1use ai_onboarding::YoungAccountBanner;
2use anthropic::AnthropicModelMode;
3use anyhow::{Context as _, Result, anyhow};
4use client::{Client, UserStore, zed_urls};
5use cloud_api_types::{OrganizationId, Plan};
6use cloud_llm_client::{
7 CLIENT_SUPPORTS_STATUS_MESSAGES_HEADER_NAME, CLIENT_SUPPORTS_STATUS_STREAM_ENDED_HEADER_NAME,
8 CLIENT_SUPPORTS_X_AI_HEADER_NAME, CompletionBody, CompletionEvent, CompletionRequestStatus,
9 CountTokensBody, CountTokensResponse, ListModelsResponse,
10 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 ANTHROPIC_PROVIDER_ID, ANTHROPIC_PROVIDER_NAME, AuthenticateError, GOOGLE_PROVIDER_ID,
23 GOOGLE_PROVIDER_NAME, IconOrSvg, LanguageModel, LanguageModelCacheConfiguration,
24 LanguageModelCompletionError, LanguageModelCompletionEvent, LanguageModelEffortLevel,
25 LanguageModelId, LanguageModelName, LanguageModelProvider, LanguageModelProviderId,
26 LanguageModelProviderName, LanguageModelProviderState, LanguageModelRequest,
27 LanguageModelToolChoice, LanguageModelToolSchemaFormat, LlmApiToken, NeedsLlmTokenRefresh,
28 OPEN_AI_PROVIDER_ID, OPEN_AI_PROVIDER_NAME, PaymentRequiredError, RateLimiter,
29 RefreshLlmTokenListener, X_AI_PROVIDER_ID, X_AI_PROVIDER_NAME, ZED_CLOUD_PROVIDER_ID,
30 ZED_CLOUD_PROVIDER_NAME,
31};
32use release_channel::AppVersion;
33use schemars::JsonSchema;
34use semver::Version;
35use serde::{Deserialize, Serialize, de::DeserializeOwned};
36use settings::SettingsStore;
37pub use settings::ZedDotDevAvailableModel as AvailableModel;
38pub use settings::ZedDotDevAvailableProvider as AvailableProvider;
39use smol::io::{AsyncReadExt, BufReader};
40use std::collections::VecDeque;
41use std::pin::Pin;
42use std::str::FromStr;
43use std::sync::Arc;
44use std::task::Poll;
45use std::time::Duration;
46use thiserror::Error;
47use ui::{TintColor, prelude::*};
48
49use crate::provider::anthropic::{
50 AnthropicEventMapper, count_anthropic_tokens_with_tiktoken, into_anthropic,
51};
52use crate::provider::google::{GoogleEventMapper, into_google};
53use crate::provider::open_ai::{
54 OpenAiEventMapper, OpenAiResponseEventMapper, count_open_ai_tokens, into_open_ai,
55 into_open_ai_response,
56};
57use crate::provider::x_ai::count_xai_tokens;
58
59const PROVIDER_ID: LanguageModelProviderId = ZED_CLOUD_PROVIDER_ID;
60const PROVIDER_NAME: LanguageModelProviderName = ZED_CLOUD_PROVIDER_NAME;
61
62#[derive(Default, Clone, Debug, PartialEq)]
63pub struct ZedDotDevSettings {
64 pub available_models: Vec<AvailableModel>,
65}
66#[derive(Default, Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]
67#[serde(tag = "type", rename_all = "lowercase")]
68pub enum ModelMode {
69 #[default]
70 Default,
71 Thinking {
72 /// The maximum number of tokens to use for reasoning. Must be lower than the model's `max_output_tokens`.
73 budget_tokens: Option<u32>,
74 },
75}
76
77impl From<ModelMode> for AnthropicModelMode {
78 fn from(value: ModelMode) -> Self {
79 match value {
80 ModelMode::Default => AnthropicModelMode::Default,
81 ModelMode::Thinking { budget_tokens } => AnthropicModelMode::Thinking { budget_tokens },
82 }
83 }
84}
85
86pub struct CloudLanguageModelProvider {
87 client: Arc<Client>,
88 state: Entity<State>,
89 _maintain_client_status: Task<()>,
90}
91
92pub struct State {
93 client: Arc<Client>,
94 llm_api_token: LlmApiToken,
95 user_store: Entity<UserStore>,
96 status: client::Status,
97 models: Vec<Arc<cloud_llm_client::LanguageModel>>,
98 default_model: Option<Arc<cloud_llm_client::LanguageModel>>,
99 default_fast_model: Option<Arc<cloud_llm_client::LanguageModel>>,
100 recommended_models: Vec<Arc<cloud_llm_client::LanguageModel>>,
101 _user_store_subscription: Subscription,
102 _settings_subscription: Subscription,
103 _llm_token_subscription: Subscription,
104}
105
106impl State {
107 fn new(
108 client: Arc<Client>,
109 user_store: Entity<UserStore>,
110 status: client::Status,
111 cx: &mut Context<Self>,
112 ) -> Self {
113 let refresh_llm_token_listener = RefreshLlmTokenListener::global(cx);
114 let llm_api_token = LlmApiToken::global(cx);
115 Self {
116 client: client.clone(),
117 llm_api_token,
118 user_store: user_store.clone(),
119 status,
120 models: Vec::new(),
121 default_model: None,
122 default_fast_model: None,
123 recommended_models: Vec::new(),
124 _user_store_subscription: cx.subscribe(
125 &user_store,
126 move |this, _user_store, event, cx| match event {
127 client::user::Event::PrivateUserInfoUpdated => {
128 let status = *client.status().borrow();
129 if status.is_signed_out() {
130 return;
131 }
132
133 let client = this.client.clone();
134 let llm_api_token = this.llm_api_token.clone();
135 let organization_id = this
136 .user_store
137 .read(cx)
138 .current_organization()
139 .map(|organization| organization.id.clone());
140 cx.spawn(async move |this, cx| {
141 let response =
142 Self::fetch_models(client, llm_api_token, organization_id).await?;
143 this.update(cx, |this, cx| this.update_models(response, cx))
144 })
145 .detach_and_log_err(cx);
146 }
147 _ => {}
148 },
149 ),
150 _settings_subscription: cx.observe_global::<SettingsStore>(|_, cx| {
151 cx.notify();
152 }),
153 _llm_token_subscription: cx.subscribe(
154 &refresh_llm_token_listener,
155 move |this, _listener, _event, cx| {
156 let client = this.client.clone();
157 let llm_api_token = this.llm_api_token.clone();
158 let organization_id = this
159 .user_store
160 .read(cx)
161 .current_organization()
162 .map(|organization| organization.id.clone());
163 cx.spawn(async move |this, cx| {
164 let response =
165 Self::fetch_models(client, llm_api_token, organization_id).await?;
166 this.update(cx, |this, cx| {
167 this.update_models(response, cx);
168 })
169 })
170 .detach_and_log_err(cx);
171 },
172 ),
173 }
174 }
175
176 fn is_signed_out(&self, cx: &App) -> bool {
177 self.user_store.read(cx).current_user().is_none()
178 }
179
180 fn authenticate(&self, cx: &mut Context<Self>) -> Task<Result<()>> {
181 let client = self.client.clone();
182 cx.spawn(async move |state, cx| {
183 client.sign_in_with_optional_connect(true, cx).await?;
184 state.update(cx, |_, cx| cx.notify())
185 })
186 }
187
188 fn update_models(&mut self, response: ListModelsResponse, cx: &mut Context<Self>) {
189 let mut models = Vec::new();
190
191 for model in response.models {
192 models.push(Arc::new(model.clone()));
193 }
194
195 self.default_model = models
196 .iter()
197 .find(|model| {
198 response
199 .default_model
200 .as_ref()
201 .is_some_and(|default_model_id| &model.id == default_model_id)
202 })
203 .cloned();
204 self.default_fast_model = models
205 .iter()
206 .find(|model| {
207 response
208 .default_fast_model
209 .as_ref()
210 .is_some_and(|default_fast_model_id| &model.id == default_fast_model_id)
211 })
212 .cloned();
213 self.recommended_models = response
214 .recommended_models
215 .iter()
216 .filter_map(|id| models.iter().find(|model| &model.id == id))
217 .cloned()
218 .collect();
219 self.models = models;
220 cx.notify();
221 }
222
223 async fn fetch_models(
224 client: Arc<Client>,
225 llm_api_token: LlmApiToken,
226 organization_id: Option<OrganizationId>,
227 ) -> Result<ListModelsResponse> {
228 let http_client = &client.http_client();
229 let token = llm_api_token.acquire(&client, organization_id).await?;
230
231 let request = http_client::Request::builder()
232 .method(Method::GET)
233 .header(CLIENT_SUPPORTS_X_AI_HEADER_NAME, "true")
234 .uri(http_client.build_zed_llm_url("/models", &[])?.as_ref())
235 .header("Authorization", format!("Bearer {token}"))
236 .body(AsyncBody::empty())?;
237 let mut response = http_client
238 .send(request)
239 .await
240 .context("failed to send list models request")?;
241
242 if response.status().is_success() {
243 let mut body = String::new();
244 response.body_mut().read_to_string(&mut body).await?;
245 Ok(serde_json::from_str(&body)?)
246 } else {
247 let mut body = String::new();
248 response.body_mut().read_to_string(&mut body).await?;
249 anyhow::bail!(
250 "error listing models.\nStatus: {:?}\nBody: {body}",
251 response.status(),
252 );
253 }
254 }
255}
256
257impl CloudLanguageModelProvider {
258 pub fn new(user_store: Entity<UserStore>, client: Arc<Client>, cx: &mut App) -> Self {
259 let mut status_rx = client.status();
260 let status = *status_rx.borrow();
261
262 let state = cx.new(|cx| State::new(client.clone(), user_store.clone(), status, cx));
263
264 let state_ref = state.downgrade();
265 let maintain_client_status = cx.spawn(async move |cx| {
266 while let Some(status) = status_rx.next().await {
267 if let Some(this) = state_ref.upgrade() {
268 _ = this.update(cx, |this, cx| {
269 if this.status != status {
270 this.status = status;
271 cx.notify();
272 }
273 });
274 } else {
275 break;
276 }
277 }
278 });
279
280 Self {
281 client,
282 state,
283 _maintain_client_status: maintain_client_status,
284 }
285 }
286
287 fn create_language_model(
288 &self,
289 model: Arc<cloud_llm_client::LanguageModel>,
290 llm_api_token: LlmApiToken,
291 user_store: Entity<UserStore>,
292 ) -> Arc<dyn LanguageModel> {
293 Arc::new(CloudLanguageModel {
294 id: LanguageModelId(SharedString::from(model.id.0.clone())),
295 model,
296 llm_api_token,
297 user_store,
298 client: self.client.clone(),
299 request_limiter: RateLimiter::new(4),
300 })
301 }
302}
303
304impl LanguageModelProviderState for CloudLanguageModelProvider {
305 type ObservableEntity = State;
306
307 fn observable_entity(&self) -> Option<Entity<Self::ObservableEntity>> {
308 Some(self.state.clone())
309 }
310}
311
312impl LanguageModelProvider for CloudLanguageModelProvider {
313 fn id(&self) -> LanguageModelProviderId {
314 PROVIDER_ID
315 }
316
317 fn name(&self) -> LanguageModelProviderName {
318 PROVIDER_NAME
319 }
320
321 fn icon(&self) -> IconOrSvg {
322 IconOrSvg::Icon(IconName::AiZed)
323 }
324
325 fn default_model(&self, cx: &App) -> Option<Arc<dyn LanguageModel>> {
326 let state = self.state.read(cx);
327 let default_model = state.default_model.clone()?;
328 let llm_api_token = state.llm_api_token.clone();
329 let user_store = state.user_store.clone();
330 Some(self.create_language_model(default_model, llm_api_token, user_store))
331 }
332
333 fn default_fast_model(&self, cx: &App) -> Option<Arc<dyn LanguageModel>> {
334 let state = self.state.read(cx);
335 let default_fast_model = state.default_fast_model.clone()?;
336 let llm_api_token = state.llm_api_token.clone();
337 let user_store = state.user_store.clone();
338 Some(self.create_language_model(default_fast_model, llm_api_token, user_store))
339 }
340
341 fn recommended_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {
342 let state = self.state.read(cx);
343 let llm_api_token = state.llm_api_token.clone();
344 let user_store = state.user_store.clone();
345 state
346 .recommended_models
347 .iter()
348 .cloned()
349 .map(|model| {
350 self.create_language_model(model, llm_api_token.clone(), user_store.clone())
351 })
352 .collect()
353 }
354
355 fn provided_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {
356 let state = self.state.read(cx);
357 let llm_api_token = state.llm_api_token.clone();
358 let user_store = state.user_store.clone();
359 state
360 .models
361 .iter()
362 .cloned()
363 .map(|model| {
364 self.create_language_model(model, llm_api_token.clone(), user_store.clone())
365 })
366 .collect()
367 }
368
369 fn is_authenticated(&self, cx: &App) -> bool {
370 let state = self.state.read(cx);
371 !state.is_signed_out(cx)
372 }
373
374 fn authenticate(&self, _cx: &mut App) -> Task<Result<(), AuthenticateError>> {
375 Task::ready(Ok(()))
376 }
377
378 fn configuration_view(
379 &self,
380 _target_agent: language_model::ConfigurationViewTargetAgent,
381 _: &mut Window,
382 cx: &mut App,
383 ) -> AnyView {
384 cx.new(|_| ConfigurationView::new(self.state.clone()))
385 .into()
386 }
387
388 fn reset_credentials(&self, _cx: &mut App) -> Task<Result<()>> {
389 Task::ready(Ok(()))
390 }
391}
392
393pub struct CloudLanguageModel {
394 id: LanguageModelId,
395 model: Arc<cloud_llm_client::LanguageModel>,
396 llm_api_token: LlmApiToken,
397 user_store: Entity<UserStore>,
398 client: Arc<Client>,
399 request_limiter: RateLimiter,
400}
401
402struct PerformLlmCompletionResponse {
403 response: Response<AsyncBody>,
404 includes_status_messages: bool,
405}
406
407impl CloudLanguageModel {
408 async fn perform_llm_completion(
409 client: Arc<Client>,
410 llm_api_token: LlmApiToken,
411 organization_id: Option<OrganizationId>,
412 app_version: Option<Version>,
413 body: CompletionBody,
414 ) -> Result<PerformLlmCompletionResponse> {
415 let http_client = &client.http_client();
416
417 let mut token = llm_api_token
418 .acquire(&client, organization_id.clone())
419 .await?;
420 let mut refreshed_token = false;
421
422 loop {
423 let request = http_client::Request::builder()
424 .method(Method::POST)
425 .uri(http_client.build_zed_llm_url("/completions", &[])?.as_ref())
426 .when_some(app_version.as_ref(), |builder, app_version| {
427 builder.header(ZED_VERSION_HEADER_NAME, app_version.to_string())
428 })
429 .header("Content-Type", "application/json")
430 .header("Authorization", format!("Bearer {token}"))
431 .header(CLIENT_SUPPORTS_STATUS_MESSAGES_HEADER_NAME, "true")
432 .header(CLIENT_SUPPORTS_STATUS_STREAM_ENDED_HEADER_NAME, "true")
433 .body(serde_json::to_string(&body)?.into())?;
434
435 let mut response = http_client.send(request).await?;
436 let status = response.status();
437 if status.is_success() {
438 let includes_status_messages = response
439 .headers()
440 .get(SERVER_SUPPORTS_STATUS_MESSAGES_HEADER_NAME)
441 .is_some();
442
443 return Ok(PerformLlmCompletionResponse {
444 response,
445 includes_status_messages,
446 });
447 }
448
449 if !refreshed_token && response.needs_llm_token_refresh() {
450 token = llm_api_token
451 .refresh(&client, organization_id.clone())
452 .await?;
453 refreshed_token = true;
454 continue;
455 }
456
457 if status == StatusCode::PAYMENT_REQUIRED {
458 return Err(anyhow!(PaymentRequiredError));
459 }
460
461 let mut body = String::new();
462 let headers = response.headers().clone();
463 response.body_mut().read_to_string(&mut body).await?;
464 return Err(anyhow!(ApiError {
465 status,
466 body,
467 headers
468 }));
469 }
470 }
471}
472
473#[derive(Debug, Error)]
474#[error("cloud language model request failed with status {status}: {body}")]
475struct ApiError {
476 status: StatusCode,
477 body: String,
478 headers: HeaderMap<HeaderValue>,
479}
480
481/// Represents error responses from Zed's cloud API.
482///
483/// Example JSON for an upstream HTTP error:
484/// ```json
485/// {
486/// "code": "upstream_http_error",
487/// "message": "Received an error from the Anthropic API: upstream connect error or disconnect/reset before headers, reset reason: connection timeout",
488/// "upstream_status": 503
489/// }
490/// ```
491#[derive(Debug, serde::Deserialize)]
492struct CloudApiError {
493 code: String,
494 message: String,
495 #[serde(default)]
496 #[serde(deserialize_with = "deserialize_optional_status_code")]
497 upstream_status: Option<StatusCode>,
498 #[serde(default)]
499 retry_after: Option<f64>,
500}
501
502fn deserialize_optional_status_code<'de, D>(deserializer: D) -> Result<Option<StatusCode>, D::Error>
503where
504 D: serde::Deserializer<'de>,
505{
506 let opt: Option<u16> = Option::deserialize(deserializer)?;
507 Ok(opt.and_then(|code| StatusCode::from_u16(code).ok()))
508}
509
510impl From<ApiError> for LanguageModelCompletionError {
511 fn from(error: ApiError) -> Self {
512 if let Ok(cloud_error) = serde_json::from_str::<CloudApiError>(&error.body) {
513 if cloud_error.code.starts_with("upstream_http_") {
514 let status = if let Some(status) = cloud_error.upstream_status {
515 status
516 } else if cloud_error.code.ends_with("_error") {
517 error.status
518 } else {
519 // If there's a status code in the code string (e.g. "upstream_http_429")
520 // then use that; otherwise, see if the JSON contains a status code.
521 cloud_error
522 .code
523 .strip_prefix("upstream_http_")
524 .and_then(|code_str| code_str.parse::<u16>().ok())
525 .and_then(|code| StatusCode::from_u16(code).ok())
526 .unwrap_or(error.status)
527 };
528
529 return LanguageModelCompletionError::UpstreamProviderError {
530 message: cloud_error.message,
531 status,
532 retry_after: cloud_error.retry_after.map(Duration::from_secs_f64),
533 };
534 }
535
536 return LanguageModelCompletionError::from_http_status(
537 PROVIDER_NAME,
538 error.status,
539 cloud_error.message,
540 None,
541 );
542 }
543
544 let retry_after = None;
545 LanguageModelCompletionError::from_http_status(
546 PROVIDER_NAME,
547 error.status,
548 error.body,
549 retry_after,
550 )
551 }
552}
553
554impl LanguageModel for CloudLanguageModel {
555 fn id(&self) -> LanguageModelId {
556 self.id.clone()
557 }
558
559 fn name(&self) -> LanguageModelName {
560 LanguageModelName::from(self.model.display_name.clone())
561 }
562
563 fn provider_id(&self) -> LanguageModelProviderId {
564 PROVIDER_ID
565 }
566
567 fn provider_name(&self) -> LanguageModelProviderName {
568 PROVIDER_NAME
569 }
570
571 fn upstream_provider_id(&self) -> LanguageModelProviderId {
572 use cloud_llm_client::LanguageModelProvider::*;
573 match self.model.provider {
574 Anthropic => ANTHROPIC_PROVIDER_ID,
575 OpenAi => OPEN_AI_PROVIDER_ID,
576 Google => GOOGLE_PROVIDER_ID,
577 XAi => X_AI_PROVIDER_ID,
578 }
579 }
580
581 fn upstream_provider_name(&self) -> LanguageModelProviderName {
582 use cloud_llm_client::LanguageModelProvider::*;
583 match self.model.provider {
584 Anthropic => ANTHROPIC_PROVIDER_NAME,
585 OpenAi => OPEN_AI_PROVIDER_NAME,
586 Google => GOOGLE_PROVIDER_NAME,
587 XAi => X_AI_PROVIDER_NAME,
588 }
589 }
590
591 fn is_latest(&self) -> bool {
592 self.model.is_latest
593 }
594
595 fn supports_tools(&self) -> bool {
596 self.model.supports_tools
597 }
598
599 fn supports_images(&self) -> bool {
600 self.model.supports_images
601 }
602
603 fn supports_thinking(&self) -> bool {
604 self.model.supports_thinking
605 }
606
607 fn supports_fast_mode(&self) -> bool {
608 self.model.supports_fast_mode
609 }
610
611 fn supported_effort_levels(&self) -> Vec<LanguageModelEffortLevel> {
612 self.model
613 .supported_effort_levels
614 .iter()
615 .map(|effort_level| LanguageModelEffortLevel {
616 name: effort_level.name.clone().into(),
617 value: effort_level.value.clone().into(),
618 is_default: effort_level.is_default.unwrap_or(false),
619 })
620 .collect()
621 }
622
623 fn supports_streaming_tools(&self) -> bool {
624 self.model.supports_streaming_tools
625 }
626
627 fn supports_tool_choice(&self, choice: LanguageModelToolChoice) -> bool {
628 match choice {
629 LanguageModelToolChoice::Auto
630 | LanguageModelToolChoice::Any
631 | LanguageModelToolChoice::None => true,
632 }
633 }
634
635 fn supports_split_token_display(&self) -> bool {
636 use cloud_llm_client::LanguageModelProvider::*;
637 matches!(self.model.provider, OpenAi | XAi)
638 }
639
640 fn telemetry_id(&self) -> String {
641 format!("zed.dev/{}", self.model.id)
642 }
643
644 fn tool_input_format(&self) -> LanguageModelToolSchemaFormat {
645 match self.model.provider {
646 cloud_llm_client::LanguageModelProvider::Anthropic
647 | cloud_llm_client::LanguageModelProvider::OpenAi => {
648 LanguageModelToolSchemaFormat::JsonSchema
649 }
650 cloud_llm_client::LanguageModelProvider::Google
651 | cloud_llm_client::LanguageModelProvider::XAi => {
652 LanguageModelToolSchemaFormat::JsonSchemaSubset
653 }
654 }
655 }
656
657 fn max_token_count(&self) -> u64 {
658 self.model.max_token_count as u64
659 }
660
661 fn max_output_tokens(&self) -> Option<u64> {
662 Some(self.model.max_output_tokens as u64)
663 }
664
665 fn cache_configuration(&self) -> Option<LanguageModelCacheConfiguration> {
666 match &self.model.provider {
667 cloud_llm_client::LanguageModelProvider::Anthropic => {
668 Some(LanguageModelCacheConfiguration {
669 min_total_token: 2_048,
670 should_speculate: true,
671 max_cache_anchors: 4,
672 })
673 }
674 cloud_llm_client::LanguageModelProvider::OpenAi
675 | cloud_llm_client::LanguageModelProvider::XAi
676 | cloud_llm_client::LanguageModelProvider::Google => None,
677 }
678 }
679
680 fn count_tokens(
681 &self,
682 request: LanguageModelRequest,
683 cx: &App,
684 ) -> BoxFuture<'static, Result<u64>> {
685 match self.model.provider {
686 cloud_llm_client::LanguageModelProvider::Anthropic => cx
687 .background_spawn(async move { count_anthropic_tokens_with_tiktoken(request) })
688 .boxed(),
689 cloud_llm_client::LanguageModelProvider::OpenAi => {
690 let model = match open_ai::Model::from_id(&self.model.id.0) {
691 Ok(model) => model,
692 Err(err) => return async move { Err(anyhow!(err)) }.boxed(),
693 };
694 count_open_ai_tokens(request, model, cx)
695 }
696 cloud_llm_client::LanguageModelProvider::XAi => {
697 let model = match x_ai::Model::from_id(&self.model.id.0) {
698 Ok(model) => model,
699 Err(err) => return async move { Err(anyhow!(err)) }.boxed(),
700 };
701 count_xai_tokens(request, model, cx)
702 }
703 cloud_llm_client::LanguageModelProvider::Google => {
704 let client = self.client.clone();
705 let llm_api_token = self.llm_api_token.clone();
706 let organization_id = self
707 .user_store
708 .read(cx)
709 .current_organization()
710 .map(|organization| organization.id.clone());
711 let model_id = self.model.id.to_string();
712 let generate_content_request =
713 into_google(request, model_id.clone(), GoogleModelMode::Default);
714 async move {
715 let http_client = &client.http_client();
716 let token = llm_api_token.acquire(&client, organization_id).await?;
717
718 let request_body = CountTokensBody {
719 provider: cloud_llm_client::LanguageModelProvider::Google,
720 model: model_id,
721 provider_request: serde_json::to_value(&google_ai::CountTokensRequest {
722 generate_content_request,
723 })?,
724 };
725 let request = http_client::Request::builder()
726 .method(Method::POST)
727 .uri(
728 http_client
729 .build_zed_llm_url("/count_tokens", &[])?
730 .as_ref(),
731 )
732 .header("Content-Type", "application/json")
733 .header("Authorization", format!("Bearer {token}"))
734 .body(serde_json::to_string(&request_body)?.into())?;
735 let mut response = http_client.send(request).await?;
736 let status = response.status();
737 let headers = response.headers().clone();
738 let mut response_body = String::new();
739 response
740 .body_mut()
741 .read_to_string(&mut response_body)
742 .await?;
743
744 if status.is_success() {
745 let response_body: CountTokensResponse =
746 serde_json::from_str(&response_body)?;
747
748 Ok(response_body.tokens as u64)
749 } else {
750 Err(anyhow!(ApiError {
751 status,
752 body: response_body,
753 headers
754 }))
755 }
756 }
757 .boxed()
758 }
759 }
760 }
761
762 fn stream_completion(
763 &self,
764 request: LanguageModelRequest,
765 cx: &AsyncApp,
766 ) -> BoxFuture<
767 'static,
768 Result<
769 BoxStream<'static, Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>,
770 LanguageModelCompletionError,
771 >,
772 > {
773 let thread_id = request.thread_id.clone();
774 let prompt_id = request.prompt_id.clone();
775 let app_version = Some(cx.update(|cx| AppVersion::global(cx)));
776 let user_store = self.user_store.clone();
777 let organization_id = cx.update(|cx| {
778 user_store
779 .read(cx)
780 .current_organization()
781 .map(|organization| organization.id.clone())
782 });
783 let thinking_allowed = request.thinking_allowed;
784 let enable_thinking = thinking_allowed && self.model.supports_thinking;
785 let provider_name = provider_name(&self.model.provider);
786 match self.model.provider {
787 cloud_llm_client::LanguageModelProvider::Anthropic => {
788 let effort = request
789 .thinking_effort
790 .as_ref()
791 .and_then(|effort| anthropic::Effort::from_str(effort).ok());
792
793 let mut request = into_anthropic(
794 request,
795 self.model.id.to_string(),
796 1.0,
797 self.model.max_output_tokens as u64,
798 if enable_thinking {
799 AnthropicModelMode::Thinking {
800 budget_tokens: Some(4_096),
801 }
802 } else {
803 AnthropicModelMode::Default
804 },
805 );
806
807 if enable_thinking && effort.is_some() {
808 request.thinking = Some(anthropic::Thinking::Adaptive);
809 request.output_config = Some(anthropic::OutputConfig { effort });
810 }
811
812 let client = self.client.clone();
813 let llm_api_token = self.llm_api_token.clone();
814 let organization_id = organization_id.clone();
815 let future = self.request_limiter.stream(async move {
816 let PerformLlmCompletionResponse {
817 response,
818 includes_status_messages,
819 } = Self::perform_llm_completion(
820 client.clone(),
821 llm_api_token,
822 organization_id,
823 app_version,
824 CompletionBody {
825 thread_id,
826 prompt_id,
827 provider: cloud_llm_client::LanguageModelProvider::Anthropic,
828 model: request.model.clone(),
829 provider_request: serde_json::to_value(&request)
830 .map_err(|e| anyhow!(e))?,
831 },
832 )
833 .await
834 .map_err(|err| match err.downcast::<ApiError>() {
835 Ok(api_err) => anyhow!(LanguageModelCompletionError::from(api_err)),
836 Err(err) => anyhow!(err),
837 })?;
838
839 let mut mapper = AnthropicEventMapper::new();
840 Ok(map_cloud_completion_events(
841 Box::pin(response_lines(response, includes_status_messages)),
842 &provider_name,
843 move |event| mapper.map_event(event),
844 ))
845 });
846 async move { Ok(future.await?.boxed()) }.boxed()
847 }
848 cloud_llm_client::LanguageModelProvider::OpenAi => {
849 let client = self.client.clone();
850 let llm_api_token = self.llm_api_token.clone();
851 let organization_id = organization_id.clone();
852 let effort = request
853 .thinking_effort
854 .as_ref()
855 .and_then(|effort| open_ai::ReasoningEffort::from_str(effort).ok());
856
857 let mut request = into_open_ai_response(
858 request,
859 &self.model.id.0,
860 self.model.supports_parallel_tool_calls,
861 true,
862 None,
863 None,
864 );
865
866 if enable_thinking && let Some(effort) = effort {
867 request.reasoning = Some(open_ai::responses::ReasoningConfig {
868 effort,
869 summary: Some(open_ai::responses::ReasoningSummaryMode::Auto),
870 });
871 }
872
873 let future = self.request_limiter.stream(async move {
874 let PerformLlmCompletionResponse {
875 response,
876 includes_status_messages,
877 } = Self::perform_llm_completion(
878 client.clone(),
879 llm_api_token,
880 organization_id,
881 app_version,
882 CompletionBody {
883 thread_id,
884 prompt_id,
885 provider: cloud_llm_client::LanguageModelProvider::OpenAi,
886 model: request.model.clone(),
887 provider_request: serde_json::to_value(&request)
888 .map_err(|e| anyhow!(e))?,
889 },
890 )
891 .await?;
892
893 let mut mapper = OpenAiResponseEventMapper::new();
894 Ok(map_cloud_completion_events(
895 Box::pin(response_lines(response, includes_status_messages)),
896 &provider_name,
897 move |event| mapper.map_event(event),
898 ))
899 });
900 async move { Ok(future.await?.boxed()) }.boxed()
901 }
902 cloud_llm_client::LanguageModelProvider::XAi => {
903 let client = self.client.clone();
904 let request = into_open_ai(
905 request,
906 &self.model.id.0,
907 self.model.supports_parallel_tool_calls,
908 false,
909 None,
910 None,
911 );
912 let llm_api_token = self.llm_api_token.clone();
913 let organization_id = organization_id.clone();
914 let future = self.request_limiter.stream(async move {
915 let PerformLlmCompletionResponse {
916 response,
917 includes_status_messages,
918 } = Self::perform_llm_completion(
919 client.clone(),
920 llm_api_token,
921 organization_id,
922 app_version,
923 CompletionBody {
924 thread_id,
925 prompt_id,
926 provider: cloud_llm_client::LanguageModelProvider::XAi,
927 model: request.model.clone(),
928 provider_request: serde_json::to_value(&request)
929 .map_err(|e| anyhow!(e))?,
930 },
931 )
932 .await?;
933
934 let mut mapper = OpenAiEventMapper::new();
935 Ok(map_cloud_completion_events(
936 Box::pin(response_lines(response, includes_status_messages)),
937 &provider_name,
938 move |event| mapper.map_event(event),
939 ))
940 });
941 async move { Ok(future.await?.boxed()) }.boxed()
942 }
943 cloud_llm_client::LanguageModelProvider::Google => {
944 let client = self.client.clone();
945 let request =
946 into_google(request, self.model.id.to_string(), GoogleModelMode::Default);
947 let llm_api_token = self.llm_api_token.clone();
948 let future = self.request_limiter.stream(async move {
949 let PerformLlmCompletionResponse {
950 response,
951 includes_status_messages,
952 } = Self::perform_llm_completion(
953 client.clone(),
954 llm_api_token,
955 organization_id,
956 app_version,
957 CompletionBody {
958 thread_id,
959 prompt_id,
960 provider: cloud_llm_client::LanguageModelProvider::Google,
961 model: request.model.model_id.clone(),
962 provider_request: serde_json::to_value(&request)
963 .map_err(|e| anyhow!(e))?,
964 },
965 )
966 .await?;
967
968 let mut mapper = GoogleEventMapper::new();
969 Ok(map_cloud_completion_events(
970 Box::pin(response_lines(response, includes_status_messages)),
971 &provider_name,
972 move |event| mapper.map_event(event),
973 ))
974 });
975 async move { Ok(future.await?.boxed()) }.boxed()
976 }
977 }
978 }
979}
980
981fn map_cloud_completion_events<T, F>(
982 stream: Pin<Box<dyn Stream<Item = Result<CompletionEvent<T>>> + Send>>,
983 provider: &LanguageModelProviderName,
984 mut map_callback: F,
985) -> BoxStream<'static, Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
986where
987 T: DeserializeOwned + 'static,
988 F: FnMut(T) -> Vec<Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
989 + Send
990 + 'static,
991{
992 let provider = provider.clone();
993 let mut stream = stream.fuse();
994
995 let mut saw_stream_ended = false;
996
997 let mut done = false;
998 let mut pending = VecDeque::new();
999
1000 stream::poll_fn(move |cx| {
1001 loop {
1002 if let Some(item) = pending.pop_front() {
1003 return Poll::Ready(Some(item));
1004 }
1005
1006 if done {
1007 return Poll::Ready(None);
1008 }
1009
1010 match stream.poll_next_unpin(cx) {
1011 Poll::Ready(Some(event)) => {
1012 let items = match event {
1013 Err(error) => {
1014 vec![Err(LanguageModelCompletionError::from(error))]
1015 }
1016 Ok(CompletionEvent::Status(CompletionRequestStatus::StreamEnded)) => {
1017 saw_stream_ended = true;
1018 vec![]
1019 }
1020 Ok(CompletionEvent::Status(status)) => {
1021 LanguageModelCompletionEvent::from_completion_request_status(
1022 status,
1023 provider.clone(),
1024 )
1025 .transpose()
1026 .map(|event| vec![event])
1027 .unwrap_or_default()
1028 }
1029 Ok(CompletionEvent::Event(event)) => map_callback(event),
1030 };
1031 pending.extend(items);
1032 }
1033 Poll::Ready(None) => {
1034 done = true;
1035
1036 if !saw_stream_ended {
1037 return Poll::Ready(Some(Err(
1038 LanguageModelCompletionError::StreamEndedUnexpectedly {
1039 provider: provider.clone(),
1040 },
1041 )));
1042 }
1043 }
1044 Poll::Pending => return Poll::Pending,
1045 }
1046 }
1047 })
1048 .boxed()
1049}
1050
1051fn provider_name(provider: &cloud_llm_client::LanguageModelProvider) -> LanguageModelProviderName {
1052 match provider {
1053 cloud_llm_client::LanguageModelProvider::Anthropic => ANTHROPIC_PROVIDER_NAME,
1054 cloud_llm_client::LanguageModelProvider::OpenAi => OPEN_AI_PROVIDER_NAME,
1055 cloud_llm_client::LanguageModelProvider::Google => GOOGLE_PROVIDER_NAME,
1056 cloud_llm_client::LanguageModelProvider::XAi => X_AI_PROVIDER_NAME,
1057 }
1058}
1059
1060fn response_lines<T: DeserializeOwned>(
1061 response: Response<AsyncBody>,
1062 includes_status_messages: bool,
1063) -> impl Stream<Item = Result<CompletionEvent<T>>> {
1064 futures::stream::try_unfold(
1065 (String::new(), BufReader::new(response.into_body())),
1066 move |(mut line, mut body)| async move {
1067 match body.read_line(&mut line).await {
1068 Ok(0) => Ok(None),
1069 Ok(_) => {
1070 let event = if includes_status_messages {
1071 serde_json::from_str::<CompletionEvent<T>>(&line)?
1072 } else {
1073 CompletionEvent::Event(serde_json::from_str::<T>(&line)?)
1074 };
1075
1076 line.clear();
1077 Ok(Some((event, (line, body))))
1078 }
1079 Err(e) => Err(e.into()),
1080 }
1081 },
1082 )
1083}
1084
1085#[derive(IntoElement, RegisterComponent)]
1086struct ZedAiConfiguration {
1087 is_connected: bool,
1088 plan: Option<Plan>,
1089 eligible_for_trial: bool,
1090 account_too_young: bool,
1091 sign_in_callback: Arc<dyn Fn(&mut Window, &mut App) + Send + Sync>,
1092}
1093
1094impl RenderOnce for ZedAiConfiguration {
1095 fn render(self, _window: &mut Window, _cx: &mut App) -> impl IntoElement {
1096 let (subscription_text, has_paid_plan) = match self.plan {
1097 Some(Plan::ZedPro) => (
1098 "You have access to Zed's hosted models through your Pro subscription.",
1099 true,
1100 ),
1101 Some(Plan::ZedProTrial) => (
1102 "You have access to Zed's hosted models through your Pro trial.",
1103 false,
1104 ),
1105 Some(Plan::ZedStudent) => (
1106 "You have access to Zed's hosted models through your Student subscription.",
1107 true,
1108 ),
1109 Some(Plan::ZedBusiness) => (
1110 "You have access to Zed's hosted models through your Organization.",
1111 true,
1112 ),
1113 Some(Plan::ZedFree) | None => (
1114 if self.eligible_for_trial {
1115 "Subscribe for access to Zed's hosted models. Start with a 14 day free trial."
1116 } else {
1117 "Subscribe for access to Zed's hosted models."
1118 },
1119 false,
1120 ),
1121 };
1122
1123 let manage_subscription_buttons = if has_paid_plan {
1124 Button::new("manage_settings", "Manage Subscription")
1125 .full_width()
1126 .label_size(LabelSize::Small)
1127 .style(ButtonStyle::Tinted(TintColor::Accent))
1128 .on_click(|_, _, cx| cx.open_url(&zed_urls::account_url(cx)))
1129 .into_any_element()
1130 } else if self.plan.is_none() || self.eligible_for_trial {
1131 Button::new("start_trial", "Start 14-day Free Pro Trial")
1132 .full_width()
1133 .style(ui::ButtonStyle::Tinted(ui::TintColor::Accent))
1134 .on_click(|_, _, cx| cx.open_url(&zed_urls::start_trial_url(cx)))
1135 .into_any_element()
1136 } else {
1137 Button::new("upgrade", "Upgrade to Pro")
1138 .full_width()
1139 .style(ui::ButtonStyle::Tinted(ui::TintColor::Accent))
1140 .on_click(|_, _, cx| cx.open_url(&zed_urls::upgrade_to_zed_pro_url(cx)))
1141 .into_any_element()
1142 };
1143
1144 if !self.is_connected {
1145 return v_flex()
1146 .gap_2()
1147 .child(Label::new("Sign in to have access to Zed's complete agentic experience with hosted models."))
1148 .child(
1149 Button::new("sign_in", "Sign In to use Zed AI")
1150 .start_icon(Icon::new(IconName::Github).size(IconSize::Small).color(Color::Muted))
1151 .full_width()
1152 .on_click({
1153 let callback = self.sign_in_callback.clone();
1154 move |_, window, cx| (callback)(window, cx)
1155 }),
1156 );
1157 }
1158
1159 v_flex().gap_2().w_full().map(|this| {
1160 if self.account_too_young {
1161 this.child(YoungAccountBanner).child(
1162 Button::new("upgrade", "Upgrade to Pro")
1163 .style(ui::ButtonStyle::Tinted(ui::TintColor::Accent))
1164 .full_width()
1165 .on_click(|_, _, cx| cx.open_url(&zed_urls::upgrade_to_zed_pro_url(cx))),
1166 )
1167 } else {
1168 this.text_sm()
1169 .child(subscription_text)
1170 .child(manage_subscription_buttons)
1171 }
1172 })
1173 }
1174}
1175
1176struct ConfigurationView {
1177 state: Entity<State>,
1178 sign_in_callback: Arc<dyn Fn(&mut Window, &mut App) + Send + Sync>,
1179}
1180
1181impl ConfigurationView {
1182 fn new(state: Entity<State>) -> Self {
1183 let sign_in_callback = Arc::new({
1184 let state = state.clone();
1185 move |_window: &mut Window, cx: &mut App| {
1186 state.update(cx, |state, cx| {
1187 state.authenticate(cx).detach_and_log_err(cx);
1188 });
1189 }
1190 });
1191
1192 Self {
1193 state,
1194 sign_in_callback,
1195 }
1196 }
1197}
1198
1199impl Render for ConfigurationView {
1200 fn render(&mut self, _: &mut Window, cx: &mut Context<Self>) -> impl IntoElement {
1201 let state = self.state.read(cx);
1202 let user_store = state.user_store.read(cx);
1203
1204 ZedAiConfiguration {
1205 is_connected: !state.is_signed_out(cx),
1206 plan: user_store.plan(),
1207 eligible_for_trial: user_store.trial_started_at().is_none(),
1208 account_too_young: user_store.account_too_young(),
1209 sign_in_callback: self.sign_in_callback.clone(),
1210 }
1211 }
1212}
1213
1214impl Component for ZedAiConfiguration {
1215 fn name() -> &'static str {
1216 "AI Configuration Content"
1217 }
1218
1219 fn sort_name() -> &'static str {
1220 "AI Configuration Content"
1221 }
1222
1223 fn scope() -> ComponentScope {
1224 ComponentScope::Onboarding
1225 }
1226
1227 fn preview(_window: &mut Window, _cx: &mut App) -> Option<AnyElement> {
1228 fn configuration(
1229 is_connected: bool,
1230 plan: Option<Plan>,
1231 eligible_for_trial: bool,
1232 account_too_young: bool,
1233 ) -> AnyElement {
1234 ZedAiConfiguration {
1235 is_connected,
1236 plan,
1237 eligible_for_trial,
1238 account_too_young,
1239 sign_in_callback: Arc::new(|_, _| {}),
1240 }
1241 .into_any_element()
1242 }
1243
1244 Some(
1245 v_flex()
1246 .p_4()
1247 .gap_4()
1248 .children(vec![
1249 single_example("Not connected", configuration(false, None, false, false)),
1250 single_example(
1251 "Accept Terms of Service",
1252 configuration(true, None, true, false),
1253 ),
1254 single_example(
1255 "No Plan - Not eligible for trial",
1256 configuration(true, None, false, false),
1257 ),
1258 single_example(
1259 "No Plan - Eligible for trial",
1260 configuration(true, None, true, false),
1261 ),
1262 single_example(
1263 "Free Plan",
1264 configuration(true, Some(Plan::ZedFree), true, false),
1265 ),
1266 single_example(
1267 "Zed Pro Trial Plan",
1268 configuration(true, Some(Plan::ZedProTrial), true, false),
1269 ),
1270 single_example(
1271 "Zed Pro Plan",
1272 configuration(true, Some(Plan::ZedPro), true, false),
1273 ),
1274 ])
1275 .into_any_element(),
1276 )
1277 }
1278}
1279
1280#[cfg(test)]
1281mod tests {
1282 use super::*;
1283 use http_client::http::{HeaderMap, StatusCode};
1284 use language_model::LanguageModelCompletionError;
1285
1286 #[test]
1287 fn test_api_error_conversion_with_upstream_http_error() {
1288 // upstream_http_error with 503 status should become ServerOverloaded
1289 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}"#;
1290
1291 let api_error = ApiError {
1292 status: StatusCode::INTERNAL_SERVER_ERROR,
1293 body: error_body.to_string(),
1294 headers: HeaderMap::new(),
1295 };
1296
1297 let completion_error: LanguageModelCompletionError = api_error.into();
1298
1299 match completion_error {
1300 LanguageModelCompletionError::UpstreamProviderError { message, .. } => {
1301 assert_eq!(
1302 message,
1303 "Received an error from the Anthropic API: upstream connect error or disconnect/reset before headers, reset reason: connection timeout"
1304 );
1305 }
1306 _ => panic!(
1307 "Expected UpstreamProviderError for upstream 503, got: {:?}",
1308 completion_error
1309 ),
1310 }
1311
1312 // upstream_http_error with 500 status should become ApiInternalServerError
1313 let error_body = r#"{"code":"upstream_http_error","message":"Received an error from the OpenAI API: internal server error","upstream_status":500}"#;
1314
1315 let api_error = ApiError {
1316 status: StatusCode::INTERNAL_SERVER_ERROR,
1317 body: error_body.to_string(),
1318 headers: HeaderMap::new(),
1319 };
1320
1321 let completion_error: LanguageModelCompletionError = api_error.into();
1322
1323 match completion_error {
1324 LanguageModelCompletionError::UpstreamProviderError { message, .. } => {
1325 assert_eq!(
1326 message,
1327 "Received an error from the OpenAI API: internal server error"
1328 );
1329 }
1330 _ => panic!(
1331 "Expected UpstreamProviderError for upstream 500, got: {:?}",
1332 completion_error
1333 ),
1334 }
1335
1336 // upstream_http_error with 429 status should become RateLimitExceeded
1337 let error_body = r#"{"code":"upstream_http_error","message":"Received an error from the Google API: rate limit exceeded","upstream_status":429}"#;
1338
1339 let api_error = ApiError {
1340 status: StatusCode::INTERNAL_SERVER_ERROR,
1341 body: error_body.to_string(),
1342 headers: HeaderMap::new(),
1343 };
1344
1345 let completion_error: LanguageModelCompletionError = api_error.into();
1346
1347 match completion_error {
1348 LanguageModelCompletionError::UpstreamProviderError { message, .. } => {
1349 assert_eq!(
1350 message,
1351 "Received an error from the Google API: rate limit exceeded"
1352 );
1353 }
1354 _ => panic!(
1355 "Expected UpstreamProviderError for upstream 429, got: {:?}",
1356 completion_error
1357 ),
1358 }
1359
1360 // Regular 500 error without upstream_http_error should remain ApiInternalServerError for Zed
1361 let error_body = "Regular internal server error";
1362
1363 let api_error = ApiError {
1364 status: StatusCode::INTERNAL_SERVER_ERROR,
1365 body: error_body.to_string(),
1366 headers: HeaderMap::new(),
1367 };
1368
1369 let completion_error: LanguageModelCompletionError = api_error.into();
1370
1371 match completion_error {
1372 LanguageModelCompletionError::ApiInternalServerError { provider, message } => {
1373 assert_eq!(provider, PROVIDER_NAME);
1374 assert_eq!(message, "Regular internal server error");
1375 }
1376 _ => panic!(
1377 "Expected ApiInternalServerError for regular 500, got: {:?}",
1378 completion_error
1379 ),
1380 }
1381
1382 // upstream_http_429 format should be converted to UpstreamProviderError
1383 let error_body = r#"{"code":"upstream_http_429","message":"Upstream Anthropic rate limit exceeded.","retry_after":30.5}"#;
1384
1385 let api_error = ApiError {
1386 status: StatusCode::INTERNAL_SERVER_ERROR,
1387 body: error_body.to_string(),
1388 headers: HeaderMap::new(),
1389 };
1390
1391 let completion_error: LanguageModelCompletionError = api_error.into();
1392
1393 match completion_error {
1394 LanguageModelCompletionError::UpstreamProviderError {
1395 message,
1396 status,
1397 retry_after,
1398 } => {
1399 assert_eq!(message, "Upstream Anthropic rate limit exceeded.");
1400 assert_eq!(status, StatusCode::TOO_MANY_REQUESTS);
1401 assert_eq!(retry_after, Some(Duration::from_secs_f64(30.5)));
1402 }
1403 _ => panic!(
1404 "Expected UpstreamProviderError for upstream_http_429, got: {:?}",
1405 completion_error
1406 ),
1407 }
1408
1409 // Invalid JSON in error body should fall back to regular error handling
1410 let error_body = "Not JSON at all";
1411
1412 let api_error = ApiError {
1413 status: StatusCode::INTERNAL_SERVER_ERROR,
1414 body: error_body.to_string(),
1415 headers: HeaderMap::new(),
1416 };
1417
1418 let completion_error: LanguageModelCompletionError = api_error.into();
1419
1420 match completion_error {
1421 LanguageModelCompletionError::ApiInternalServerError { provider, .. } => {
1422 assert_eq!(provider, PROVIDER_NAME);
1423 }
1424 _ => panic!(
1425 "Expected ApiInternalServerError for invalid JSON, got: {:?}",
1426 completion_error
1427 ),
1428 }
1429 }
1430}