1use super::open_ai::count_open_ai_tokens;
2use crate::{
3 settings::AllLanguageModelSettings, CloudModel, LanguageModel, LanguageModelId,
4 LanguageModelName, LanguageModelProviderId, LanguageModelProviderName,
5 LanguageModelProviderState, LanguageModelRequest, RateLimiter, ZedModel,
6};
7use anthropic::AnthropicError;
8use anyhow::{anyhow, bail, Context as _, Result};
9use client::{Client, PerformCompletionParams, UserStore, EXPIRED_LLM_TOKEN_HEADER_NAME};
10use collections::BTreeMap;
11use feature_flags::{FeatureFlagAppExt, ZedPro};
12use futures::{future::BoxFuture, stream::BoxStream, AsyncBufReadExt, FutureExt, StreamExt};
13use gpui::{
14 AnyElement, AnyView, AppContext, AsyncAppContext, FontWeight, Model, ModelContext,
15 Subscription, Task,
16};
17use http_client::{AsyncBody, HttpClient, Method, Response};
18use schemars::JsonSchema;
19use serde::{Deserialize, Serialize};
20use serde_json::value::RawValue;
21use settings::{Settings, SettingsStore};
22use smol::{
23 io::{AsyncReadExt, BufReader},
24 lock::{RwLock, RwLockUpgradableReadGuard, RwLockWriteGuard},
25};
26use std::{future, sync::Arc};
27use strum::IntoEnumIterator;
28use ui::prelude::*;
29
30use crate::{LanguageModelAvailability, LanguageModelProvider};
31
32use super::anthropic::count_anthropic_tokens;
33
34pub const PROVIDER_ID: &str = "zed.dev";
35pub const PROVIDER_NAME: &str = "Zed";
36
37#[derive(Default, Clone, Debug, PartialEq)]
38pub struct ZedDotDevSettings {
39 pub available_models: Vec<AvailableModel>,
40}
41
42#[derive(Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]
43#[serde(rename_all = "lowercase")]
44pub enum AvailableProvider {
45 Anthropic,
46 OpenAi,
47 Google,
48}
49
50#[derive(Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]
51pub struct AvailableModel {
52 provider: AvailableProvider,
53 name: String,
54 max_tokens: usize,
55 tool_override: Option<String>,
56}
57
58pub struct CloudLanguageModelProvider {
59 client: Arc<Client>,
60 llm_api_token: LlmApiToken,
61 state: gpui::Model<State>,
62 _maintain_client_status: Task<()>,
63}
64
65pub struct State {
66 client: Arc<Client>,
67 user_store: Model<UserStore>,
68 status: client::Status,
69 accept_terms: Option<Task<Result<()>>>,
70 _subscription: Subscription,
71}
72
73impl State {
74 fn is_signed_out(&self) -> bool {
75 self.status.is_signed_out()
76 }
77
78 fn authenticate(&self, cx: &mut ModelContext<Self>) -> Task<Result<()>> {
79 let client = self.client.clone();
80 cx.spawn(move |this, mut cx| async move {
81 client.authenticate_and_connect(true, &cx).await?;
82 this.update(&mut cx, |_, cx| cx.notify())
83 })
84 }
85
86 fn has_accepted_terms_of_service(&self, cx: &AppContext) -> bool {
87 self.user_store
88 .read(cx)
89 .current_user_has_accepted_terms()
90 .unwrap_or(false)
91 }
92
93 fn accept_terms_of_service(&mut self, cx: &mut ModelContext<Self>) {
94 let user_store = self.user_store.clone();
95 self.accept_terms = Some(cx.spawn(move |this, mut cx| async move {
96 let _ = user_store
97 .update(&mut cx, |store, cx| store.accept_terms_of_service(cx))?
98 .await;
99 this.update(&mut cx, |this, cx| {
100 this.accept_terms = None;
101 cx.notify()
102 })
103 }));
104 }
105}
106
107impl CloudLanguageModelProvider {
108 pub fn new(user_store: Model<UserStore>, client: Arc<Client>, cx: &mut AppContext) -> Self {
109 let mut status_rx = client.status();
110 let status = *status_rx.borrow();
111
112 let state = cx.new_model(|cx| State {
113 client: client.clone(),
114 user_store,
115 status,
116 accept_terms: None,
117 _subscription: cx.observe_global::<SettingsStore>(|_, cx| {
118 cx.notify();
119 }),
120 });
121
122 let state_ref = state.downgrade();
123 let maintain_client_status = cx.spawn(|mut cx| async move {
124 while let Some(status) = status_rx.next().await {
125 if let Some(this) = state_ref.upgrade() {
126 _ = this.update(&mut cx, |this, cx| {
127 if this.status != status {
128 this.status = status;
129 cx.notify();
130 }
131 });
132 } else {
133 break;
134 }
135 }
136 });
137
138 Self {
139 client,
140 state,
141 llm_api_token: LlmApiToken::default(),
142 _maintain_client_status: maintain_client_status,
143 }
144 }
145}
146
147impl LanguageModelProviderState for CloudLanguageModelProvider {
148 type ObservableEntity = State;
149
150 fn observable_entity(&self) -> Option<gpui::Model<Self::ObservableEntity>> {
151 Some(self.state.clone())
152 }
153}
154
155impl LanguageModelProvider for CloudLanguageModelProvider {
156 fn id(&self) -> LanguageModelProviderId {
157 LanguageModelProviderId(PROVIDER_ID.into())
158 }
159
160 fn name(&self) -> LanguageModelProviderName {
161 LanguageModelProviderName(PROVIDER_NAME.into())
162 }
163
164 fn icon(&self) -> IconName {
165 IconName::AiZed
166 }
167
168 fn provided_models(&self, cx: &AppContext) -> Vec<Arc<dyn LanguageModel>> {
169 let mut models = BTreeMap::default();
170
171 if cx.is_staff() {
172 for model in anthropic::Model::iter() {
173 if !matches!(model, anthropic::Model::Custom { .. }) {
174 models.insert(model.id().to_string(), CloudModel::Anthropic(model));
175 }
176 }
177 for model in open_ai::Model::iter() {
178 if !matches!(model, open_ai::Model::Custom { .. }) {
179 models.insert(model.id().to_string(), CloudModel::OpenAi(model));
180 }
181 }
182 for model in google_ai::Model::iter() {
183 if !matches!(model, google_ai::Model::Custom { .. }) {
184 models.insert(model.id().to_string(), CloudModel::Google(model));
185 }
186 }
187 for model in ZedModel::iter() {
188 models.insert(model.id().to_string(), CloudModel::Zed(model));
189 }
190
191 // Override with available models from settings
192 for model in &AllLanguageModelSettings::get_global(cx)
193 .zed_dot_dev
194 .available_models
195 {
196 let model = match model.provider {
197 AvailableProvider::Anthropic => {
198 CloudModel::Anthropic(anthropic::Model::Custom {
199 name: model.name.clone(),
200 max_tokens: model.max_tokens,
201 tool_override: model.tool_override.clone(),
202 })
203 }
204 AvailableProvider::OpenAi => CloudModel::OpenAi(open_ai::Model::Custom {
205 name: model.name.clone(),
206 max_tokens: model.max_tokens,
207 }),
208 AvailableProvider::Google => CloudModel::Google(google_ai::Model::Custom {
209 name: model.name.clone(),
210 max_tokens: model.max_tokens,
211 }),
212 };
213 models.insert(model.id().to_string(), model.clone());
214 }
215 } else {
216 models.insert(
217 anthropic::Model::Claude3_5Sonnet.id().to_string(),
218 CloudModel::Anthropic(anthropic::Model::Claude3_5Sonnet),
219 );
220 }
221
222 models
223 .into_values()
224 .map(|model| {
225 Arc::new(CloudLanguageModel {
226 id: LanguageModelId::from(model.id().to_string()),
227 model,
228 llm_api_token: self.llm_api_token.clone(),
229 client: self.client.clone(),
230 request_limiter: RateLimiter::new(4),
231 }) as Arc<dyn LanguageModel>
232 })
233 .collect()
234 }
235
236 fn is_authenticated(&self, cx: &AppContext) -> bool {
237 !self.state.read(cx).is_signed_out()
238 }
239
240 fn authenticate(&self, _cx: &mut AppContext) -> Task<Result<()>> {
241 Task::ready(Ok(()))
242 }
243
244 fn configuration_view(&self, cx: &mut WindowContext) -> AnyView {
245 cx.new_view(|_cx| ConfigurationView {
246 state: self.state.clone(),
247 })
248 .into()
249 }
250
251 fn must_accept_terms(&self, cx: &AppContext) -> bool {
252 !self.state.read(cx).has_accepted_terms_of_service(cx)
253 }
254
255 fn render_accept_terms(&self, cx: &mut WindowContext) -> Option<AnyElement> {
256 let state = self.state.read(cx);
257
258 let terms = [(
259 "anthropic_terms_of_service",
260 "Anthropic Terms of Service",
261 "https://www.anthropic.com/legal/consumer-terms",
262 )]
263 .map(|(id, label, url)| {
264 Button::new(id, label)
265 .style(ButtonStyle::Subtle)
266 .icon(IconName::ExternalLink)
267 .icon_size(IconSize::XSmall)
268 .icon_color(Color::Muted)
269 .on_click(move |_, cx| cx.open_url(url))
270 });
271
272 if state.has_accepted_terms_of_service(cx) {
273 None
274 } else {
275 let disabled = state.accept_terms.is_some();
276 Some(
277 v_flex()
278 .child(Label::new("Terms & Conditions").weight(FontWeight::SEMIBOLD))
279 .child("Please read and accept the terms and conditions of Zed AI and our provider partners to continue.")
280 .child(v_flex().m_2().gap_1().children(terms))
281 .child(
282 h_flex().justify_end().mt_1().child(
283 Button::new("accept_terms", "Accept")
284 .disabled(disabled)
285 .on_click({
286 let state = self.state.downgrade();
287 move |_, cx| {
288 state
289 .update(cx, |state, cx| {
290 state.accept_terms_of_service(cx)
291 })
292 .ok();
293 }
294 }),
295 ),
296 )
297 .into_any(),
298 )
299 }
300 }
301
302 fn reset_credentials(&self, _cx: &mut AppContext) -> Task<Result<()>> {
303 Task::ready(Ok(()))
304 }
305}
306
307pub struct CloudLanguageModel {
308 id: LanguageModelId,
309 model: CloudModel,
310 llm_api_token: LlmApiToken,
311 client: Arc<Client>,
312 request_limiter: RateLimiter,
313}
314
315#[derive(Clone, Default)]
316struct LlmApiToken(Arc<RwLock<Option<String>>>);
317
318impl CloudLanguageModel {
319 async fn perform_llm_completion(
320 client: Arc<Client>,
321 llm_api_token: LlmApiToken,
322 body: PerformCompletionParams,
323 ) -> Result<Response<AsyncBody>> {
324 let http_client = &client.http_client();
325
326 let mut token = llm_api_token.acquire(&client).await?;
327 let mut did_retry = false;
328
329 let response = loop {
330 let request = http_client::Request::builder()
331 .method(Method::POST)
332 .uri(http_client.build_zed_llm_url("/completion", &[])?.as_ref())
333 .header("Content-Type", "application/json")
334 .header("Authorization", format!("Bearer {token}"))
335 .body(serde_json::to_string(&body)?.into())?;
336 let mut response = http_client.send(request).await?;
337 if response.status().is_success() {
338 break response;
339 } else if !did_retry
340 && response
341 .headers()
342 .get(EXPIRED_LLM_TOKEN_HEADER_NAME)
343 .is_some()
344 {
345 did_retry = true;
346 token = llm_api_token.refresh(&client).await?;
347 } else {
348 let mut body = String::new();
349 response.body_mut().read_to_string(&mut body).await?;
350 break Err(anyhow!(
351 "cloud language model completion failed with status {}: {body}",
352 response.status()
353 ))?;
354 }
355 };
356
357 Ok(response)
358 }
359}
360
361impl LanguageModel for CloudLanguageModel {
362 fn id(&self) -> LanguageModelId {
363 self.id.clone()
364 }
365
366 fn name(&self) -> LanguageModelName {
367 LanguageModelName::from(self.model.display_name().to_string())
368 }
369
370 fn provider_id(&self) -> LanguageModelProviderId {
371 LanguageModelProviderId(PROVIDER_ID.into())
372 }
373
374 fn provider_name(&self) -> LanguageModelProviderName {
375 LanguageModelProviderName(PROVIDER_NAME.into())
376 }
377
378 fn telemetry_id(&self) -> String {
379 format!("zed.dev/{}", self.model.id())
380 }
381
382 fn availability(&self) -> LanguageModelAvailability {
383 self.model.availability()
384 }
385
386 fn max_token_count(&self) -> usize {
387 self.model.max_token_count()
388 }
389
390 fn count_tokens(
391 &self,
392 request: LanguageModelRequest,
393 cx: &AppContext,
394 ) -> BoxFuture<'static, Result<usize>> {
395 match self.model.clone() {
396 CloudModel::Anthropic(_) => count_anthropic_tokens(request, cx),
397 CloudModel::OpenAi(model) => count_open_ai_tokens(request, model, cx),
398 CloudModel::Google(model) => {
399 let client = self.client.clone();
400 let request = request.into_google(model.id().into());
401 let request = google_ai::CountTokensRequest {
402 contents: request.contents,
403 };
404 async move {
405 let request = serde_json::to_string(&request)?;
406 let response = client
407 .request(proto::CountLanguageModelTokens {
408 provider: proto::LanguageModelProvider::Google as i32,
409 request,
410 })
411 .await?;
412 Ok(response.token_count as usize)
413 }
414 .boxed()
415 }
416 CloudModel::Zed(_) => {
417 count_open_ai_tokens(request, open_ai::Model::ThreePointFiveTurbo, cx)
418 }
419 }
420 }
421
422 fn stream_completion(
423 &self,
424 request: LanguageModelRequest,
425 _cx: &AsyncAppContext,
426 ) -> BoxFuture<'static, Result<BoxStream<'static, Result<String>>>> {
427 match &self.model {
428 CloudModel::Anthropic(model) => {
429 let request = request.into_anthropic(model.id().into());
430 let client = self.client.clone();
431 let llm_api_token = self.llm_api_token.clone();
432 let future = self.request_limiter.stream(async move {
433 let response = Self::perform_llm_completion(
434 client.clone(),
435 llm_api_token,
436 PerformCompletionParams {
437 provider: client::LanguageModelProvider::Anthropic,
438 model: request.model.clone(),
439 provider_request: RawValue::from_string(serde_json::to_string(
440 &request,
441 )?)?,
442 },
443 )
444 .await?;
445 let body = BufReader::new(response.into_body());
446 let stream = futures::stream::try_unfold(body, move |mut body| async move {
447 let mut buffer = String::new();
448 match body.read_line(&mut buffer).await {
449 Ok(0) => Ok(None),
450 Ok(_) => {
451 let event: anthropic::Event = serde_json::from_str(&buffer)
452 .context("failed to parse Anthropic event")?;
453 Ok(Some((event, body)))
454 }
455 Err(err) => Err(AnthropicError::Other(err.into())),
456 }
457 });
458
459 Ok(anthropic::extract_text_from_events(stream))
460 });
461 async move {
462 Ok(future
463 .await?
464 .map(|result| result.map_err(|err| anyhow!(err)))
465 .boxed())
466 }
467 .boxed()
468 }
469 CloudModel::OpenAi(model) => {
470 let client = self.client.clone();
471 let request = request.into_open_ai(model.id().into());
472 let llm_api_token = self.llm_api_token.clone();
473 let future = self.request_limiter.stream(async move {
474 let response = Self::perform_llm_completion(
475 client.clone(),
476 llm_api_token,
477 PerformCompletionParams {
478 provider: client::LanguageModelProvider::OpenAi,
479 model: request.model.clone(),
480 provider_request: RawValue::from_string(serde_json::to_string(
481 &request,
482 )?)?,
483 },
484 )
485 .await?;
486 let body = BufReader::new(response.into_body());
487 let stream = futures::stream::try_unfold(body, move |mut body| async move {
488 let mut buffer = String::new();
489 match body.read_line(&mut buffer).await {
490 Ok(0) => Ok(None),
491 Ok(_) => {
492 let event: open_ai::ResponseStreamEvent =
493 serde_json::from_str(&buffer)?;
494 Ok(Some((event, body)))
495 }
496 Err(e) => Err(e.into()),
497 }
498 });
499
500 Ok(open_ai::extract_text_from_events(stream))
501 });
502 async move { Ok(future.await?.boxed()) }.boxed()
503 }
504 CloudModel::Google(model) => {
505 let client = self.client.clone();
506 let request = request.into_google(model.id().into());
507 let llm_api_token = self.llm_api_token.clone();
508 let future = self.request_limiter.stream(async move {
509 let response = Self::perform_llm_completion(
510 client.clone(),
511 llm_api_token,
512 PerformCompletionParams {
513 provider: client::LanguageModelProvider::Google,
514 model: request.model.clone(),
515 provider_request: RawValue::from_string(serde_json::to_string(
516 &request,
517 )?)?,
518 },
519 )
520 .await?;
521 let body = BufReader::new(response.into_body());
522 let stream = futures::stream::try_unfold(body, move |mut body| async move {
523 let mut buffer = String::new();
524 match body.read_line(&mut buffer).await {
525 Ok(0) => Ok(None),
526 Ok(_) => {
527 let event: google_ai::GenerateContentResponse =
528 serde_json::from_str(&buffer)?;
529 Ok(Some((event, body)))
530 }
531 Err(e) => Err(e.into()),
532 }
533 });
534
535 Ok(google_ai::extract_text_from_events(stream))
536 });
537 async move { Ok(future.await?.boxed()) }.boxed()
538 }
539 CloudModel::Zed(model) => {
540 let client = self.client.clone();
541 let mut request = request.into_open_ai(model.id().into());
542 request.max_tokens = Some(4000);
543 let llm_api_token = self.llm_api_token.clone();
544 let future = self.request_limiter.stream(async move {
545 let response = Self::perform_llm_completion(
546 client.clone(),
547 llm_api_token,
548 PerformCompletionParams {
549 provider: client::LanguageModelProvider::Zed,
550 model: request.model.clone(),
551 provider_request: RawValue::from_string(serde_json::to_string(
552 &request,
553 )?)?,
554 },
555 )
556 .await?;
557 let body = BufReader::new(response.into_body());
558 let stream = futures::stream::try_unfold(body, move |mut body| async move {
559 let mut buffer = String::new();
560 match body.read_line(&mut buffer).await {
561 Ok(0) => Ok(None),
562 Ok(_) => {
563 let event: open_ai::ResponseStreamEvent =
564 serde_json::from_str(&buffer)?;
565 Ok(Some((event, body)))
566 }
567 Err(e) => Err(e.into()),
568 }
569 });
570
571 Ok(open_ai::extract_text_from_events(stream))
572 });
573 async move { Ok(future.await?.boxed()) }.boxed()
574 }
575 }
576 }
577
578 fn use_any_tool(
579 &self,
580 request: LanguageModelRequest,
581 tool_name: String,
582 tool_description: String,
583 input_schema: serde_json::Value,
584 _cx: &AsyncAppContext,
585 ) -> BoxFuture<'static, Result<serde_json::Value>> {
586 match &self.model {
587 CloudModel::Anthropic(model) => {
588 let client = self.client.clone();
589 let mut request = request.into_anthropic(model.tool_model_id().into());
590 request.tool_choice = Some(anthropic::ToolChoice::Tool {
591 name: tool_name.clone(),
592 });
593 request.tools = vec![anthropic::Tool {
594 name: tool_name.clone(),
595 description: tool_description,
596 input_schema,
597 }];
598
599 let llm_api_token = self.llm_api_token.clone();
600 self.request_limiter
601 .run(async move {
602 let response = Self::perform_llm_completion(
603 client.clone(),
604 llm_api_token,
605 PerformCompletionParams {
606 provider: client::LanguageModelProvider::Anthropic,
607 model: request.model.clone(),
608 provider_request: RawValue::from_string(serde_json::to_string(
609 &request,
610 )?)?,
611 },
612 )
613 .await?;
614
615 let mut tool_use_index = None;
616 let mut tool_input = String::new();
617 let mut body = BufReader::new(response.into_body());
618 let mut line = String::new();
619 while body.read_line(&mut line).await? > 0 {
620 let event: anthropic::Event = serde_json::from_str(&line)?;
621 line.clear();
622
623 match event {
624 anthropic::Event::ContentBlockStart {
625 content_block,
626 index,
627 } => {
628 if let anthropic::Content::ToolUse { name, .. } = content_block
629 {
630 if name == tool_name {
631 tool_use_index = Some(index);
632 }
633 }
634 }
635 anthropic::Event::ContentBlockDelta { index, delta } => match delta
636 {
637 anthropic::ContentDelta::TextDelta { .. } => {}
638 anthropic::ContentDelta::InputJsonDelta { partial_json } => {
639 if Some(index) == tool_use_index {
640 tool_input.push_str(&partial_json);
641 }
642 }
643 },
644 anthropic::Event::ContentBlockStop { index } => {
645 if Some(index) == tool_use_index {
646 return Ok(serde_json::from_str(&tool_input)?);
647 }
648 }
649 _ => {}
650 }
651 }
652
653 if tool_use_index.is_some() {
654 Err(anyhow!("tool content incomplete"))
655 } else {
656 Err(anyhow!("tool not used"))
657 }
658 })
659 .boxed()
660 }
661 CloudModel::OpenAi(model) => {
662 let mut request = request.into_open_ai(model.id().into());
663 let client = self.client.clone();
664 let mut function = open_ai::FunctionDefinition {
665 name: tool_name.clone(),
666 description: None,
667 parameters: None,
668 };
669 let func = open_ai::ToolDefinition::Function {
670 function: function.clone(),
671 };
672 request.tool_choice = Some(open_ai::ToolChoice::Other(func.clone()));
673 // Fill in description and params separately, as they're not needed for tool_choice field.
674 function.description = Some(tool_description);
675 function.parameters = Some(input_schema);
676 request.tools = vec![open_ai::ToolDefinition::Function { function }];
677
678 let llm_api_token = self.llm_api_token.clone();
679 self.request_limiter
680 .run(async move {
681 let response = Self::perform_llm_completion(
682 client.clone(),
683 llm_api_token,
684 PerformCompletionParams {
685 provider: client::LanguageModelProvider::OpenAi,
686 model: request.model.clone(),
687 provider_request: RawValue::from_string(serde_json::to_string(
688 &request,
689 )?)?,
690 },
691 )
692 .await?;
693
694 let mut body = BufReader::new(response.into_body());
695 let mut line = String::new();
696 let mut load_state = None;
697
698 while body.read_line(&mut line).await? > 0 {
699 let part: open_ai::ResponseStreamEvent = serde_json::from_str(&line)?;
700 line.clear();
701
702 for choice in part.choices {
703 let Some(tool_calls) = choice.delta.tool_calls else {
704 continue;
705 };
706
707 for call in tool_calls {
708 if let Some(func) = call.function {
709 if func.name.as_deref() == Some(tool_name.as_str()) {
710 load_state = Some((String::default(), call.index));
711 }
712 if let Some((arguments, (output, index))) =
713 func.arguments.zip(load_state.as_mut())
714 {
715 if call.index == *index {
716 output.push_str(&arguments);
717 }
718 }
719 }
720 }
721 }
722 }
723
724 if let Some((arguments, _)) = load_state {
725 return Ok(serde_json::from_str(&arguments)?);
726 } else {
727 bail!("tool not used");
728 }
729 })
730 .boxed()
731 }
732 CloudModel::Google(_) => {
733 future::ready(Err(anyhow!("tool use not implemented for Google AI"))).boxed()
734 }
735 CloudModel::Zed(model) => {
736 // All Zed models are OpenAI-based at the time of writing.
737 let mut request = request.into_open_ai(model.id().into());
738 let client = self.client.clone();
739 let mut function = open_ai::FunctionDefinition {
740 name: tool_name.clone(),
741 description: None,
742 parameters: None,
743 };
744 let func = open_ai::ToolDefinition::Function {
745 function: function.clone(),
746 };
747 request.tool_choice = Some(open_ai::ToolChoice::Other(func.clone()));
748 // Fill in description and params separately, as they're not needed for tool_choice field.
749 function.description = Some(tool_description);
750 function.parameters = Some(input_schema);
751 request.tools = vec![open_ai::ToolDefinition::Function { function }];
752
753 let llm_api_token = self.llm_api_token.clone();
754 self.request_limiter
755 .run(async move {
756 let response = Self::perform_llm_completion(
757 client.clone(),
758 llm_api_token,
759 PerformCompletionParams {
760 provider: client::LanguageModelProvider::Zed,
761 model: request.model.clone(),
762 provider_request: RawValue::from_string(serde_json::to_string(
763 &request,
764 )?)?,
765 },
766 )
767 .await?;
768
769 let mut body = BufReader::new(response.into_body());
770 let mut line = String::new();
771 let mut load_state = None;
772
773 while body.read_line(&mut line).await? > 0 {
774 let part: open_ai::ResponseStreamEvent = serde_json::from_str(&line)?;
775 line.clear();
776
777 for choice in part.choices {
778 let Some(tool_calls) = choice.delta.tool_calls else {
779 continue;
780 };
781
782 for call in tool_calls {
783 if let Some(func) = call.function {
784 if func.name.as_deref() == Some(tool_name.as_str()) {
785 load_state = Some((String::default(), call.index));
786 }
787 if let Some((arguments, (output, index))) =
788 func.arguments.zip(load_state.as_mut())
789 {
790 if call.index == *index {
791 output.push_str(&arguments);
792 }
793 }
794 }
795 }
796 }
797 }
798 if let Some((arguments, _)) = load_state {
799 return Ok(serde_json::from_str(&arguments)?);
800 } else {
801 bail!("tool not used");
802 }
803 })
804 .boxed()
805 }
806 }
807 }
808}
809
810impl LlmApiToken {
811 async fn acquire(&self, client: &Arc<Client>) -> Result<String> {
812 let lock = self.0.upgradable_read().await;
813 if let Some(token) = lock.as_ref() {
814 Ok(token.to_string())
815 } else {
816 Self::fetch(RwLockUpgradableReadGuard::upgrade(lock).await, &client).await
817 }
818 }
819
820 async fn refresh(&self, client: &Arc<Client>) -> Result<String> {
821 Self::fetch(self.0.write().await, &client).await
822 }
823
824 async fn fetch<'a>(
825 mut lock: RwLockWriteGuard<'a, Option<String>>,
826 client: &Arc<Client>,
827 ) -> Result<String> {
828 let response = client.request(proto::GetLlmToken {}).await?;
829 *lock = Some(response.token.clone());
830 Ok(response.token.clone())
831 }
832}
833
834struct ConfigurationView {
835 state: gpui::Model<State>,
836}
837
838impl ConfigurationView {
839 fn authenticate(&mut self, cx: &mut ViewContext<Self>) {
840 self.state.update(cx, |state, cx| {
841 state.authenticate(cx).detach_and_log_err(cx);
842 });
843 cx.notify();
844 }
845}
846
847impl Render for ConfigurationView {
848 fn render(&mut self, cx: &mut ViewContext<Self>) -> impl IntoElement {
849 const ZED_AI_URL: &str = "https://zed.dev/ai";
850 const ACCOUNT_SETTINGS_URL: &str = "https://zed.dev/account";
851
852 let is_connected = !self.state.read(cx).is_signed_out();
853 let plan = self.state.read(cx).user_store.read(cx).current_plan();
854 let must_accept_terms = !self.state.read(cx).has_accepted_terms_of_service(cx);
855
856 let is_pro = plan == Some(proto::Plan::ZedPro);
857
858 if is_connected {
859 v_flex()
860 .gap_3()
861 .max_w_4_5()
862 .when(must_accept_terms, |this| {
863 this.child(Label::new(
864 "You must accept the terms of service to use this provider.",
865 ))
866 })
867 .child(Label::new(
868 if is_pro {
869 "You have full access to Zed's hosted models from Anthropic, OpenAI, Google with faster speeds and higher limits through Zed Pro."
870 } else {
871 "You have basic access to models from Anthropic through the Zed AI Free plan."
872 }))
873 .children(if is_pro {
874 Some(
875 h_flex().child(
876 Button::new("manage_settings", "Manage Subscription")
877 .style(ButtonStyle::Filled)
878 .on_click(
879 cx.listener(|_, _, cx| cx.open_url(ACCOUNT_SETTINGS_URL)),
880 ),
881 ),
882 )
883 } else if cx.has_flag::<ZedPro>() {
884 Some(
885 h_flex()
886 .gap_2()
887 .child(
888 Button::new("learn_more", "Learn more")
889 .style(ButtonStyle::Subtle)
890 .on_click(cx.listener(|_, _, cx| cx.open_url(ZED_AI_URL))),
891 )
892 .child(
893 Button::new("upgrade", "Upgrade")
894 .style(ButtonStyle::Subtle)
895 .color(Color::Accent)
896 .on_click(
897 cx.listener(|_, _, cx| cx.open_url(ACCOUNT_SETTINGS_URL)),
898 ),
899 ),
900 )
901 } else {
902 None
903 })
904 } else {
905 v_flex()
906 .gap_6()
907 .child(Label::new("Use the zed.dev to access language models."))
908 .child(
909 v_flex()
910 .gap_2()
911 .child(
912 Button::new("sign_in", "Sign in")
913 .icon_color(Color::Muted)
914 .icon(IconName::Github)
915 .icon_position(IconPosition::Start)
916 .style(ButtonStyle::Filled)
917 .full_width()
918 .on_click(cx.listener(move |this, _, cx| this.authenticate(cx))),
919 )
920 .child(
921 div().flex().w_full().items_center().child(
922 Label::new("Sign in to enable collaboration.")
923 .color(Color::Muted)
924 .size(LabelSize::Small),
925 ),
926 ),
927 )
928 }
929 }
930}