"upstream_http_error","message":"Received an error from the Anthropic API: upstream connect error or disconnect/reset before headers, reset reason: connection timeout","upstream_status":503}"#;
1248
1249 let api_error = ApiError {
1250 status: StatusCode::INTERNAL_SERVER_ERROR,
1251 body: error_body.to_string(),
1252 headers: HeaderMap::new(),
1253 };
1254
1255 let completion_error: LanguageModelCompletionError = api_error.into();
1256
1257 match completion_error {
1258 LanguageModelCompletionError::UpstreamProviderError { message, .. } => {
1259 assert_eq!(
1260 message,
1261 "Received an error from the Anthropic API: upstream connect error or disconnect/reset before headers, reset reason: connection timeout"
1262 );
1263 }
1264 _ => panic!(
1265 "Expected UpstreamProviderError for upstream 503, got: {:?}",
1266 completion_error
1267 ),
1268 }
1269
1270 // upstream_http_error with 500 status should become ApiInternalServerError
1271 let error_body = r#"{"code":"upstream_http_error","message":"Received an error from the OpenAI API: internal server error","upstream_status":500}"#;
1272
1273 let api_error = ApiError {
1274 status: StatusCode::INTERNAL_SERVER_ERROR,
1275 body: error_body.to_string(),
1276 headers: HeaderMap::new(),
1277 };
1278
1279 let completion_error: LanguageModelCompletionError = api_error.into();
1280
1281 match completion_error {
1282 LanguageModelCompletionError::UpstreamProviderError { message, .. } => {
1283 assert_eq!(
1284 message,
1285 "Received an error from the OpenAI API: internal server error"
1286 );
1287 }
1288 _ => panic!(
1289 "Expected UpstreamProviderError for upstream 500, got: {:?}",
1290 completion_error
1291 ),
1292 }
1293
1294 // upstream_http_error with 429 status should become RateLimitExceeded
1295 let error_body = r#"{"code":"upstream_http_error","message":"Received an error from the Google API: rate limit exceeded","upstream_status":429}"#;
1296
1297 let api_error = ApiError {
1298 status: StatusCode::INTERNAL_SERVER_ERROR,
1299 body: error_body.to_string(),
1300 headers: HeaderMap::new(),
1301 };
1302
1303 let completion_error: LanguageModelCompletionError = api_error.into();
1304
1305 match completion_error {
1306 LanguageModelCompletionError::UpstreamProviderError { message, .. } => {
1307 assert_eq!(
1308 message,
1309 "Received an error from the Google API: rate limit exceeded"
1310 );
1311 }
1312 _ => panic!(
1313 "Expected UpstreamProviderError for upstream 429, got: {:?}",
1314 completion_error
1315 ),
1316 }
1317
1318 // Regular 500 error without upstream_http_error should remain ApiInternalServerError for Zed
1319 let error_body = "Regular internal server error";
1320
1321 let api_error = ApiError {
1322 status: StatusCode::INTERNAL_SERVER_ERROR,
1323 body: error_body.to_string(),
1324 headers: HeaderMap::new(),
1325 };
1326
1327 let completion_error: LanguageModelCompletionError = api_error.into();
1328
1329 match completion_error {
1330 LanguageModelCompletionError::ApiInternalServerError { provider, message } => {
1331 assert_eq!(provider, PROVIDER_NAME);
1332 assert_eq!(message, "Regular internal server error");
1333 }
1334 _ => panic!(
1335 "Expected ApiInternalServerError for regular 500, got: {:?}",
1336 completion_error
1337 ),
1338 }
1339
1340 // upstream_http_429 format should be converted to UpstreamProviderError
1341 let error_body = r#"{"code":"upstream_http_429","message":"Upstream Anthropic rate limit exceeded.","retry_after":30.5}"#;
1342
1343 let api_error = ApiError {
1344 status: StatusCode::INTERNAL_SERVER_ERROR,
1345 body: error_body.to_string(),
1346 headers: HeaderMap::new(),
1347 };
1348
1349 let completion_error: LanguageModelCompletionError = api_error.into();
1350
1351 match completion_error {
1352 LanguageModelCompletionError::UpstreamProviderError {
1353 message,
1354 status,
1355 retry_after,
1356 } => {
1357 assert_eq!(message, "Upstream Anthropic rate limit exceeded.");
1358 assert_eq!(status, StatusCode::TOO_MANY_REQUESTS);
1359 assert_eq!(retry_after, Some(Duration::from_secs_f64(30.5)));
1360 }
1361 _ => panic!(
1362 "Expected UpstreamProviderError for upstream_http_429, got: {:?}",
1363 completion_error
1364 ),
1365 }
1366
1367 // Invalid JSON in error body should fall back to regular error handling
1368 let error_body = "Not JSON at all";
1369
1370 let api_error = ApiError {
1371 status: StatusCode::INTERNAL_SERVER_ERROR,
1372 body: error_body.to_string(),
1373 headers: HeaderMap::new(),
1374 };
1375
1376 let completion_error: LanguageModelCompletionError = api_error.into();
1377
1378 match completion_error {
1379 LanguageModelCompletionError::ApiInternalServerError { provider, .. } => {
1380 assert_eq!(provider, PROVIDER_NAME);
1381 }
1382 _ => panic!(
1383 "Expected ApiInternalServerError for invalid JSON, got: {:?}",
1384 completion_error
1385 ),
1386 }
1387 }
1388}