1use ai_onboarding::YoungAccountBanner;
2use anthropic::AnthropicModelMode;
3use anyhow::{Context as _, Result, anyhow};
4use chrono::{DateTime, Utc};
5use client::{Client, ModelRequestUsage, UserStore, zed_urls};
6use cloud_llm_client::{
7 CLIENT_SUPPORTS_STATUS_MESSAGES_HEADER_NAME, CURRENT_PLAN_HEADER_NAME, CompletionBody,
8 CompletionEvent, CompletionRequestStatus, CountTokensBody, CountTokensResponse,
9 EXPIRED_LLM_TOKEN_HEADER_NAME, ListModelsResponse, MODEL_REQUESTS_RESOURCE_HEADER_VALUE, Plan,
10 PlanV1, PlanV2, SERVER_SUPPORTS_STATUS_MESSAGES_HEADER_NAME,
11 SUBSCRIPTION_LIMIT_RESOURCE_HEADER_NAME, TOOL_USE_LIMIT_REACHED_HEADER_NAME,
12 ZED_VERSION_HEADER_NAME,
13};
14use futures::{
15 AsyncBufReadExt, FutureExt, Stream, StreamExt, future::BoxFuture, stream::BoxStream,
16};
17use google_ai::GoogleModelMode;
18use gpui::{
19 AnyElement, AnyView, App, AsyncApp, Context, Entity, SemanticVersion, Subscription, Task,
20};
21use http_client::http::{HeaderMap, HeaderValue};
22use http_client::{AsyncBody, HttpClient, Method, Response, StatusCode};
23use language_model::{
24 AuthenticateError, LanguageModel, LanguageModelCacheConfiguration,
25 LanguageModelCompletionError, LanguageModelCompletionEvent, LanguageModelId, LanguageModelName,
26 LanguageModelProvider, LanguageModelProviderId, LanguageModelProviderName,
27 LanguageModelProviderState, LanguageModelRequest, LanguageModelToolChoice,
28 LanguageModelToolSchemaFormat, LlmApiToken, ModelRequestLimitReachedError,
29 PaymentRequiredError, RateLimiter, RefreshLlmTokenListener,
30};
31use release_channel::AppVersion;
32use schemars::JsonSchema;
33use serde::{Deserialize, Serialize, de::DeserializeOwned};
34use settings::SettingsStore;
35use smol::io::{AsyncReadExt, BufReader};
36use std::pin::Pin;
37use std::str::FromStr as _;
38use std::sync::Arc;
39use std::time::Duration;
40use thiserror::Error;
41use ui::{TintColor, prelude::*};
42use util::{ResultExt as _, maybe};
43
44use crate::provider::anthropic::{AnthropicEventMapper, count_anthropic_tokens, into_anthropic};
45use crate::provider::google::{GoogleEventMapper, into_google};
46use crate::provider::open_ai::{OpenAiEventMapper, count_open_ai_tokens, into_open_ai};
47
48const PROVIDER_ID: LanguageModelProviderId = language_model::ZED_CLOUD_PROVIDER_ID;
49const PROVIDER_NAME: LanguageModelProviderName = language_model::ZED_CLOUD_PROVIDER_NAME;
50
51#[derive(Default, Clone, Debug, PartialEq)]
52pub struct ZedDotDevSettings {
53 pub available_models: Vec<AvailableModel>,
54}
55
56#[derive(Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]
57#[serde(rename_all = "lowercase")]
58pub enum AvailableProvider {
59 Anthropic,
60 OpenAi,
61 Google,
62}
63
64#[derive(Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]
65pub struct AvailableModel {
66 /// The provider of the language model.
67 pub provider: AvailableProvider,
68 /// The model's name in the provider's API. e.g. claude-3-5-sonnet-20240620
69 pub name: String,
70 /// The name displayed in the UI, such as in the assistant panel model dropdown menu.
71 pub display_name: Option<String>,
72 /// The size of the context window, indicating the maximum number of tokens the model can process.
73 pub max_tokens: usize,
74 /// The maximum number of output tokens allowed by the model.
75 pub max_output_tokens: Option<u64>,
76 /// The maximum number of completion tokens allowed by the model (o1-* only)
77 pub max_completion_tokens: Option<u64>,
78 /// Override this model with a different Anthropic model for tool calls.
79 pub tool_override: Option<String>,
80 /// Indicates whether this custom model supports caching.
81 pub cache_configuration: Option<LanguageModelCacheConfiguration>,
82 /// The default temperature to use for this model.
83 pub default_temperature: Option<f32>,
84 /// Any extra beta headers to provide when using the model.
85 #[serde(default)]
86 pub extra_beta_headers: Vec<String>,
87 /// The model's mode (e.g. thinking)
88 pub mode: Option<ModelMode>,
89}
90
91#[derive(Default, Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]
92#[serde(tag = "type", rename_all = "lowercase")]
93pub enum ModelMode {
94 #[default]
95 Default,
96 Thinking {
97 /// The maximum number of tokens to use for reasoning. Must be lower than the model's `max_output_tokens`.
98 budget_tokens: Option<u32>,
99 },
100}
101
102impl From<ModelMode> for AnthropicModelMode {
103 fn from(value: ModelMode) -> Self {
104 match value {
105 ModelMode::Default => AnthropicModelMode::Default,
106 ModelMode::Thinking { budget_tokens } => AnthropicModelMode::Thinking { budget_tokens },
107 }
108 }
109}
110
111pub struct CloudLanguageModelProvider {
112 client: Arc<Client>,
113 state: gpui::Entity<State>,
114 _maintain_client_status: Task<()>,
115}
116
117pub struct State {
118 client: Arc<Client>,
119 llm_api_token: LlmApiToken,
120 user_store: Entity<UserStore>,
121 status: client::Status,
122 models: Vec<Arc<cloud_llm_client::LanguageModel>>,
123 default_model: Option<Arc<cloud_llm_client::LanguageModel>>,
124 default_fast_model: Option<Arc<cloud_llm_client::LanguageModel>>,
125 recommended_models: Vec<Arc<cloud_llm_client::LanguageModel>>,
126 _fetch_models_task: Task<()>,
127 _settings_subscription: Subscription,
128 _llm_token_subscription: Subscription,
129}
130
131impl State {
132 fn new(
133 client: Arc<Client>,
134 user_store: Entity<UserStore>,
135 status: client::Status,
136 cx: &mut Context<Self>,
137 ) -> Self {
138 let refresh_llm_token_listener = RefreshLlmTokenListener::global(cx);
139 let mut current_user = user_store.read(cx).watch_current_user();
140 Self {
141 client: client.clone(),
142 llm_api_token: LlmApiToken::default(),
143 user_store,
144 status,
145 models: Vec::new(),
146 default_model: None,
147 default_fast_model: None,
148 recommended_models: Vec::new(),
149 _fetch_models_task: cx.spawn(async move |this, cx| {
150 maybe!(async move {
151 let (client, llm_api_token) = this
152 .read_with(cx, |this, _cx| (client.clone(), this.llm_api_token.clone()))?;
153
154 while current_user.borrow().is_none() {
155 current_user.next().await;
156 }
157
158 let response =
159 Self::fetch_models(client.clone(), llm_api_token.clone()).await?;
160 this.update(cx, |this, cx| this.update_models(response, cx))?;
161 anyhow::Ok(())
162 })
163 .await
164 .context("failed to fetch Zed models")
165 .log_err();
166 }),
167 _settings_subscription: cx.observe_global::<SettingsStore>(|_, cx| {
168 cx.notify();
169 }),
170 _llm_token_subscription: cx.subscribe(
171 &refresh_llm_token_listener,
172 move |this, _listener, _event, cx| {
173 let client = this.client.clone();
174 let llm_api_token = this.llm_api_token.clone();
175 cx.spawn(async move |this, cx| {
176 llm_api_token.refresh(&client).await?;
177 let response = Self::fetch_models(client, llm_api_token).await?;
178 this.update(cx, |this, cx| {
179 this.update_models(response, cx);
180 })
181 })
182 .detach_and_log_err(cx);
183 },
184 ),
185 }
186 }
187
188 fn is_signed_out(&self, cx: &App) -> bool {
189 self.user_store.read(cx).current_user().is_none()
190 }
191
192 fn authenticate(&self, cx: &mut Context<Self>) -> Task<Result<()>> {
193 let client = self.client.clone();
194 cx.spawn(async move |state, cx| {
195 client.sign_in_with_optional_connect(true, cx).await?;
196 state.update(cx, |_, cx| cx.notify())
197 })
198 }
199 fn update_models(&mut self, response: ListModelsResponse, cx: &mut Context<Self>) {
200 let mut models = Vec::new();
201
202 for model in response.models {
203 models.push(Arc::new(model.clone()));
204
205 // Right now we represent thinking variants of models as separate models on the client,
206 // so we need to insert variants for any model that supports thinking.
207 if model.supports_thinking {
208 models.push(Arc::new(cloud_llm_client::LanguageModel {
209 id: cloud_llm_client::LanguageModelId(format!("{}-thinking", model.id).into()),
210 display_name: format!("{} Thinking", model.display_name),
211 ..model
212 }));
213 }
214 }
215
216 self.default_model = models
217 .iter()
218 .find(|model| {
219 response
220 .default_model
221 .as_ref()
222 .is_some_and(|default_model_id| &model.id == default_model_id)
223 })
224 .cloned();
225 self.default_fast_model = models
226 .iter()
227 .find(|model| {
228 response
229 .default_fast_model
230 .as_ref()
231 .is_some_and(|default_fast_model_id| &model.id == default_fast_model_id)
232 })
233 .cloned();
234 self.recommended_models = response
235 .recommended_models
236 .iter()
237 .filter_map(|id| models.iter().find(|model| &model.id == id))
238 .cloned()
239 .collect();
240 self.models = models;
241 cx.notify();
242 }
243
244 async fn fetch_models(
245 client: Arc<Client>,
246 llm_api_token: LlmApiToken,
247 ) -> Result<ListModelsResponse> {
248 let http_client = &client.http_client();
249 let token = llm_api_token.acquire(&client).await?;
250
251 let request = http_client::Request::builder()
252 .method(Method::GET)
253 .uri(http_client.build_zed_llm_url("/models", &[])?.as_ref())
254 .header("Authorization", format!("Bearer {token}"))
255 .body(AsyncBody::empty())?;
256 let mut response = http_client
257 .send(request)
258 .await
259 .context("failed to send list models request")?;
260
261 if response.status().is_success() {
262 let mut body = String::new();
263 response.body_mut().read_to_string(&mut body).await?;
264 Ok(serde_json::from_str(&body)?)
265 } else {
266 let mut body = String::new();
267 response.body_mut().read_to_string(&mut body).await?;
268 anyhow::bail!(
269 "error listing models.\nStatus: {:?}\nBody: {body}",
270 response.status(),
271 );
272 }
273 }
274}
275
276impl CloudLanguageModelProvider {
277 pub fn new(user_store: Entity<UserStore>, client: Arc<Client>, cx: &mut App) -> Self {
278 let mut status_rx = client.status();
279 let status = *status_rx.borrow();
280
281 let state = cx.new(|cx| State::new(client.clone(), user_store.clone(), status, cx));
282
283 let state_ref = state.downgrade();
284 let maintain_client_status = cx.spawn(async move |cx| {
285 while let Some(status) = status_rx.next().await {
286 if let Some(this) = state_ref.upgrade() {
287 _ = this.update(cx, |this, cx| {
288 if this.status != status {
289 this.status = status;
290 cx.notify();
291 }
292 });
293 } else {
294 break;
295 }
296 }
297 });
298
299 Self {
300 client,
301 state,
302 _maintain_client_status: maintain_client_status,
303 }
304 }
305
306 fn create_language_model(
307 &self,
308 model: Arc<cloud_llm_client::LanguageModel>,
309 llm_api_token: LlmApiToken,
310 ) -> Arc<dyn LanguageModel> {
311 Arc::new(CloudLanguageModel {
312 id: LanguageModelId(SharedString::from(model.id.0.clone())),
313 model,
314 llm_api_token,
315 client: self.client.clone(),
316 request_limiter: RateLimiter::new(4),
317 })
318 }
319}
320
321impl LanguageModelProviderState for CloudLanguageModelProvider {
322 type ObservableEntity = State;
323
324 fn observable_entity(&self) -> Option<gpui::Entity<Self::ObservableEntity>> {
325 Some(self.state.clone())
326 }
327}
328
329impl LanguageModelProvider for CloudLanguageModelProvider {
330 fn id(&self) -> LanguageModelProviderId {
331 PROVIDER_ID
332 }
333
334 fn name(&self) -> LanguageModelProviderName {
335 PROVIDER_NAME
336 }
337
338 fn icon(&self) -> IconName {
339 IconName::AiZed
340 }
341
342 fn default_model(&self, cx: &App) -> Option<Arc<dyn LanguageModel>> {
343 let default_model = self.state.read(cx).default_model.clone()?;
344 let llm_api_token = self.state.read(cx).llm_api_token.clone();
345 Some(self.create_language_model(default_model, llm_api_token))
346 }
347
348 fn default_fast_model(&self, cx: &App) -> Option<Arc<dyn LanguageModel>> {
349 let default_fast_model = self.state.read(cx).default_fast_model.clone()?;
350 let llm_api_token = self.state.read(cx).llm_api_token.clone();
351 Some(self.create_language_model(default_fast_model, llm_api_token))
352 }
353
354 fn recommended_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {
355 let llm_api_token = self.state.read(cx).llm_api_token.clone();
356 self.state
357 .read(cx)
358 .recommended_models
359 .iter()
360 .cloned()
361 .map(|model| self.create_language_model(model, llm_api_token.clone()))
362 .collect()
363 }
364
365 fn provided_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {
366 let llm_api_token = self.state.read(cx).llm_api_token.clone();
367 self.state
368 .read(cx)
369 .models
370 .iter()
371 .cloned()
372 .map(|model| self.create_language_model(model, llm_api_token.clone()))
373 .collect()
374 }
375
376 fn is_authenticated(&self, cx: &App) -> bool {
377 let state = self.state.read(cx);
378 !state.is_signed_out(cx)
379 }
380
381 fn authenticate(&self, _cx: &mut App) -> Task<Result<(), AuthenticateError>> {
382 Task::ready(Ok(()))
383 }
384
385 fn configuration_view(
386 &self,
387 _target_agent: language_model::ConfigurationViewTargetAgent,
388 _: &mut Window,
389 cx: &mut App,
390 ) -> AnyView {
391 cx.new(|_| ConfigurationView::new(self.state.clone()))
392 .into()
393 }
394
395 fn reset_credentials(&self, _cx: &mut App) -> Task<Result<()>> {
396 Task::ready(Ok(()))
397 }
398}
399
400pub struct CloudLanguageModel {
401 id: LanguageModelId,
402 model: Arc<cloud_llm_client::LanguageModel>,
403 llm_api_token: LlmApiToken,
404 client: Arc<Client>,
405 request_limiter: RateLimiter,
406}
407
408struct PerformLlmCompletionResponse {
409 response: Response<AsyncBody>,
410 usage: Option<ModelRequestUsage>,
411 tool_use_limit_reached: bool,
412 includes_status_messages: bool,
413}
414
415impl CloudLanguageModel {
416 async fn perform_llm_completion(
417 client: Arc<Client>,
418 llm_api_token: LlmApiToken,
419 app_version: Option<SemanticVersion>,
420 body: CompletionBody,
421 ) -> Result<PerformLlmCompletionResponse> {
422 let http_client = &client.http_client();
423
424 let mut token = llm_api_token.acquire(&client).await?;
425 let mut refreshed_token = false;
426
427 loop {
428 let request_builder = http_client::Request::builder()
429 .method(Method::POST)
430 .uri(http_client.build_zed_llm_url("/completions", &[])?.as_ref());
431 let request_builder = if let Some(app_version) = app_version {
432 request_builder.header(ZED_VERSION_HEADER_NAME, app_version.to_string())
433 } else {
434 request_builder
435 };
436
437 let request = request_builder
438 .header("Content-Type", "application/json")
439 .header("Authorization", format!("Bearer {token}"))
440 .header(CLIENT_SUPPORTS_STATUS_MESSAGES_HEADER_NAME, "true")
441 .body(serde_json::to_string(&body)?.into())?;
442 let mut response = http_client.send(request).await?;
443 let status = response.status();
444 if status.is_success() {
445 let includes_status_messages = response
446 .headers()
447 .get(SERVER_SUPPORTS_STATUS_MESSAGES_HEADER_NAME)
448 .is_some();
449
450 let tool_use_limit_reached = response
451 .headers()
452 .get(TOOL_USE_LIMIT_REACHED_HEADER_NAME)
453 .is_some();
454
455 let usage = if includes_status_messages {
456 None
457 } else {
458 ModelRequestUsage::from_headers(response.headers()).ok()
459 };
460
461 return Ok(PerformLlmCompletionResponse {
462 response,
463 usage,
464 includes_status_messages,
465 tool_use_limit_reached,
466 });
467 }
468
469 if !refreshed_token
470 && response
471 .headers()
472 .get(EXPIRED_LLM_TOKEN_HEADER_NAME)
473 .is_some()
474 {
475 token = llm_api_token.refresh(&client).await?;
476 refreshed_token = true;
477 continue;
478 }
479
480 if status == StatusCode::FORBIDDEN
481 && response
482 .headers()
483 .get(SUBSCRIPTION_LIMIT_RESOURCE_HEADER_NAME)
484 .is_some()
485 {
486 if let Some(MODEL_REQUESTS_RESOURCE_HEADER_VALUE) = response
487 .headers()
488 .get(SUBSCRIPTION_LIMIT_RESOURCE_HEADER_NAME)
489 .and_then(|resource| resource.to_str().ok())
490 && let Some(plan) = response
491 .headers()
492 .get(CURRENT_PLAN_HEADER_NAME)
493 .and_then(|plan| plan.to_str().ok())
494 .and_then(|plan| cloud_llm_client::PlanV1::from_str(plan).ok())
495 .map(Plan::V1)
496 {
497 return Err(anyhow!(ModelRequestLimitReachedError { plan }));
498 }
499 } else if status == StatusCode::PAYMENT_REQUIRED {
500 return Err(anyhow!(PaymentRequiredError));
501 }
502
503 let mut body = String::new();
504 let headers = response.headers().clone();
505 response.body_mut().read_to_string(&mut body).await?;
506 return Err(anyhow!(ApiError {
507 status,
508 body,
509 headers
510 }));
511 }
512 }
513}
514
515#[derive(Debug, Error)]
516#[error("cloud language model request failed with status {status}: {body}")]
517struct ApiError {
518 status: StatusCode,
519 body: String,
520 headers: HeaderMap<HeaderValue>,
521}
522
523/// Represents error responses from Zed's cloud API.
524///
525/// Example JSON for an upstream HTTP error:
526/// ```json
527/// {
528/// "code": "upstream_http_error",
529/// "message": "Received an error from the Anthropic API: upstream connect error or disconnect/reset before headers, reset reason: connection timeout",
530/// "upstream_status": 503
531/// }
532/// ```
533#[derive(Debug, serde::Deserialize)]
534struct CloudApiError {
535 code: String,
536 message: String,
537 #[serde(default)]
538 #[serde(deserialize_with = "deserialize_optional_status_code")]
539 upstream_status: Option<StatusCode>,
540 #[serde(default)]
541 retry_after: Option<f64>,
542}
543
544fn deserialize_optional_status_code<'de, D>(deserializer: D) -> Result<Option<StatusCode>, D::Error>
545where
546 D: serde::Deserializer<'de>,
547{
548 let opt: Option<u16> = Option::deserialize(deserializer)?;
549 Ok(opt.and_then(|code| StatusCode::from_u16(code).ok()))
550}
551
552impl From<ApiError> for LanguageModelCompletionError {
553 fn from(error: ApiError) -> Self {
554 if let Ok(cloud_error) = serde_json::from_str::<CloudApiError>(&error.body) {
555 if cloud_error.code.starts_with("upstream_http_") {
556 let status = if let Some(status) = cloud_error.upstream_status {
557 status
558 } else if cloud_error.code.ends_with("_error") {
559 error.status
560 } else {
561 // If there's a status code in the code string (e.g. "upstream_http_429")
562 // then use that; otherwise, see if the JSON contains a status code.
563 cloud_error
564 .code
565 .strip_prefix("upstream_http_")
566 .and_then(|code_str| code_str.parse::<u16>().ok())
567 .and_then(|code| StatusCode::from_u16(code).ok())
568 .unwrap_or(error.status)
569 };
570
571 return LanguageModelCompletionError::UpstreamProviderError {
572 message: cloud_error.message,
573 status,
574 retry_after: cloud_error.retry_after.map(Duration::from_secs_f64),
575 };
576 }
577
578 return LanguageModelCompletionError::from_http_status(
579 PROVIDER_NAME,
580 error.status,
581 cloud_error.message,
582 None,
583 );
584 }
585
586 let retry_after = None;
587 LanguageModelCompletionError::from_http_status(
588 PROVIDER_NAME,
589 error.status,
590 error.body,
591 retry_after,
592 )
593 }
594}
595
596impl LanguageModel for CloudLanguageModel {
597 fn id(&self) -> LanguageModelId {
598 self.id.clone()
599 }
600
601 fn name(&self) -> LanguageModelName {
602 LanguageModelName::from(self.model.display_name.clone())
603 }
604
605 fn provider_id(&self) -> LanguageModelProviderId {
606 PROVIDER_ID
607 }
608
609 fn provider_name(&self) -> LanguageModelProviderName {
610 PROVIDER_NAME
611 }
612
613 fn upstream_provider_id(&self) -> LanguageModelProviderId {
614 use cloud_llm_client::LanguageModelProvider::*;
615 match self.model.provider {
616 Anthropic => language_model::ANTHROPIC_PROVIDER_ID,
617 OpenAi => language_model::OPEN_AI_PROVIDER_ID,
618 Google => language_model::GOOGLE_PROVIDER_ID,
619 }
620 }
621
622 fn upstream_provider_name(&self) -> LanguageModelProviderName {
623 use cloud_llm_client::LanguageModelProvider::*;
624 match self.model.provider {
625 Anthropic => language_model::ANTHROPIC_PROVIDER_NAME,
626 OpenAi => language_model::OPEN_AI_PROVIDER_NAME,
627 Google => language_model::GOOGLE_PROVIDER_NAME,
628 }
629 }
630
631 fn supports_tools(&self) -> bool {
632 self.model.supports_tools
633 }
634
635 fn supports_images(&self) -> bool {
636 self.model.supports_images
637 }
638
639 fn supports_tool_choice(&self, choice: LanguageModelToolChoice) -> bool {
640 match choice {
641 LanguageModelToolChoice::Auto
642 | LanguageModelToolChoice::Any
643 | LanguageModelToolChoice::None => true,
644 }
645 }
646
647 fn supports_burn_mode(&self) -> bool {
648 self.model.supports_max_mode
649 }
650
651 fn telemetry_id(&self) -> String {
652 format!("zed.dev/{}", self.model.id)
653 }
654
655 fn tool_input_format(&self) -> LanguageModelToolSchemaFormat {
656 match self.model.provider {
657 cloud_llm_client::LanguageModelProvider::Anthropic
658 | cloud_llm_client::LanguageModelProvider::OpenAi => {
659 LanguageModelToolSchemaFormat::JsonSchema
660 }
661 cloud_llm_client::LanguageModelProvider::Google => {
662 LanguageModelToolSchemaFormat::JsonSchemaSubset
663 }
664 }
665 }
666
667 fn max_token_count(&self) -> u64 {
668 self.model.max_token_count as u64
669 }
670
671 fn max_token_count_in_burn_mode(&self) -> Option<u64> {
672 self.model
673 .max_token_count_in_max_mode
674 .filter(|_| self.model.supports_max_mode)
675 .map(|max_token_count| max_token_count as u64)
676 }
677
678 fn cache_configuration(&self) -> Option<LanguageModelCacheConfiguration> {
679 match &self.model.provider {
680 cloud_llm_client::LanguageModelProvider::Anthropic => {
681 Some(LanguageModelCacheConfiguration {
682 min_total_token: 2_048,
683 should_speculate: true,
684 max_cache_anchors: 4,
685 })
686 }
687 cloud_llm_client::LanguageModelProvider::OpenAi
688 | cloud_llm_client::LanguageModelProvider::Google => None,
689 }
690 }
691
692 fn count_tokens(
693 &self,
694 request: LanguageModelRequest,
695 cx: &App,
696 ) -> BoxFuture<'static, Result<u64>> {
697 match self.model.provider {
698 cloud_llm_client::LanguageModelProvider::Anthropic => {
699 count_anthropic_tokens(request, cx)
700 }
701 cloud_llm_client::LanguageModelProvider::OpenAi => {
702 let model = match open_ai::Model::from_id(&self.model.id.0) {
703 Ok(model) => model,
704 Err(err) => return async move { Err(anyhow!(err)) }.boxed(),
705 };
706 count_open_ai_tokens(request, model, cx)
707 }
708 cloud_llm_client::LanguageModelProvider::Google => {
709 let client = self.client.clone();
710 let llm_api_token = self.llm_api_token.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).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 intent = request.intent;
776 let mode = request.mode;
777 let app_version = cx.update(|cx| AppVersion::global(cx)).ok();
778 let thinking_allowed = request.thinking_allowed;
779 match self.model.provider {
780 cloud_llm_client::LanguageModelProvider::Anthropic => {
781 let request = into_anthropic(
782 request,
783 self.model.id.to_string(),
784 1.0,
785 self.model.max_output_tokens as u64,
786 if thinking_allowed && self.model.id.0.ends_with("-thinking") {
787 AnthropicModelMode::Thinking {
788 budget_tokens: Some(4_096),
789 }
790 } else {
791 AnthropicModelMode::Default
792 },
793 );
794 let client = self.client.clone();
795 let llm_api_token = self.llm_api_token.clone();
796 let future = self.request_limiter.stream(async move {
797 let PerformLlmCompletionResponse {
798 response,
799 usage,
800 includes_status_messages,
801 tool_use_limit_reached,
802 } = Self::perform_llm_completion(
803 client.clone(),
804 llm_api_token,
805 app_version,
806 CompletionBody {
807 thread_id,
808 prompt_id,
809 intent,
810 mode,
811 provider: cloud_llm_client::LanguageModelProvider::Anthropic,
812 model: request.model.clone(),
813 provider_request: serde_json::to_value(&request)
814 .map_err(|e| anyhow!(e))?,
815 },
816 )
817 .await
818 .map_err(|err| match err.downcast::<ApiError>() {
819 Ok(api_err) => anyhow!(LanguageModelCompletionError::from(api_err)),
820 Err(err) => anyhow!(err),
821 })?;
822
823 let mut mapper = AnthropicEventMapper::new();
824 Ok(map_cloud_completion_events(
825 Box::pin(
826 response_lines(response, includes_status_messages)
827 .chain(usage_updated_event(usage))
828 .chain(tool_use_limit_reached_event(tool_use_limit_reached)),
829 ),
830 move |event| mapper.map_event(event),
831 ))
832 });
833 async move { Ok(future.await?.boxed()) }.boxed()
834 }
835 cloud_llm_client::LanguageModelProvider::OpenAi => {
836 let client = self.client.clone();
837 let model = match open_ai::Model::from_id(&self.model.id.0) {
838 Ok(model) => model,
839 Err(err) => return async move { Err(anyhow!(err).into()) }.boxed(),
840 };
841 let request = into_open_ai(
842 request,
843 model.id(),
844 model.supports_parallel_tool_calls(),
845 model.supports_prompt_cache_key(),
846 None,
847 None,
848 );
849 let llm_api_token = self.llm_api_token.clone();
850 let future = self.request_limiter.stream(async move {
851 let PerformLlmCompletionResponse {
852 response,
853 usage,
854 includes_status_messages,
855 tool_use_limit_reached,
856 } = Self::perform_llm_completion(
857 client.clone(),
858 llm_api_token,
859 app_version,
860 CompletionBody {
861 thread_id,
862 prompt_id,
863 intent,
864 mode,
865 provider: cloud_llm_client::LanguageModelProvider::OpenAi,
866 model: request.model.clone(),
867 provider_request: serde_json::to_value(&request)
868 .map_err(|e| anyhow!(e))?,
869 },
870 )
871 .await?;
872
873 let mut mapper = OpenAiEventMapper::new();
874 Ok(map_cloud_completion_events(
875 Box::pin(
876 response_lines(response, includes_status_messages)
877 .chain(usage_updated_event(usage))
878 .chain(tool_use_limit_reached_event(tool_use_limit_reached)),
879 ),
880 move |event| mapper.map_event(event),
881 ))
882 });
883 async move { Ok(future.await?.boxed()) }.boxed()
884 }
885 cloud_llm_client::LanguageModelProvider::Google => {
886 let client = self.client.clone();
887 let request =
888 into_google(request, self.model.id.to_string(), GoogleModelMode::Default);
889 let llm_api_token = self.llm_api_token.clone();
890 let future = self.request_limiter.stream(async move {
891 let PerformLlmCompletionResponse {
892 response,
893 usage,
894 includes_status_messages,
895 tool_use_limit_reached,
896 } = Self::perform_llm_completion(
897 client.clone(),
898 llm_api_token,
899 app_version,
900 CompletionBody {
901 thread_id,
902 prompt_id,
903 intent,
904 mode,
905 provider: cloud_llm_client::LanguageModelProvider::Google,
906 model: request.model.model_id.clone(),
907 provider_request: serde_json::to_value(&request)
908 .map_err(|e| anyhow!(e))?,
909 },
910 )
911 .await?;
912
913 let mut mapper = GoogleEventMapper::new();
914 Ok(map_cloud_completion_events(
915 Box::pin(
916 response_lines(response, includes_status_messages)
917 .chain(usage_updated_event(usage))
918 .chain(tool_use_limit_reached_event(tool_use_limit_reached)),
919 ),
920 move |event| mapper.map_event(event),
921 ))
922 });
923 async move { Ok(future.await?.boxed()) }.boxed()
924 }
925 }
926 }
927}
928
929fn map_cloud_completion_events<T, F>(
930 stream: Pin<Box<dyn Stream<Item = Result<CompletionEvent<T>>> + Send>>,
931 mut map_callback: F,
932) -> BoxStream<'static, Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
933where
934 T: DeserializeOwned + 'static,
935 F: FnMut(T) -> Vec<Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
936 + Send
937 + 'static,
938{
939 stream
940 .flat_map(move |event| {
941 futures::stream::iter(match event {
942 Err(error) => {
943 vec![Err(LanguageModelCompletionError::from(error))]
944 }
945 Ok(CompletionEvent::Status(event)) => {
946 vec![Ok(LanguageModelCompletionEvent::StatusUpdate(event))]
947 }
948 Ok(CompletionEvent::Event(event)) => map_callback(event),
949 })
950 })
951 .boxed()
952}
953
954fn usage_updated_event<T>(
955 usage: Option<ModelRequestUsage>,
956) -> impl Stream<Item = Result<CompletionEvent<T>>> {
957 futures::stream::iter(usage.map(|usage| {
958 Ok(CompletionEvent::Status(
959 CompletionRequestStatus::UsageUpdated {
960 amount: usage.amount as usize,
961 limit: usage.limit,
962 },
963 ))
964 }))
965}
966
967fn tool_use_limit_reached_event<T>(
968 tool_use_limit_reached: bool,
969) -> impl Stream<Item = Result<CompletionEvent<T>>> {
970 futures::stream::iter(tool_use_limit_reached.then(|| {
971 Ok(CompletionEvent::Status(
972 CompletionRequestStatus::ToolUseLimitReached,
973 ))
974 }))
975}
976
977fn response_lines<T: DeserializeOwned>(
978 response: Response<AsyncBody>,
979 includes_status_messages: bool,
980) -> impl Stream<Item = Result<CompletionEvent<T>>> {
981 futures::stream::try_unfold(
982 (String::new(), BufReader::new(response.into_body())),
983 move |(mut line, mut body)| async move {
984 match body.read_line(&mut line).await {
985 Ok(0) => Ok(None),
986 Ok(_) => {
987 let event = if includes_status_messages {
988 serde_json::from_str::<CompletionEvent<T>>(&line)?
989 } else {
990 CompletionEvent::Event(serde_json::from_str::<T>(&line)?)
991 };
992
993 line.clear();
994 Ok(Some((event, (line, body))))
995 }
996 Err(e) => Err(e.into()),
997 }
998 },
999 )
1000}
1001
1002#[derive(IntoElement, RegisterComponent)]
1003struct ZedAiConfiguration {
1004 is_connected: bool,
1005 plan: Option<Plan>,
1006 subscription_period: Option<(DateTime<Utc>, DateTime<Utc>)>,
1007 eligible_for_trial: bool,
1008 account_too_young: bool,
1009 sign_in_callback: Arc<dyn Fn(&mut Window, &mut App) + Send + Sync>,
1010}
1011
1012impl RenderOnce for ZedAiConfiguration {
1013 fn render(self, _window: &mut Window, _cx: &mut App) -> impl IntoElement {
1014 let young_account_banner = YoungAccountBanner;
1015
1016 let is_pro = self.plan.is_some_and(|plan| {
1017 matches!(plan, Plan::V1(PlanV1::ZedPro) | Plan::V2(PlanV2::ZedPro))
1018 });
1019 let subscription_text = match (self.plan, self.subscription_period) {
1020 (Some(Plan::V1(PlanV1::ZedPro) | Plan::V2(PlanV2::ZedPro)), Some(_)) => {
1021 "You have access to Zed's hosted models through your Pro subscription."
1022 }
1023 (Some(Plan::V1(PlanV1::ZedProTrial) | Plan::V2(PlanV2::ZedProTrial)), Some(_)) => {
1024 "You have access to Zed's hosted models through your Pro trial."
1025 }
1026 (Some(Plan::V1(PlanV1::ZedFree) | Plan::V2(PlanV2::ZedFree)), Some(_)) => {
1027 "You have basic access to Zed's hosted models through the Free plan."
1028 }
1029 _ => {
1030 if self.eligible_for_trial {
1031 "Subscribe for access to Zed's hosted models. Start with a 14 day free trial."
1032 } else {
1033 "Subscribe for access to Zed's hosted models."
1034 }
1035 }
1036 };
1037
1038 let manage_subscription_buttons = if is_pro {
1039 Button::new("manage_settings", "Manage Subscription")
1040 .full_width()
1041 .style(ButtonStyle::Tinted(TintColor::Accent))
1042 .on_click(|_, _, cx| cx.open_url(&zed_urls::account_url(cx)))
1043 .into_any_element()
1044 } else if self.plan.is_none() || self.eligible_for_trial {
1045 Button::new("start_trial", "Start 14-day Free Pro Trial")
1046 .full_width()
1047 .style(ui::ButtonStyle::Tinted(ui::TintColor::Accent))
1048 .on_click(|_, _, cx| cx.open_url(&zed_urls::start_trial_url(cx)))
1049 .into_any_element()
1050 } else {
1051 Button::new("upgrade", "Upgrade to Pro")
1052 .full_width()
1053 .style(ui::ButtonStyle::Tinted(ui::TintColor::Accent))
1054 .on_click(|_, _, cx| cx.open_url(&zed_urls::upgrade_to_zed_pro_url(cx)))
1055 .into_any_element()
1056 };
1057
1058 if !self.is_connected {
1059 return v_flex()
1060 .gap_2()
1061 .child(Label::new("Sign in to have access to Zed's complete agentic experience with hosted models."))
1062 .child(
1063 Button::new("sign_in", "Sign In to use Zed AI")
1064 .icon_color(Color::Muted)
1065 .icon(IconName::Github)
1066 .icon_size(IconSize::Small)
1067 .icon_position(IconPosition::Start)
1068 .full_width()
1069 .on_click({
1070 let callback = self.sign_in_callback.clone();
1071 move |_, window, cx| (callback)(window, cx)
1072 }),
1073 );
1074 }
1075
1076 v_flex().gap_2().w_full().map(|this| {
1077 if self.account_too_young {
1078 this.child(young_account_banner).child(
1079 Button::new("upgrade", "Upgrade to Pro")
1080 .style(ui::ButtonStyle::Tinted(ui::TintColor::Accent))
1081 .full_width()
1082 .on_click(|_, _, cx| cx.open_url(&zed_urls::upgrade_to_zed_pro_url(cx))),
1083 )
1084 } else {
1085 this.text_sm()
1086 .child(subscription_text)
1087 .child(manage_subscription_buttons)
1088 }
1089 })
1090 }
1091}
1092
1093struct ConfigurationView {
1094 state: Entity<State>,
1095 sign_in_callback: Arc<dyn Fn(&mut Window, &mut App) + Send + Sync>,
1096}
1097
1098impl ConfigurationView {
1099 fn new(state: Entity<State>) -> Self {
1100 let sign_in_callback = Arc::new({
1101 let state = state.clone();
1102 move |_window: &mut Window, cx: &mut App| {
1103 state.update(cx, |state, cx| {
1104 state.authenticate(cx).detach_and_log_err(cx);
1105 });
1106 }
1107 });
1108
1109 Self {
1110 state,
1111 sign_in_callback,
1112 }
1113 }
1114}
1115
1116impl Render for ConfigurationView {
1117 fn render(&mut self, _: &mut Window, cx: &mut Context<Self>) -> impl IntoElement {
1118 let state = self.state.read(cx);
1119 let user_store = state.user_store.read(cx);
1120
1121 ZedAiConfiguration {
1122 is_connected: !state.is_signed_out(cx),
1123 plan: user_store.plan(),
1124 subscription_period: user_store.subscription_period(),
1125 eligible_for_trial: user_store.trial_started_at().is_none(),
1126 account_too_young: user_store.account_too_young(),
1127 sign_in_callback: self.sign_in_callback.clone(),
1128 }
1129 }
1130}
1131
1132impl Component for ZedAiConfiguration {
1133 fn name() -> &'static str {
1134 "AI Configuration Content"
1135 }
1136
1137 fn sort_name() -> &'static str {
1138 "AI Configuration Content"
1139 }
1140
1141 fn scope() -> ComponentScope {
1142 ComponentScope::Onboarding
1143 }
1144
1145 fn preview(_window: &mut Window, _cx: &mut App) -> Option<AnyElement> {
1146 fn configuration(
1147 is_connected: bool,
1148 plan: Option<Plan>,
1149 eligible_for_trial: bool,
1150 account_too_young: bool,
1151 ) -> AnyElement {
1152 ZedAiConfiguration {
1153 is_connected,
1154 plan,
1155 subscription_period: plan
1156 .is_some()
1157 .then(|| (Utc::now(), Utc::now() + chrono::Duration::days(7))),
1158 eligible_for_trial,
1159 account_too_young,
1160 sign_in_callback: Arc::new(|_, _| {}),
1161 }
1162 .into_any_element()
1163 }
1164
1165 Some(
1166 v_flex()
1167 .p_4()
1168 .gap_4()
1169 .children(vec![
1170 single_example("Not connected", configuration(false, None, false, false)),
1171 single_example(
1172 "Accept Terms of Service",
1173 configuration(true, None, true, false),
1174 ),
1175 single_example(
1176 "No Plan - Not eligible for trial",
1177 configuration(true, None, false, false),
1178 ),
1179 single_example(
1180 "No Plan - Eligible for trial",
1181 configuration(true, None, true, false),
1182 ),
1183 single_example(
1184 "Free Plan",
1185 configuration(true, Some(Plan::V1(PlanV1::ZedFree)), true, false),
1186 ),
1187 single_example(
1188 "Zed Pro Trial Plan",
1189 configuration(true, Some(Plan::V1(PlanV1::ZedProTrial)), true, false),
1190 ),
1191 single_example(
1192 "Zed Pro Plan",
1193 configuration(true, Some(Plan::V1(PlanV1::ZedPro)), true, false),
1194 ),
1195 ])
1196 .into_any_element(),
1197 )
1198 }
1199}
1200
1201#[cfg(test)]
1202mod tests {
1203 use super::*;
1204 use http_client::http::{HeaderMap, StatusCode};
1205 use language_model::LanguageModelCompletionError;
1206
1207 #[test]
1208 fn test_api_error_conversion_with_upstream_http_error() {
1209 // upstream_http_error with 503 status should become ServerOverloaded
1210 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}"#;
1211
1212 let api_error = ApiError {
1213 status: StatusCode::INTERNAL_SERVER_ERROR,
1214 body: error_body.to_string(),
1215 headers: HeaderMap::new(),
1216 };
1217
1218 let completion_error: LanguageModelCompletionError = api_error.into();
1219
1220 match completion_error {
1221 LanguageModelCompletionError::UpstreamProviderError { message, .. } => {
1222 assert_eq!(
1223 message,
1224 "Received an error from the Anthropic API: upstream connect error or disconnect/reset before headers, reset reason: connection timeout"
1225 );
1226 }
1227 _ => panic!(
1228 "Expected UpstreamProviderError for upstream 503, got: {:?}",
1229 completion_error
1230 ),
1231 }
1232
1233 // upstream_http_error with 500 status should become ApiInternalServerError
1234 let error_body = r#"{"code":"upstream_http_error","message":"Received an error from the OpenAI API: internal server error","upstream_status":500}"#;
1235
1236 let api_error = ApiError {
1237 status: StatusCode::INTERNAL_SERVER_ERROR,
1238 body: error_body.to_string(),
1239 headers: HeaderMap::new(),
1240 };
1241
1242 let completion_error: LanguageModelCompletionError = api_error.into();
1243
1244 match completion_error {
1245 LanguageModelCompletionError::UpstreamProviderError { message, .. } => {
1246 assert_eq!(
1247 message,
1248 "Received an error from the OpenAI API: internal server error"
1249 );
1250 }
1251 _ => panic!(
1252 "Expected UpstreamProviderError for upstream 500, got: {:?}",
1253 completion_error
1254 ),
1255 }
1256
1257 // upstream_http_error with 429 status should become RateLimitExceeded
1258 let error_body = r#"{"code":"upstream_http_error","message":"Received an error from the Google API: rate limit exceeded","upstream_status":429}"#;
1259
1260 let api_error = ApiError {
1261 status: StatusCode::INTERNAL_SERVER_ERROR,
1262 body: error_body.to_string(),
1263 headers: HeaderMap::new(),
1264 };
1265
1266 let completion_error: LanguageModelCompletionError = api_error.into();
1267
1268 match completion_error {
1269 LanguageModelCompletionError::UpstreamProviderError { message, .. } => {
1270 assert_eq!(
1271 message,
1272 "Received an error from the Google API: rate limit exceeded"
1273 );
1274 }
1275 _ => panic!(
1276 "Expected UpstreamProviderError for upstream 429, got: {:?}",
1277 completion_error
1278 ),
1279 }
1280
1281 // Regular 500 error without upstream_http_error should remain ApiInternalServerError for Zed
1282 let error_body = "Regular internal server error";
1283
1284 let api_error = ApiError {
1285 status: StatusCode::INTERNAL_SERVER_ERROR,
1286 body: error_body.to_string(),
1287 headers: HeaderMap::new(),
1288 };
1289
1290 let completion_error: LanguageModelCompletionError = api_error.into();
1291
1292 match completion_error {
1293 LanguageModelCompletionError::ApiInternalServerError { provider, message } => {
1294 assert_eq!(provider, PROVIDER_NAME);
1295 assert_eq!(message, "Regular internal server error");
1296 }
1297 _ => panic!(
1298 "Expected ApiInternalServerError for regular 500, got: {:?}",
1299 completion_error
1300 ),
1301 }
1302
1303 // upstream_http_429 format should be converted to UpstreamProviderError
1304 let error_body = r#"{"code":"upstream_http_429","message":"Upstream Anthropic rate limit exceeded.","retry_after":30.5}"#;
1305
1306 let api_error = ApiError {
1307 status: StatusCode::INTERNAL_SERVER_ERROR,
1308 body: error_body.to_string(),
1309 headers: HeaderMap::new(),
1310 };
1311
1312 let completion_error: LanguageModelCompletionError = api_error.into();
1313
1314 match completion_error {
1315 LanguageModelCompletionError::UpstreamProviderError {
1316 message,
1317 status,
1318 retry_after,
1319 } => {
1320 assert_eq!(message, "Upstream Anthropic rate limit exceeded.");
1321 assert_eq!(status, StatusCode::TOO_MANY_REQUESTS);
1322 assert_eq!(retry_after, Some(Duration::from_secs_f64(30.5)));
1323 }
1324 _ => panic!(
1325 "Expected UpstreamProviderError for upstream_http_429, got: {:?}",
1326 completion_error
1327 ),
1328 }
1329
1330 // Invalid JSON in error body should fall back to regular error handling
1331 let error_body = "Not JSON at all";
1332
1333 let api_error = ApiError {
1334 status: StatusCode::INTERNAL_SERVER_ERROR,
1335 body: error_body.to_string(),
1336 headers: HeaderMap::new(),
1337 };
1338
1339 let completion_error: LanguageModelCompletionError = api_error.into();
1340
1341 match completion_error {
1342 LanguageModelCompletionError::ApiInternalServerError { provider, .. } => {
1343 assert_eq!(provider, PROVIDER_NAME);
1344 }
1345 _ => panic!(
1346 "Expected ApiInternalServerError for invalid JSON, got: {:?}",
1347 completion_error
1348 ),
1349 }
1350 }
1351}