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