1use crate::AllLanguageModelSettings;
2use crate::ui::InstructionListItem;
3use anthropic::{
4 AnthropicError, AnthropicModelMode, ContentDelta, Event, ResponseContent, ToolResultContent,
5 ToolResultPart, Usage,
6};
7use anyhow::{Context as _, Result, anyhow};
8use collections::{BTreeMap, HashMap};
9use credentials_provider::CredentialsProvider;
10use editor::{Editor, EditorElement, EditorStyle};
11use futures::Stream;
12use futures::{FutureExt, StreamExt, future::BoxFuture, stream::BoxStream};
13use gpui::{
14 AnyView, App, AsyncApp, Context, Entity, FontStyle, Subscription, Task, TextStyle, WhiteSpace,
15};
16use http_client::HttpClient;
17use language_model::{
18 AuthenticateError, ConfigurationViewTargetAgent, LanguageModel,
19 LanguageModelCacheConfiguration, LanguageModelCompletionError, LanguageModelId,
20 LanguageModelName, LanguageModelProvider, LanguageModelProviderId, LanguageModelProviderName,
21 LanguageModelProviderState, LanguageModelRequest, LanguageModelToolChoice,
22 LanguageModelToolResultContent, MessageContent, RateLimiter, Role,
23};
24use language_model::{LanguageModelCompletionEvent, LanguageModelToolUse, StopReason};
25use schemars::JsonSchema;
26use serde::{Deserialize, Serialize};
27use settings::{Settings, SettingsStore};
28use std::pin::Pin;
29use std::str::FromStr;
30use std::sync::Arc;
31use strum::IntoEnumIterator;
32use theme::ThemeSettings;
33use ui::{Icon, IconName, List, Tooltip, prelude::*};
34use util::ResultExt;
35
36const PROVIDER_ID: LanguageModelProviderId = language_model::ANTHROPIC_PROVIDER_ID;
37const PROVIDER_NAME: LanguageModelProviderName = language_model::ANTHROPIC_PROVIDER_NAME;
38
39#[derive(Default, Clone, Debug, PartialEq)]
40pub struct AnthropicSettings {
41 pub api_url: String,
42 /// Extend Zed's list of Anthropic models.
43 pub available_models: Vec<AvailableModel>,
44}
45
46#[derive(Clone, Debug, PartialEq, Serialize, Deserialize, JsonSchema)]
47pub struct AvailableModel {
48 /// The model's name in the Anthropic API. e.g. claude-3-5-sonnet-latest, claude-3-opus-20240229, etc
49 pub name: String,
50 /// The model's name in Zed's UI, such as in the model selector dropdown menu in the assistant panel.
51 pub display_name: Option<String>,
52 /// The model's context window size.
53 pub max_tokens: u64,
54 /// A model `name` to substitute when calling tools, in case the primary model doesn't support tool calling.
55 pub tool_override: Option<String>,
56 /// Configuration of Anthropic's caching API.
57 pub cache_configuration: Option<LanguageModelCacheConfiguration>,
58 pub max_output_tokens: Option<u64>,
59 pub default_temperature: Option<f32>,
60 #[serde(default)]
61 pub extra_beta_headers: Vec<String>,
62 /// The model's mode (e.g. thinking)
63 pub mode: Option<ModelMode>,
64}
65
66#[derive(Clone, Debug, Default, 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
86impl From<AnthropicModelMode> for ModelMode {
87 fn from(value: AnthropicModelMode) -> Self {
88 match value {
89 AnthropicModelMode::Default => ModelMode::Default,
90 AnthropicModelMode::Thinking { budget_tokens } => ModelMode::Thinking { budget_tokens },
91 }
92 }
93}
94
95pub struct AnthropicLanguageModelProvider {
96 http_client: Arc<dyn HttpClient>,
97 state: gpui::Entity<State>,
98}
99
100const ANTHROPIC_API_KEY_VAR: &str = "ANTHROPIC_API_KEY";
101
102pub struct State {
103 api_key: Option<String>,
104 api_key_from_env: bool,
105 _subscription: Subscription,
106}
107
108impl State {
109 fn reset_api_key(&self, cx: &mut Context<Self>) -> Task<Result<()>> {
110 let credentials_provider = <dyn CredentialsProvider>::global(cx);
111 let api_url = AllLanguageModelSettings::get_global(cx)
112 .anthropic
113 .api_url
114 .clone();
115 cx.spawn(async move |this, cx| {
116 credentials_provider
117 .delete_credentials(&api_url, cx)
118 .await
119 .ok();
120 this.update(cx, |this, cx| {
121 this.api_key = None;
122 this.api_key_from_env = false;
123 cx.notify();
124 })
125 })
126 }
127
128 fn set_api_key(&mut self, api_key: String, cx: &mut Context<Self>) -> Task<Result<()>> {
129 let credentials_provider = <dyn CredentialsProvider>::global(cx);
130 let api_url = AllLanguageModelSettings::get_global(cx)
131 .anthropic
132 .api_url
133 .clone();
134 cx.spawn(async move |this, cx| {
135 credentials_provider
136 .write_credentials(&api_url, "Bearer", api_key.as_bytes(), cx)
137 .await
138 .ok();
139
140 this.update(cx, |this, cx| {
141 this.api_key = Some(api_key);
142 cx.notify();
143 })
144 })
145 }
146
147 fn is_authenticated(&self) -> bool {
148 self.api_key.is_some()
149 }
150
151 fn authenticate(&self, cx: &mut Context<Self>) -> Task<Result<(), AuthenticateError>> {
152 if self.is_authenticated() {
153 return Task::ready(Ok(()));
154 }
155
156 let key = AnthropicLanguageModelProvider::api_key(cx);
157
158 cx.spawn(async move |this, cx| {
159 let key = key.await?;
160
161 this.update(cx, |this, cx| {
162 this.api_key = Some(key.key);
163 this.api_key_from_env = key.from_env;
164 cx.notify();
165 })?;
166
167 Ok(())
168 })
169 }
170}
171
172pub struct ApiKey {
173 pub key: String,
174 pub from_env: bool,
175}
176
177impl AnthropicLanguageModelProvider {
178 pub fn new(http_client: Arc<dyn HttpClient>, cx: &mut App) -> Self {
179 let state = cx.new(|cx| State {
180 api_key: None,
181 api_key_from_env: false,
182 _subscription: cx.observe_global::<SettingsStore>(|_, cx| {
183 cx.notify();
184 }),
185 });
186
187 Self { http_client, state }
188 }
189
190 fn create_language_model(&self, model: anthropic::Model) -> Arc<dyn LanguageModel> {
191 Arc::new(AnthropicModel {
192 id: LanguageModelId::from(model.id().to_string()),
193 model,
194 state: self.state.clone(),
195 http_client: self.http_client.clone(),
196 request_limiter: RateLimiter::new(4),
197 })
198 }
199
200 pub fn api_key(cx: &mut App) -> Task<Result<ApiKey, AuthenticateError>> {
201 let credentials_provider = <dyn CredentialsProvider>::global(cx);
202 let api_url = AllLanguageModelSettings::get_global(cx)
203 .anthropic
204 .api_url
205 .clone();
206
207 if let Ok(key) = std::env::var(ANTHROPIC_API_KEY_VAR) {
208 Task::ready(Ok(ApiKey {
209 key,
210 from_env: true,
211 }))
212 } else {
213 cx.spawn(async move |cx| {
214 let (_, api_key) = credentials_provider
215 .read_credentials(&api_url, cx)
216 .await?
217 .ok_or(AuthenticateError::CredentialsNotFound)?;
218
219 Ok(ApiKey {
220 key: String::from_utf8(api_key).context("invalid {PROVIDER_NAME} API key")?,
221 from_env: false,
222 })
223 })
224 }
225 }
226}
227
228impl LanguageModelProviderState for AnthropicLanguageModelProvider {
229 type ObservableEntity = State;
230
231 fn observable_entity(&self) -> Option<gpui::Entity<Self::ObservableEntity>> {
232 Some(self.state.clone())
233 }
234}
235
236impl LanguageModelProvider for AnthropicLanguageModelProvider {
237 fn id(&self) -> LanguageModelProviderId {
238 PROVIDER_ID
239 }
240
241 fn name(&self) -> LanguageModelProviderName {
242 PROVIDER_NAME
243 }
244
245 fn icon(&self) -> IconName {
246 IconName::AiAnthropic
247 }
248
249 fn default_model(&self, _cx: &App) -> Option<Arc<dyn LanguageModel>> {
250 Some(self.create_language_model(anthropic::Model::default()))
251 }
252
253 fn default_fast_model(&self, _cx: &App) -> Option<Arc<dyn LanguageModel>> {
254 Some(self.create_language_model(anthropic::Model::default_fast()))
255 }
256
257 fn recommended_models(&self, _cx: &App) -> Vec<Arc<dyn LanguageModel>> {
258 [
259 anthropic::Model::ClaudeSonnet4,
260 anthropic::Model::ClaudeSonnet4Thinking,
261 ]
262 .into_iter()
263 .map(|model| self.create_language_model(model))
264 .collect()
265 }
266
267 fn provided_models(&self, cx: &App) -> Vec<Arc<dyn LanguageModel>> {
268 let mut models = BTreeMap::default();
269
270 // Add base models from anthropic::Model::iter()
271 for model in anthropic::Model::iter() {
272 if !matches!(model, anthropic::Model::Custom { .. }) {
273 models.insert(model.id().to_string(), model);
274 }
275 }
276
277 // Override with available models from settings
278 for model in AllLanguageModelSettings::get_global(cx)
279 .anthropic
280 .available_models
281 .iter()
282 {
283 models.insert(
284 model.name.clone(),
285 anthropic::Model::Custom {
286 name: model.name.clone(),
287 display_name: model.display_name.clone(),
288 max_tokens: model.max_tokens,
289 tool_override: model.tool_override.clone(),
290 cache_configuration: model.cache_configuration.as_ref().map(|config| {
291 anthropic::AnthropicModelCacheConfiguration {
292 max_cache_anchors: config.max_cache_anchors,
293 should_speculate: config.should_speculate,
294 min_total_token: config.min_total_token,
295 }
296 }),
297 max_output_tokens: model.max_output_tokens,
298 default_temperature: model.default_temperature,
299 extra_beta_headers: model.extra_beta_headers.clone(),
300 mode: model.mode.clone().unwrap_or_default().into(),
301 },
302 );
303 }
304
305 models
306 .into_values()
307 .map(|model| self.create_language_model(model))
308 .collect()
309 }
310
311 fn is_authenticated(&self, cx: &App) -> bool {
312 self.state.read(cx).is_authenticated()
313 }
314
315 fn authenticate(&self, cx: &mut App) -> Task<Result<(), AuthenticateError>> {
316 self.state.update(cx, |state, cx| state.authenticate(cx))
317 }
318
319 fn configuration_view(
320 &self,
321 target_agent: ConfigurationViewTargetAgent,
322 window: &mut Window,
323 cx: &mut App,
324 ) -> AnyView {
325 cx.new(|cx| ConfigurationView::new(self.state.clone(), target_agent, window, cx))
326 .into()
327 }
328
329 fn reset_credentials(&self, cx: &mut App) -> Task<Result<()>> {
330 self.state.update(cx, |state, cx| state.reset_api_key(cx))
331 }
332}
333
334pub struct AnthropicModel {
335 id: LanguageModelId,
336 model: anthropic::Model,
337 state: gpui::Entity<State>,
338 http_client: Arc<dyn HttpClient>,
339 request_limiter: RateLimiter,
340}
341
342pub fn count_anthropic_tokens(
343 request: LanguageModelRequest,
344 cx: &App,
345) -> BoxFuture<'static, Result<u64>> {
346 cx.background_spawn(async move {
347 let messages = request.messages;
348 let mut tokens_from_images = 0;
349 let mut string_messages = Vec::with_capacity(messages.len());
350
351 for message in messages {
352 use language_model::MessageContent;
353
354 let mut string_contents = String::new();
355
356 for content in message.content {
357 match content {
358 MessageContent::Text(text) => {
359 string_contents.push_str(&text);
360 }
361 MessageContent::Thinking { .. } => {
362 // Thinking blocks are not included in the input token count.
363 }
364 MessageContent::RedactedThinking(_) => {
365 // Thinking blocks are not included in the input token count.
366 }
367 MessageContent::Image(image) => {
368 tokens_from_images += image.estimate_tokens();
369 }
370 MessageContent::ToolUse(_tool_use) => {
371 // TODO: Estimate token usage from tool uses.
372 }
373 MessageContent::ToolResult(tool_result) => match &tool_result.content {
374 LanguageModelToolResultContent::Text(text) => {
375 string_contents.push_str(text);
376 }
377 LanguageModelToolResultContent::Image(image) => {
378 tokens_from_images += image.estimate_tokens();
379 }
380 },
381 }
382 }
383
384 if !string_contents.is_empty() {
385 string_messages.push(tiktoken_rs::ChatCompletionRequestMessage {
386 role: match message.role {
387 Role::User => "user".into(),
388 Role::Assistant => "assistant".into(),
389 Role::System => "system".into(),
390 },
391 content: Some(string_contents),
392 name: None,
393 function_call: None,
394 });
395 }
396 }
397
398 // Tiktoken doesn't yet support these models, so we manually use the
399 // same tokenizer as GPT-4.
400 tiktoken_rs::num_tokens_from_messages("gpt-4", &string_messages)
401 .map(|tokens| (tokens + tokens_from_images) as u64)
402 })
403 .boxed()
404}
405
406impl AnthropicModel {
407 fn stream_completion(
408 &self,
409 request: anthropic::Request,
410 cx: &AsyncApp,
411 ) -> BoxFuture<
412 'static,
413 Result<
414 BoxStream<'static, Result<anthropic::Event, AnthropicError>>,
415 LanguageModelCompletionError,
416 >,
417 > {
418 let http_client = self.http_client.clone();
419
420 let Ok((api_key, api_url)) = cx.read_entity(&self.state, |state, cx| {
421 let settings = &AllLanguageModelSettings::get_global(cx).anthropic;
422 (state.api_key.clone(), settings.api_url.clone())
423 }) else {
424 return futures::future::ready(Err(anyhow!("App state dropped").into())).boxed();
425 };
426
427 async move {
428 let Some(api_key) = api_key else {
429 return Err(LanguageModelCompletionError::NoApiKey {
430 provider: PROVIDER_NAME,
431 });
432 };
433 let request =
434 anthropic::stream_completion(http_client.as_ref(), &api_url, &api_key, request);
435 request.await.map_err(Into::into)
436 }
437 .boxed()
438 }
439}
440
441impl LanguageModel for AnthropicModel {
442 fn id(&self) -> LanguageModelId {
443 self.id.clone()
444 }
445
446 fn name(&self) -> LanguageModelName {
447 LanguageModelName::from(self.model.display_name().to_string())
448 }
449
450 fn provider_id(&self) -> LanguageModelProviderId {
451 PROVIDER_ID
452 }
453
454 fn provider_name(&self) -> LanguageModelProviderName {
455 PROVIDER_NAME
456 }
457
458 fn supports_tools(&self) -> bool {
459 true
460 }
461
462 fn supports_images(&self) -> bool {
463 true
464 }
465
466 fn supports_tool_choice(&self, choice: LanguageModelToolChoice) -> bool {
467 match choice {
468 LanguageModelToolChoice::Auto
469 | LanguageModelToolChoice::Any
470 | LanguageModelToolChoice::None => true,
471 }
472 }
473
474 fn telemetry_id(&self) -> String {
475 format!("anthropic/{}", self.model.id())
476 }
477
478 fn api_key(&self, cx: &App) -> Option<String> {
479 self.state.read(cx).api_key.clone()
480 }
481
482 fn max_token_count(&self) -> u64 {
483 self.model.max_token_count()
484 }
485
486 fn max_output_tokens(&self) -> Option<u64> {
487 Some(self.model.max_output_tokens())
488 }
489
490 fn count_tokens(
491 &self,
492 request: LanguageModelRequest,
493 cx: &App,
494 ) -> BoxFuture<'static, Result<u64>> {
495 count_anthropic_tokens(request, cx)
496 }
497
498 fn stream_completion(
499 &self,
500 request: LanguageModelRequest,
501 cx: &AsyncApp,
502 ) -> BoxFuture<
503 'static,
504 Result<
505 BoxStream<'static, Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>,
506 LanguageModelCompletionError,
507 >,
508 > {
509 let request = into_anthropic(
510 request,
511 self.model.request_id().into(),
512 self.model.default_temperature(),
513 self.model.max_output_tokens(),
514 self.model.mode(),
515 );
516 let request = self.stream_completion(request, cx);
517 let future = self.request_limiter.stream(async move {
518 let response = request.await?;
519 Ok(AnthropicEventMapper::new().map_stream(response))
520 });
521 async move { Ok(future.await?.boxed()) }.boxed()
522 }
523
524 fn cache_configuration(&self) -> Option<LanguageModelCacheConfiguration> {
525 self.model
526 .cache_configuration()
527 .map(|config| LanguageModelCacheConfiguration {
528 max_cache_anchors: config.max_cache_anchors,
529 should_speculate: config.should_speculate,
530 min_total_token: config.min_total_token,
531 })
532 }
533}
534
535pub fn into_anthropic(
536 request: LanguageModelRequest,
537 model: String,
538 default_temperature: f32,
539 max_output_tokens: u64,
540 mode: AnthropicModelMode,
541) -> anthropic::Request {
542 let mut new_messages: Vec<anthropic::Message> = Vec::new();
543 let mut system_message = String::new();
544
545 for message in request.messages {
546 if message.contents_empty() {
547 continue;
548 }
549
550 match message.role {
551 Role::User | Role::Assistant => {
552 let mut anthropic_message_content: Vec<anthropic::RequestContent> = message
553 .content
554 .into_iter()
555 .filter_map(|content| match content {
556 MessageContent::Text(text) => {
557 let text = if text.chars().last().is_some_and(|c| c.is_whitespace()) {
558 text.trim_end().to_string()
559 } else {
560 text
561 };
562 if !text.is_empty() {
563 Some(anthropic::RequestContent::Text {
564 text,
565 cache_control: None,
566 })
567 } else {
568 None
569 }
570 }
571 MessageContent::Thinking {
572 text: thinking,
573 signature,
574 } => {
575 if !thinking.is_empty() {
576 Some(anthropic::RequestContent::Thinking {
577 thinking,
578 signature: signature.unwrap_or_default(),
579 cache_control: None,
580 })
581 } else {
582 None
583 }
584 }
585 MessageContent::RedactedThinking(data) => {
586 if !data.is_empty() {
587 Some(anthropic::RequestContent::RedactedThinking { data })
588 } else {
589 None
590 }
591 }
592 MessageContent::Image(image) => Some(anthropic::RequestContent::Image {
593 source: anthropic::ImageSource {
594 source_type: "base64".to_string(),
595 media_type: "image/png".to_string(),
596 data: image.source.to_string(),
597 },
598 cache_control: None,
599 }),
600 MessageContent::ToolUse(tool_use) => {
601 Some(anthropic::RequestContent::ToolUse {
602 id: tool_use.id.to_string(),
603 name: tool_use.name.to_string(),
604 input: tool_use.input,
605 cache_control: None,
606 })
607 }
608 MessageContent::ToolResult(tool_result) => {
609 Some(anthropic::RequestContent::ToolResult {
610 tool_use_id: tool_result.tool_use_id.to_string(),
611 is_error: tool_result.is_error,
612 content: match tool_result.content {
613 LanguageModelToolResultContent::Text(text) => {
614 ToolResultContent::Plain(text.to_string())
615 }
616 LanguageModelToolResultContent::Image(image) => {
617 ToolResultContent::Multipart(vec![ToolResultPart::Image {
618 source: anthropic::ImageSource {
619 source_type: "base64".to_string(),
620 media_type: "image/png".to_string(),
621 data: image.source.to_string(),
622 },
623 }])
624 }
625 },
626 cache_control: None,
627 })
628 }
629 })
630 .collect();
631 let anthropic_role = match message.role {
632 Role::User => anthropic::Role::User,
633 Role::Assistant => anthropic::Role::Assistant,
634 Role::System => unreachable!("System role should never occur here"),
635 };
636 if let Some(last_message) = new_messages.last_mut()
637 && last_message.role == anthropic_role
638 {
639 last_message.content.extend(anthropic_message_content);
640 continue;
641 }
642
643 // Mark the last segment of the message as cached
644 if message.cache {
645 let cache_control_value = Some(anthropic::CacheControl {
646 cache_type: anthropic::CacheControlType::Ephemeral,
647 });
648 for message_content in anthropic_message_content.iter_mut().rev() {
649 match message_content {
650 anthropic::RequestContent::RedactedThinking { .. } => {
651 // Caching is not possible, fallback to next message
652 }
653 anthropic::RequestContent::Text { cache_control, .. }
654 | anthropic::RequestContent::Thinking { cache_control, .. }
655 | anthropic::RequestContent::Image { cache_control, .. }
656 | anthropic::RequestContent::ToolUse { cache_control, .. }
657 | anthropic::RequestContent::ToolResult { cache_control, .. } => {
658 *cache_control = cache_control_value;
659 break;
660 }
661 }
662 }
663 }
664
665 new_messages.push(anthropic::Message {
666 role: anthropic_role,
667 content: anthropic_message_content,
668 });
669 }
670 Role::System => {
671 if !system_message.is_empty() {
672 system_message.push_str("\n\n");
673 }
674 system_message.push_str(&message.string_contents());
675 }
676 }
677 }
678
679 anthropic::Request {
680 model,
681 messages: new_messages,
682 max_tokens: max_output_tokens,
683 system: if system_message.is_empty() {
684 None
685 } else {
686 Some(anthropic::StringOrContents::String(system_message))
687 },
688 thinking: if request.thinking_allowed
689 && let AnthropicModelMode::Thinking { budget_tokens } = mode
690 {
691 Some(anthropic::Thinking::Enabled { budget_tokens })
692 } else {
693 None
694 },
695 tools: request
696 .tools
697 .into_iter()
698 .map(|tool| anthropic::Tool {
699 name: tool.name,
700 description: tool.description,
701 input_schema: tool.input_schema,
702 })
703 .collect(),
704 tool_choice: request.tool_choice.map(|choice| match choice {
705 LanguageModelToolChoice::Auto => anthropic::ToolChoice::Auto,
706 LanguageModelToolChoice::Any => anthropic::ToolChoice::Any,
707 LanguageModelToolChoice::None => anthropic::ToolChoice::None,
708 }),
709 metadata: None,
710 stop_sequences: Vec::new(),
711 temperature: request.temperature.or(Some(default_temperature)),
712 top_k: None,
713 top_p: None,
714 }
715}
716
717pub struct AnthropicEventMapper {
718 tool_uses_by_index: HashMap<usize, RawToolUse>,
719 usage: Usage,
720 stop_reason: StopReason,
721}
722
723impl AnthropicEventMapper {
724 pub fn new() -> Self {
725 Self {
726 tool_uses_by_index: HashMap::default(),
727 usage: Usage::default(),
728 stop_reason: StopReason::EndTurn,
729 }
730 }
731
732 pub fn map_stream(
733 mut self,
734 events: Pin<Box<dyn Send + Stream<Item = Result<Event, AnthropicError>>>>,
735 ) -> impl Stream<Item = Result<LanguageModelCompletionEvent, LanguageModelCompletionError>>
736 {
737 events.flat_map(move |event| {
738 futures::stream::iter(match event {
739 Ok(event) => self.map_event(event),
740 Err(error) => vec![Err(error.into())],
741 })
742 })
743 }
744
745 pub fn map_event(
746 &mut self,
747 event: Event,
748 ) -> Vec<Result<LanguageModelCompletionEvent, LanguageModelCompletionError>> {
749 match event {
750 Event::ContentBlockStart {
751 index,
752 content_block,
753 } => match content_block {
754 ResponseContent::Text { text } => {
755 vec![Ok(LanguageModelCompletionEvent::Text(text))]
756 }
757 ResponseContent::Thinking { thinking } => {
758 vec![Ok(LanguageModelCompletionEvent::Thinking {
759 text: thinking,
760 signature: None,
761 })]
762 }
763 ResponseContent::RedactedThinking { data } => {
764 vec![Ok(LanguageModelCompletionEvent::RedactedThinking { data })]
765 }
766 ResponseContent::ToolUse { id, name, .. } => {
767 self.tool_uses_by_index.insert(
768 index,
769 RawToolUse {
770 id,
771 name,
772 input_json: String::new(),
773 },
774 );
775 Vec::new()
776 }
777 },
778 Event::ContentBlockDelta { index, delta } => match delta {
779 ContentDelta::TextDelta { text } => {
780 vec![Ok(LanguageModelCompletionEvent::Text(text))]
781 }
782 ContentDelta::ThinkingDelta { thinking } => {
783 vec![Ok(LanguageModelCompletionEvent::Thinking {
784 text: thinking,
785 signature: None,
786 })]
787 }
788 ContentDelta::SignatureDelta { signature } => {
789 vec![Ok(LanguageModelCompletionEvent::Thinking {
790 text: "".to_string(),
791 signature: Some(signature),
792 })]
793 }
794 ContentDelta::InputJsonDelta { partial_json } => {
795 if let Some(tool_use) = self.tool_uses_by_index.get_mut(&index) {
796 tool_use.input_json.push_str(&partial_json);
797
798 // Try to convert invalid (incomplete) JSON into
799 // valid JSON that serde can accept, e.g. by closing
800 // unclosed delimiters. This way, we can update the
801 // UI with whatever has been streamed back so far.
802 if let Ok(input) = serde_json::Value::from_str(
803 &partial_json_fixer::fix_json(&tool_use.input_json),
804 ) {
805 return vec![Ok(LanguageModelCompletionEvent::ToolUse(
806 LanguageModelToolUse {
807 id: tool_use.id.clone().into(),
808 name: tool_use.name.clone().into(),
809 is_input_complete: false,
810 raw_input: tool_use.input_json.clone(),
811 input,
812 },
813 ))];
814 }
815 }
816 vec![]
817 }
818 },
819 Event::ContentBlockStop { index } => {
820 if let Some(tool_use) = self.tool_uses_by_index.remove(&index) {
821 let input_json = tool_use.input_json.trim();
822 let input_value = if input_json.is_empty() {
823 Ok(serde_json::Value::Object(serde_json::Map::default()))
824 } else {
825 serde_json::Value::from_str(input_json)
826 };
827 let event_result = match input_value {
828 Ok(input) => Ok(LanguageModelCompletionEvent::ToolUse(
829 LanguageModelToolUse {
830 id: tool_use.id.into(),
831 name: tool_use.name.into(),
832 is_input_complete: true,
833 input,
834 raw_input: tool_use.input_json.clone(),
835 },
836 )),
837 Err(json_parse_err) => {
838 Ok(LanguageModelCompletionEvent::ToolUseJsonParseError {
839 id: tool_use.id.into(),
840 tool_name: tool_use.name.into(),
841 raw_input: input_json.into(),
842 json_parse_error: json_parse_err.to_string(),
843 })
844 }
845 };
846
847 vec![event_result]
848 } else {
849 Vec::new()
850 }
851 }
852 Event::MessageStart { message } => {
853 update_usage(&mut self.usage, &message.usage);
854 vec![
855 Ok(LanguageModelCompletionEvent::UsageUpdate(convert_usage(
856 &self.usage,
857 ))),
858 Ok(LanguageModelCompletionEvent::StartMessage {
859 message_id: message.id,
860 }),
861 ]
862 }
863 Event::MessageDelta { delta, usage } => {
864 update_usage(&mut self.usage, &usage);
865 if let Some(stop_reason) = delta.stop_reason.as_deref() {
866 self.stop_reason = match stop_reason {
867 "end_turn" => StopReason::EndTurn,
868 "max_tokens" => StopReason::MaxTokens,
869 "tool_use" => StopReason::ToolUse,
870 "refusal" => StopReason::Refusal,
871 _ => {
872 log::error!("Unexpected anthropic stop_reason: {stop_reason}");
873 StopReason::EndTurn
874 }
875 };
876 }
877 vec![Ok(LanguageModelCompletionEvent::UsageUpdate(
878 convert_usage(&self.usage),
879 ))]
880 }
881 Event::MessageStop => {
882 vec![Ok(LanguageModelCompletionEvent::Stop(self.stop_reason))]
883 }
884 Event::Error { error } => {
885 vec![Err(error.into())]
886 }
887 _ => Vec::new(),
888 }
889 }
890}
891
892struct RawToolUse {
893 id: String,
894 name: String,
895 input_json: String,
896}
897
898/// Updates usage data by preferring counts from `new`.
899fn update_usage(usage: &mut Usage, new: &Usage) {
900 if let Some(input_tokens) = new.input_tokens {
901 usage.input_tokens = Some(input_tokens);
902 }
903 if let Some(output_tokens) = new.output_tokens {
904 usage.output_tokens = Some(output_tokens);
905 }
906 if let Some(cache_creation_input_tokens) = new.cache_creation_input_tokens {
907 usage.cache_creation_input_tokens = Some(cache_creation_input_tokens);
908 }
909 if let Some(cache_read_input_tokens) = new.cache_read_input_tokens {
910 usage.cache_read_input_tokens = Some(cache_read_input_tokens);
911 }
912}
913
914fn convert_usage(usage: &Usage) -> language_model::TokenUsage {
915 language_model::TokenUsage {
916 input_tokens: usage.input_tokens.unwrap_or(0),
917 output_tokens: usage.output_tokens.unwrap_or(0),
918 cache_creation_input_tokens: usage.cache_creation_input_tokens.unwrap_or(0),
919 cache_read_input_tokens: usage.cache_read_input_tokens.unwrap_or(0),
920 }
921}
922
923struct ConfigurationView {
924 api_key_editor: Entity<Editor>,
925 state: gpui::Entity<State>,
926 load_credentials_task: Option<Task<()>>,
927 target_agent: ConfigurationViewTargetAgent,
928}
929
930impl ConfigurationView {
931 const PLACEHOLDER_TEXT: &'static str = "sk-ant-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx";
932
933 fn new(
934 state: gpui::Entity<State>,
935 target_agent: ConfigurationViewTargetAgent,
936 window: &mut Window,
937 cx: &mut Context<Self>,
938 ) -> Self {
939 cx.observe(&state, |_, _, cx| {
940 cx.notify();
941 })
942 .detach();
943
944 let load_credentials_task = Some(cx.spawn({
945 let state = state.clone();
946 async move |this, cx| {
947 if let Some(task) = state
948 .update(cx, |state, cx| state.authenticate(cx))
949 .log_err()
950 {
951 // We don't log an error, because "not signed in" is also an error.
952 let _ = task.await;
953 }
954 this.update(cx, |this, cx| {
955 this.load_credentials_task = None;
956 cx.notify();
957 })
958 .log_err();
959 }
960 }));
961
962 Self {
963 api_key_editor: cx.new(|cx| {
964 let mut editor = Editor::single_line(window, cx);
965 editor.set_placeholder_text(Self::PLACEHOLDER_TEXT, cx);
966 editor
967 }),
968 state,
969 load_credentials_task,
970 target_agent,
971 }
972 }
973
974 fn save_api_key(&mut self, _: &menu::Confirm, window: &mut Window, cx: &mut Context<Self>) {
975 let api_key = self.api_key_editor.read(cx).text(cx);
976 if api_key.is_empty() {
977 return;
978 }
979
980 let state = self.state.clone();
981 cx.spawn_in(window, async move |_, cx| {
982 state
983 .update(cx, |state, cx| state.set_api_key(api_key, cx))?
984 .await
985 })
986 .detach_and_log_err(cx);
987
988 cx.notify();
989 }
990
991 fn reset_api_key(&mut self, window: &mut Window, cx: &mut Context<Self>) {
992 self.api_key_editor
993 .update(cx, |editor, cx| editor.set_text("", window, cx));
994
995 let state = self.state.clone();
996 cx.spawn_in(window, async move |_, cx| {
997 state.update(cx, |state, cx| state.reset_api_key(cx))?.await
998 })
999 .detach_and_log_err(cx);
1000
1001 cx.notify();
1002 }
1003
1004 fn render_api_key_editor(&self, cx: &mut Context<Self>) -> impl IntoElement {
1005 let settings = ThemeSettings::get_global(cx);
1006 let text_style = TextStyle {
1007 color: cx.theme().colors().text,
1008 font_family: settings.ui_font.family.clone(),
1009 font_features: settings.ui_font.features.clone(),
1010 font_fallbacks: settings.ui_font.fallbacks.clone(),
1011 font_size: rems(0.875).into(),
1012 font_weight: settings.ui_font.weight,
1013 font_style: FontStyle::Normal,
1014 line_height: relative(1.3),
1015 white_space: WhiteSpace::Normal,
1016 ..Default::default()
1017 };
1018 EditorElement::new(
1019 &self.api_key_editor,
1020 EditorStyle {
1021 background: cx.theme().colors().editor_background,
1022 local_player: cx.theme().players().local(),
1023 text: text_style,
1024 ..Default::default()
1025 },
1026 )
1027 }
1028
1029 fn should_render_editor(&self, cx: &mut Context<Self>) -> bool {
1030 !self.state.read(cx).is_authenticated()
1031 }
1032}
1033
1034impl Render for ConfigurationView {
1035 fn render(&mut self, _: &mut Window, cx: &mut Context<Self>) -> impl IntoElement {
1036 let env_var_set = self.state.read(cx).api_key_from_env;
1037
1038 if self.load_credentials_task.is_some() {
1039 div().child(Label::new("Loading credentials...")).into_any()
1040 } else if self.should_render_editor(cx) {
1041 v_flex()
1042 .size_full()
1043 .on_action(cx.listener(Self::save_api_key))
1044 .child(Label::new(format!("To use {}, you need to add an API key. Follow these steps:", match &self.target_agent {
1045 ConfigurationViewTargetAgent::ZedAgent => "Zed's agent with Anthropic".into(),
1046 ConfigurationViewTargetAgent::Other(agent) => agent.clone(),
1047 })))
1048 .child(
1049 List::new()
1050 .child(
1051 InstructionListItem::new(
1052 "Create one by visiting",
1053 Some("Anthropic's settings"),
1054 Some("https://console.anthropic.com/settings/keys")
1055 )
1056 )
1057 .child(
1058 InstructionListItem::text_only("Paste your API key below and hit enter to start using the agent")
1059 )
1060 )
1061 .child(
1062 h_flex()
1063 .w_full()
1064 .my_2()
1065 .px_2()
1066 .py_1()
1067 .bg(cx.theme().colors().editor_background)
1068 .border_1()
1069 .border_color(cx.theme().colors().border)
1070 .rounded_sm()
1071 .child(self.render_api_key_editor(cx)),
1072 )
1073 .child(
1074 Label::new(
1075 format!("You can also assign the {ANTHROPIC_API_KEY_VAR} environment variable and restart Zed."),
1076 )
1077 .size(LabelSize::Small)
1078 .color(Color::Muted),
1079 )
1080 .into_any()
1081 } else {
1082 h_flex()
1083 .mt_1()
1084 .p_1()
1085 .justify_between()
1086 .rounded_md()
1087 .border_1()
1088 .border_color(cx.theme().colors().border)
1089 .bg(cx.theme().colors().background)
1090 .child(
1091 h_flex()
1092 .gap_1()
1093 .child(Icon::new(IconName::Check).color(Color::Success))
1094 .child(Label::new(if env_var_set {
1095 format!("API key set in {ANTHROPIC_API_KEY_VAR} environment variable.")
1096 } else {
1097 "API key configured.".to_string()
1098 })),
1099 )
1100 .child(
1101 Button::new("reset-key", "Reset Key")
1102 .label_size(LabelSize::Small)
1103 .icon(Some(IconName::Trash))
1104 .icon_size(IconSize::Small)
1105 .icon_position(IconPosition::Start)
1106 .disabled(env_var_set)
1107 .when(env_var_set, |this| {
1108 this.tooltip(Tooltip::text(format!("To reset your API key, unset the {ANTHROPIC_API_KEY_VAR} environment variable.")))
1109 })
1110 .on_click(cx.listener(|this, _, window, cx| this.reset_api_key(window, cx))),
1111 )
1112 .into_any()
1113 }
1114 }
1115}
1116
1117#[cfg(test)]
1118mod tests {
1119 use super::*;
1120 use anthropic::AnthropicModelMode;
1121 use language_model::{LanguageModelRequestMessage, MessageContent};
1122
1123 #[test]
1124 fn test_cache_control_only_on_last_segment() {
1125 let request = LanguageModelRequest {
1126 messages: vec![LanguageModelRequestMessage {
1127 role: Role::User,
1128 content: vec![
1129 MessageContent::Text("Some prompt".to_string()),
1130 MessageContent::Image(language_model::LanguageModelImage::empty()),
1131 MessageContent::Image(language_model::LanguageModelImage::empty()),
1132 MessageContent::Image(language_model::LanguageModelImage::empty()),
1133 MessageContent::Image(language_model::LanguageModelImage::empty()),
1134 ],
1135 cache: true,
1136 }],
1137 thread_id: None,
1138 prompt_id: None,
1139 intent: None,
1140 mode: None,
1141 stop: vec![],
1142 temperature: None,
1143 tools: vec![],
1144 tool_choice: None,
1145 thinking_allowed: true,
1146 };
1147
1148 let anthropic_request = into_anthropic(
1149 request,
1150 "claude-3-5-sonnet".to_string(),
1151 0.7,
1152 4096,
1153 AnthropicModelMode::Default,
1154 );
1155
1156 assert_eq!(anthropic_request.messages.len(), 1);
1157
1158 let message = &anthropic_request.messages[0];
1159 assert_eq!(message.content.len(), 5);
1160
1161 assert!(matches!(
1162 message.content[0],
1163 anthropic::RequestContent::Text {
1164 cache_control: None,
1165 ..
1166 }
1167 ));
1168 for i in 1..3 {
1169 assert!(matches!(
1170 message.content[i],
1171 anthropic::RequestContent::Image {
1172 cache_control: None,
1173 ..
1174 }
1175 ));
1176 }
1177
1178 assert!(matches!(
1179 message.content[4],
1180 anthropic::RequestContent::Image {
1181 cache_control: Some(anthropic::CacheControl {
1182 cache_type: anthropic::CacheControlType::Ephemeral,
1183 }),
1184 ..
1185 }
1186 ));
1187 }
1188}